small_avatar_AdrianR3.png

thumbnail_avatar_AdrianR3.png

avatar_AdrianR3.png

Adrian

GET IN TOUCH

#### Sesame goes viral, LynxJS and the component-based approach, and Gemini Robotics and its Gemini 2.0-based models are truly impressive.


In our latest newsletter, we've gathered the most fascinating updates from the tech world, sprinkled with intriguing tidbits from its many orbits. Our subscribers were the first to get their hands on these cutting-edge insights, delivered straight to their inboxes.

#### Main Topic: **Sesame’s AI Voice Tech Goes Viral: Meet the Open-Source Model Behind Maya**

![MAIN TOPIC_sesame.jpg](https://bitnoise-cms.btn.fyi/uploads/MAIN_TOPIC_sesame_d6c35a1901.jpg)

Sesame gained viral attention for its lifelike voice technology. Co-founded by Oculus’ Brendan Iribe, the company has open-sourced CSM-1B, the base model powering its realistic voice assistant, Maya. This 1-billion-parameter model, licensed under Apache 2.0, generates RVQ audio codes, similar to Google’s SoundStream and Meta’s Encodec.

The conversation experience with Maya is highly realistic and remarkably natural, making interactions feel smooth and engaging. [**Try it yourself and see the future of AI voice technology in action!**](https://www.sesame.com/research/crossing_the_uncanny_valley_of_voice?#demo)

#### Tool of the month: **Lynx - Lightweight and Efficient JavaScript Framework**

![TOOL OF THE MONTH_Lynx.jpg](https://bitnoise-cms.btn.fyi/uploads/TOOL_OF_THE_MONTH_Lynx_f9ef15dc4d.jpg)
LynxJS is a lightweight JavaScript framework for building web applications with a component-based approach. It features a Virtual DOM, reactive data binding, and a modular structure for flexibility and performance. Designed to be fast and efficient, it’s a great alternative to heavier frameworks like React or Vue.

[**Read more.**](https://lynxjs.org/index.html)


#### Trending

1.[**Watch a video showcasing the capabilities of the new Gemini Robotics models,**](https://www.youtube.com/watch?v=4MvGnmmP3c0) powered by DeepMind AI. Their interactivity, dexterity, and precision are truly impressive!

2. Anders Hejlsberg, Microsoft’s Technical Fellow and lead architect of TypeScript, [**unveils a new TypeScript**](https://devblogs.microsoft.com/typescript/typescript-native-port/) port designed to power the next generation of high-performance developer tools.

#### Around the Web

📱Light, the company behind minimalist phones, is launching the Light Phone III on March 27. [**Check out its features and design.**](https://techcrunch.com/2025/03/13/minimalist-light-phone-iii-launches-march-27/)

🏁 [**Skype will close**](https://www.bbc.com/news/articles/cn7vxlrvxyeo) in May.

📍 An update to Google’s Android tracking app now [**displays contacts who have shared their location with you.**](https://www.theverge.com/news/630222/android-march-find-my-device-people-location-sharing-update)

thumbnail_tech news march 2025.jpg

tech news march 2025.jpg

Tech News March 2025 - Newsletter Highlights

Frontend Engineering

UX/UI Design

Research & Discovery

AI For Your Business

Services

Case Studies

Career

About us

Blog

English, acknowledged as an international and global language, is spoken by approximately 2 billion people, making it the most popular language despite its lack of official status. Companies and creators aiming to communicate effectively with a broad audience often choose this language, believing it will attract more customers and build stronger relationships.

But what if there's a better and, more importantly, faster way to reach customers and gain global audience recognition, **regardless of language barriers?**

New AI tools are rapidly advancing, introducing surprising innovations in various fields. Regarding language matters, there are already several "strong players" in the market, but in our opinion, **HeyGen** deserves the most attention. It's not just an excellent tool for generating virtual avatars (although truly impressive), but, most importantly, it effortlessly **allows you to create video content in any language worldwide, in your own voice!**

Sounds like magic? Yes, the level of technological advancement truly astonishes!

Impressed by HeyGen's capabilities, we thought about how to use it to create engaging content on social media. It was early December, and the holiday atmosphere was building up, so we almost simultaneously came up with the idea: we'll create video holiday greetings for our international clients. We wrote the greeting content and decided which languages we wanted to speak. Then came the technical part, which was my task. I delved into the instructions on the tool's website, familiarized myself with its operation, and did a few trials... And there it is! You can see the results [HERE](https://www.instagram.com/reel/C1HGi0jqR4W/?utm_source=ig_web_copy_link&igsh=MzRlODBiNWFlZA==)

#### Quick guide

The whole process was gratifying and sometimes even amusing, but in the end, I was left with a few thoughts. Therefore, I present **a quick guide and some important tips that will help you create a satisfying avatar speaking in any language in the world:**

**Video** - you must upload a minimum 2-minute video sample to the tool, where you or another person speaks any text. Of course, **good video quality is essential**. You can use a phone, but the better the quality of the recording is, the better the final result, so I recommend using a camera.
TIP 1: Use a tripod and set up the best, straight frame. Ensure good lighting on the person's face, ideally with no shadows. Also, **think about the background** - keep it neat, tidy, and clean - remember, **it will always be visible behind your avatar**, and you probably don't want to cut it out in post-production cables sticking out somewhere ;). Of course, to avoid background problems, you can use a green screen.
TIP 2: For the best effect, edit the video in an editing program first - adjust colors, play with contrast, etc. Do everything to make the material satisfying before uploading it to HeyGen.

**Speech/sound** - an excellent audio track right after the video is crucial for achieving a satisfying effect. The content of your speech is entirely irrelevant; it's about how you say it. Speak naturally, but try to pause between some sentences.
TIP 1: The instructions on the HeyGen website suggest taking a 1-2 second break after each sentence, but this has a bad effect. Speak clearly, loudly, and energetically. And the most important, **emotions in the voice matter - a sad tone will result in a sad avatar**.
TIP 2: It's good to "spice up" the soundtrack before uploading it to the tool. Basic editing is enough: noise reduction and gentle voice strengthening. (If you don't know how to do this, you can find plenty of short tutorials on YouTube on how to do it).

**Posture and appearance** - stand upright and have a natural facial expression. Look at the camera constantly, but focus only a little on it. From my observations, natural eye movements to the sides are better than excessive focus on the lens. You can also gesture - but don't overdo it - don't do anything you wouldn't do in a regular conversation. **Round, large gestures are not suitable.**
TIP1: What you wear for recording depends on you, but remember that due to the possibility of glitches in the avatar, HeyGen does not recommend wearing jewellery and glasses, and excessive facial hair is also not advisable.



#### All set?

Upload the video to the tool. The program will also ask you to add a short confirmation of your identity. Click "consent" and wait a few minutes until Hey Gen generates your avatar. And that's it!

![blog-hey gen_grafika 1_high (1).png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_hey_gen_grafika_1_high_1_261a9d6980.png)

Enter any text in the box below. Add pauses by inserting the clock icon. **Pay attention to spelling and punctuation - this mainly affects the tone of the generated speech.** On the right side, you can choose language. The program will automatically translate the text. Click “submit” and enjoy the results.
**In the free plan, you can create up to a 2-minute video.**

![blog-hey gen_grafika 2_high (1).png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_hey_gen_grafika_2_high_1_a5f6e0349e.png)

👁️ Interestingly, **the tool also allows the translation of previously recorded content**. On the bottom left of the HeyGen homepage, there is a "Video Translate" button.
Upload your video, choose one of the available languages, and set the number of speakers. The results of such translation are less natural but still very impressive.

![blog-hey gen_grafika 3_cover_high (1).png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_hey_gen_grafika_3_cover_high_1_c80f0297f2.png)


The tool surprises and opens a new way to create international content. The primary and indisputable advantages of this solution are:
- Maximum time savings
- Super easy content creation
- Effortless translation of content
- Ability to localize ready-made content for any market
- Relatively low subscription cost - starting from $24/month for 15 minutes of generated video

#### The immense potential

In today’s connected world, tools such as HeyGen stands out as a transformative tool, breaking down language barriers to unify global communication. This journey, from creating multilingual holiday greetings to mastering its innovative features, showcases the shift towards universally accessible content.

The experience with HeyGen reveals the immense potential of AI in bridging communication gaps, heralding a future where stories and messages are boundless. In this new era, content speaks every language, making global understanding and connection not just possible, but seamless.

large_blog-hey gen_cover_high.png

medium_blog-hey gen_cover_high.png

small_blog-hey gen_cover_high.png

thumbnail_blog-hey gen_cover_high.png

blog-hey gen_cover_high.png

Beyond language borders with HeyGen: crafting videos that speak to the world!

In an age where technology intertwines with every aspect of our lives, why not let it play cupid this Valentine's Day? Artificial Intelligence has evolved from a mere concept into a tool capable of expressing the deepest emotions of love and affection. Whether you're a tech enthusiast or simply looking for a unique way to surprise your significant other, AI offers many possibilities to make this Valentine’s Day unforgettable. Below, we delve into how you can use AI tools like ChatGPT, Midjourney, and Suno.Ai to craft personalised tokens of love.

Here's a guide to Valentine's Day gestures for the forgetful.


#### Use ChatGPT to craft The Perfect Love Poem

ChatGPT, an AI-driven chatbot, is revolutionising how we think about writing. It uses advanced algorithms to generate text based on the prompts you provide. This Valentine's, it can help you articulate your feelings through a bespoke love poem.

**How to use ChatGPT for a love poem:**

**Access ChatGPT:** Create an account on https://chat.openai.com/. Easy and simple. 
**Set the tone:** Initiate the conversation with a prompt like “Write a romantic love poem.”
**Personalize:** Add details to make the poem unique. Try “Include themes of enduring love, shared adventures, and the joy we find in each other’s company.” Ask it to write about their favourite dishes, your first date, and unforgettable moments.
**Generate & Refine:** Review the generated poem. Refine it with additional instructions or ask for revisions if it's imperfect. If you have a favourite artist, let their works serve as a guide to the tone. 
**Sample Prompt:** “Create a love poem that captures the essence of a five-year journey together, filled with both laughter and challenges, highlighting the strength of our bond. Get inspired by Shakespeare's style.”
![blog-walentynki-chatgpt 1_high.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_walentynki_chatgpt_1_high_aeb756c371.png)


#### Midjourney: designing personalised graphics

Midjourney is an AI tool designed for creating visual content. This tool can help you design a heartfelt graphic or illustration visually representing your love story.

**How to Use Midjourney:**

**Access the tool:** Join the Midjourney discord server from this [link](https://discord.com/invite/midjourney) and choose your subscription plan by simply writing /subscribe on any Midjourney’s channel (or on the website [https://www.midjourney.com/account](https://www.midjourney.com/account))
**Describe your vision:** type in a detailed description. For a Valentine’s Day theme, you might use “two lovers watching the sunset, with the quote “Together is our favourite place to be”.” Again - you can ask for inspiration from your favourite artist!
**Customise:** add details about colours, style, or elements you want included.
**Review & Adjust:** examine the generated graphic. You can adjust your instructions based on the outcome.
**Sample Prompt:** “illustration of a couple surrounded by books and coffee, capturing a cosy, intimate moment with a soft pastel colour palette.”
![blog-walentynki-midjourney_high.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_walentynki_midjourney_high_b2a6d5764f.png)



#### Suno AI: Composing Your Love Ballad

Suno AI is an AI that specialises in music composition. With Suno.Ai, you can create a love ballad that melodically expresses your feelings, making for a deeply personal and unique gift.

**How to Use Suno AI:**

**Select your style:** create an account on suno.ai, choose a musical style that resonates with your love story or select "Custom Mode" and let the magic happen.
**Input Lyrics:** use your ChatGPT-generated poem or any special phrases as lyrics. Or generate a completely new text;  it’s totally up to you. Pay attention to marking in the text what the [verse] and the [chorus] are. You will get the best results when the text is rhythmic and rhyming.
**Generate the music:** allow Suno AI to compose the music based on your input.
**Refine:** listen and make any necessary adjustments to ensure the music perfectly encapsulates your feelings.
**Sample Prompt:** “Compose a gentle acoustic ballad that reflects a journey of growth and love, incorporating lyrics about overcoming life’s storms together.”
![blog-walentynki-suno.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_walentynki_suno_2627672f69.png)



This Valentine’s Day, embrace the power of AI to create something genuinely heartwarming and personal. From a poem that eloquently expresses your feelings to a custom graphic that visually depicts your love or a ballad that sings your heart’s song, the possibilities are as limitless as your affection. Let these AI innovations help you express the timeless emotion of love in a modern and touching way.

Happy Valentine’s Day! May your celebration be filled with love, innovation, and the magic of AI-crafted memories.

large_cień+blog-valentynki_cover_high.png

medium_cień+blog-valentynki_cover_high.png

small_cień+blog-valentynki_cover_high.png

thumbnail_cień+blog-valentynki_cover_high.png

cień+blog-valentynki_cover_high.png

Harnessing AI for Love: a Valentine’s Day Guide

![blog-AI ŁUKASZ R_grafika 1_WSTĘPpng.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_LUKASZ_R_grafika_1_WSTE_Ppng_b7da5fba06.png)

Having nurtured a passion for technology from a young age, I've developed a keen understanding of its potential to streamline processes and enhance productivity. It's easy to assume that the utility of available market tools is universally recognized; however, I am aware that the true value of these resources may not be immediately apparent to everyone. Many of these accessible solutions can significantly benefit your business. Allow me to provide a concise overview of several key strategies you can implement immediately. Remarkably, you can leverage many of these solutions without the need for extensive technical knowledge or the employment of engineers.

So, let's dive in.

#### Applications and services

![blog-AI ŁUKASZ R_LOGO-CHAT GPT MIDJOURNEY.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_LUKASZ_R_LOGO_CHAT_GPT_MIDJOURNEY_421841dac2.png)

Let's start with the dedicated tools that you, as a business owner, can begin using even today. There are two types of integrations:

##### **Direct**

There are already a few tools that I can recommend for you to start using. Some of them are obvious, and some are not.

You can start with popular AI tools like:

- **ChatGPT**: It can assist you in various ways, including automating customer service through intelligent and responsive communication, providing content creation for marketing materials, aiding in data analysis by summarizing large volumes of text, and enhancing productivity by answering queries and generating reports.
    
- **Midjourney**: This tool focuses on generating high-quality visual content. It could be a game-changer in areas like marketing, branding, and product development. It enables the rapid creation of visual assets, helping you enhance your online presence, attract more customers, and effectively communicate your brand message.

Or you can dig deeper and try using some less popular tools like:

- **HeyGen**: Helps you create avatars and multilingual voices for videos. It's a cost-effective and time-saving solution, eliminating the need for traditional, resource-intensive video production methods. Its user-friendly nature makes high-quality video production accessible, even for those without technical expertise. Adopting this AI-driven approach can significantly enhance your marketing efforts.
[Check out Dominika's article about HeyGen. ](https://www.bitnoise.pl/blog/beyond-language-borders-with-HeyGen)

- **MacWhisper**: This tool can quickly and easily transcribe your audio files into text using OpenAI's state-of-the-art transcription technology, Whisper. Whether you're recording a meeting, lecture, or any other important audio, it accurately transcribes your files into text.

##### **Indirect**

It's highly possible that you are already using some form of AI in your daily life, such as the autocomplete feature in your Gmail account, or you are using tools like Notion, which has already introduced support for text generation in its tool. You can [explore]( https://www.notion.so/help/guides/using-notion-ai) how it can improve your daily work. 


#### Prompt engineering

![blog-AI ŁUKASZ R_PROMPT.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_LUKASZ_R_PROMPT_5f0817fd10.png)

It's highly likely that you've already engaged in prompt engineering, perhaps by experimenting with the capabilities of tools like ChatGPT, Dall-e, or Midjourney. But what exactly is prompt engineering?

Prompt engineering is the process of carefully designing and formulating prompts (inputs) to communicate with AI models effectively. The goal is to elicit the most accurate, relevant, and coherent responses from the AI. This involves understanding the AI's capabilities and limitations and crafting prompts that are clear, specific, and structured in a way that guides the AI towards providing the desired output. It's a crucial skill for effective AI interactions.

So, what can you do, and where can you seek advice? I would suggest the following:

- Start simple and observe the results.
- Don't be afraid; try new things and observe how the model reacts to your prompts.
- Seek inspiration by observing what others have done.
- Read official and community-based documentation, such as: [OpenAI's guide on prompt engineering](https://platform.openai.com/docs/guides/prompt-engineering) or a great example of multiple resources [here](https://roadmap.sh/prompt-engineering). An amazing guide can be found at [LearnPrompting.org](https://learnprompting.org/docs/intro).
- You can also exchange prompts with your team and build a unique internal knowledge base for your organization.


#### No-code integrations and automation

![blog-AI ŁUKASZ R_LOGO_MAKE-ZAPIERpng.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_LUKASZ_R_LOGO_MAKE_ZAPIE_Rpng_339656f2f0.png)

There are several tools and services that you can start using with a little bit of knowledge, either by learning new things or by asking someone who is 'more technical' on your team. What can you do using these tools? You can automate your processes, leveraging, for example, ChatGPT to help you as one piece of your workflow puzzle. Imagine that every time you receive a new email, ChatGPT automatically checks the business context of this email. If it's, for example, some kind of critical situation (e.g., a system is down), you receive a direct notification via SMS or even a direct call where the bot reads you this email.

This is the easiest possible way to use different pieces of services, connect them with current AI capabilities, and optimize your internal processes on a daily basis for you or the business you are running.

You can check out platforms like [Make.com](https://make.com/), [IFTTT.com](https://ifttt.com/), or [Zapier.com](https://zapier.com/) if you are interested, or contact us so we can help you as well.

#### Foundation models

![blog-AI ŁUKASZ R_LOGO-OPENAI-LLAMA.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_LUKASZ_R_LOGO_OPENAI_LLAMA_2e95e7b1f2.png)

This is the point where you will probably need an external or internal technical provider to help you utilize the currently available technology. But for now, I just want to give you a high-level idea of what you can achieve using Large Language Models (LLMs) for commercial or private use.

What is a foundation model?

A foundation model is a type of artificial intelligence model that is trained on a large, diverse dataset from a wide range of sources. These models are called "foundation" because they provide a broad, general understanding of the world, language, images, or other data types, which can then be fine-tuned or adapted for a wide variety of specific tasks without needing to be trained from scratch for each new task.

So, you can use commercial models like the one provided by OpenAI or utilize open-source models like Mistral or Lama. However, using them relies on large, universal datasets of data that were used to train the model.

How can you use Large Language Models (LLMs) with your own data? There are two options that we have already built for our internal needs, both of which are based on our internal knowledge.

#### Finetuning

![blog-AI ŁUKASZ R_grafika 2_FINETUNING.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_LUKASZ_R_grafika_2_FINETUNING_c7c57f8119.png)

It's a process where a pre-trained model is further trained on a smaller, specific dataset to adapt it for a particular task. This process leverages the broad knowledge the model has acquired during its initial extensive training on a large and diverse dataset, allowing it to quickly learn the nuances of the new task with relatively few additional examples.

This approach is efficient and effective, requiring less data and computational resources than training a model from scratch for each new task.

You can think of it as creating a general model tailored to your specific needs or specialized tasks within various industries. It still requires data for training and resources with computing power to fine-tune the model, but the process takes only minutes or hours rather than weeks or months, as is the case with foundation models.

How can you use it in your business?
- Chatbot for Customer Service - Fine-tuning a foundation model on a dataset specific to a company's products or services can create highly effective chatbots for customer service.
- Personalized Education - Educational content can be used to fine-tune models to create personalized tutoring systems that cater to the unique learning styles and needs of individual students.
- Sentiment Analysis - Companies can fine-tune models on customer reviews, social media posts, and other feedback to perform sentiment analysis tailored to their products or services.

But you may ask, what if my data is constantly changing or if I need to analyze large documents on a daily basis? The idea of fine-tuning the model for each document might seem impractical. This is where we can introduce another idea:

#### RAG architecture

![blog-AI ŁUKASZ R_grafika 3_RAG.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_LUKASZ_R_grafika_3_RAG_dfb81670f3.png)

It is an innovative approach that combines the best of two worlds in natural language processing: retrieval-based and generative models. It was introduced to enhance the ability of generative models to produce more accurate, relevant, and informed content by leveraging external knowledge sources.

How can it help you?

One of the examples we've already built for our internal needs is a Knowledge-Enhanced Chatbot - imagine a bot on your Slack/Teams/whatever which can provide detailed, accurate, and contextually relevant responses to user queries by fetching information from a vast database of domain-specific knowledge. This capability makes it particularly useful in areas where expertise and up-to-date information are crucial. The domain doesn't really matter so it could be useful in:
- **Education**: Assisting in learning and research by sourcing detailed explanations, examples, and references from educational materials.
- **Finance**: Providing personalized investment advice by accessing real-time market data, financial news, and analysis reports.
- **HR**: Can automate the initial screening process of resumes and applications by retrieving information from job descriptions, required qualifications, and ideal candidate profiles. It can then generate a summary or score for each applicant, highlighting how well their background, skills, and experiences match the job criteria.


#### But can I create my own LLM?

Yes, although it's unlikely that you would. You need time and money to collect data, training time can range from weeks to months. For example, GPT-3, with its 175 billion parameters, was trained for weeks on thousands of GPUs. OpenAI’s GPT-3 is rumoured to have cost around $4.6 million in direct computing costs. Building a foundation LLM is a resource-intensive endeavour that demands careful planning and significant investment.


At the same time, currently available tools for end users and engineering teams are more than sufficient to utilise Large Language Models (LLMs) to accelerate business processes, optimize our time, and automate tasks.

large_blog-Łukasz AI_cover.png

medium_blog-Łukasz AI_cover.png

small_blog-Łukasz AI_cover.png

thumbnail_blog-Łukasz AI_cover.png

blog-Łukasz AI_cover.png

How AI can supercharge your business in 2024 on multiple levels of your cost and effort

In a recent episode of PozTechTalks, Łukasz Roth, along with technology experts Piotr Baranek (Bitnoise)  and Patryk Ziemkowski (Apptension), delved into the world of AI tools designed to aid developers. Their discussion illuminated the potential and challenges of three such tools, shedding light on their practical applications.



**OpenCommit** focuses on automating commit message generation. While it offers customization options and integrates with VS Code for ease of use, its reliance on external APIs (specifically OpenAI) and its detailed output, which sometimes includes irrelevant data, were noted as drawbacks.

**Codium**, another tool highlighted, excels in automatic test generation, particularly for React applications. Despite its innovative approach to generating both use cases and unit tests, its unpredictability and over-reliance on specific file structures limit its effectiveness. The tool's capacity to mock functions and hooks was praised, but its deterministic nature and occasional inaccuracies in test generation were points of concern.

The discussion also touched upon **Stepsize** AI, aimed more at project managers than developers, which generates reports on team performance and project metrics. While potentially useful for large projects with extensive data, scepticism was voiced regarding the accuracy and reliability of its automated reports. The inability to verify the underlying data processing and the tool's paid model add layers of complexity to its adoption.

![blog-AI Tools for Developers_high.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_Tools_for_Developers_high_164b6eb1ee.png)

Each tool presents unique capabilities designed to streamline the development process, from simplifying commit messages with **OpenCommit** and enhancing testing procedures with **Codium** to providing project insights through **Stepsize** AI. However, their limitations, such as detailed but sometimes irrelevant output, unpredictability, and data accuracy concerns, underscore the importance of human oversight in AI-assisted development.

As AI continues to evolve, these tools represent a step forward in automating and optimizing development workflows. Still, they also highlight the need for continued refinement and integration with human expertise.

This exploration into AI tools for developers ends here, but the journey doesn't stop. Stay tuned for the next part of the #PozTechTalks podcast, where we'll dive into three more innovative tools designed to elevate the coding experience. Each tool promises to bring unique advantages to the table, addressing different aspects of development and project management. Don't miss out on uncovering more AI solutions that could revolutionize your workflow.

Check out the full episode (in Polish)!

AI Toolkit Review: Devs Perspective #1

large_blog-AI Tools for Developers_cover_high.png

medium_blog-AI Tools for Developers_cover_high.png

small_blog-AI Tools for Developers_cover_high.png

thumbnail_blog-AI Tools for Developers_cover_high.png

blog-AI Tools for Developers_cover_high.png

Exploring AI Tools for Developers: Insights from PozTechTalks

#### Choose wisely - how does the version affect your images?

Configuring Midjourney optimally makes further work with the bot much easier. You can access the settings in any Discord channel where you and the bot are (only you will see it) or in a conversation with the bot using the **/settings** command.

At the beginning, you will find a menu (dropdown) where you can choose which version of MJ you will use as the default one and its parameters.

Users are usually advised to work with the latest version, but it's a fascinating process to see how different editions interpret the prompt. Midjourney version 6 was released on December 21, 2023, which makes it a very fresh model at the time of publication of this article. However, version 5, which has dominated the last few months, is equally worth attention. The biggest difference is in the understanding of prompts. Version 5 is sensitive to words with high aesthetic boost (e.g. -core, -punk) and references to cultural and pop-cultural aesthetics. It is, therefore, difficult to write longer prompts in it without disturbing the balance of keywords. Version 6 understands natural language much better and accepts long descriptions, combining them quite logically. However, it requires an accurate and precise definition of the scene conditions - from the shooting technique through the description of the shot to the detailed description of the background. Which one is better? You have to check it yourself.

![blog-Midjourney guiden_grafika_menu.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guiden_grafika_menu_a8a3fff0e3.png)

#### First things first

The first line contains styling parameters. We can set the stylise and style parameters. The raw style is the only ready-made style option in 6 (you will find more of them in Niji – the Midjourney anime model –) and focuses on faithfully reproducing the prompt and the references used in the prompt. Each option: stylise low, stylize med, stylize high and stylize very high corresponds to setting the –s parameter to 50, 100, 250 and 750.

How does this affect the image? CRAZY. Like every parameter, anyway. The Midjourney bot has been trained to create the most "artistic" images possible, including consistency of composition, references, colours and lighting. The higher this coefficient is raised, the more "artistry" the bot will add. However, this comes at the cost of faithfulness to the prompt. In the above examples, I used a powerful archetype - shieldmaiden - which will generate a Nordic female warrior. I added a problematic condition - a blue tattoo on the face - and a very vague definition of the background - fog.

It is easy to see how the image and its style change as the styling increases - becoming increasingly generically mid-journey. Images with -s above 350 are usually very nice but need help staying true to the details of the prompt.

At the same time, an interesting side effect appears in version 6 - above the value of 200; characters tend to stand backwards if their position is not specified in the prompt. It is also important that the bot in version 6 adds less "beauty" and focuses more on trying to reproduce the exact situation/conditions that the user has described of what the user has entered. It tries to bring the dial in and keep it in the image. Styled at level 750, the warriors in version 5 are almost identical; in version 6, they are very different and definitely not as Instagram-beautiful.

To maintain an interesting composition, the Midjourney bot expands the background by adding new elements to it. In turn, complex details such as a tattoo disappear at the expense of making the image more beautiful. By default, in /settings, you will probably find the med setting, but there are plenty of fields for experiments. **The stylised parameter can be changed by 1, taking values from 0 to 1000**. Each version of MJ is also "sensitive" to the stylised parameter in a different way, and (current) version 6 is very sensitive to it.
![blog-Midjourney guide_prompt.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guide_prompt_a25e8a4389.png)


![blog-Midjourney guide_grafika 1.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guide_grafika_1_848a2e5790.png)

![blog-Midjourney guide_grafika 2.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guide_grafika_2_f78146a553.png)

![blog-Midjourney guide_grafika 3.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guide_grafika_3_24e24f7569.png)
![blog-Midjourney guide_grafika 4.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guide_grafika_4_87b68bd600.png)
![blog-Midjourney guide_grafika 5.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guide_grafika_5_fd94779c8f.png)

In the next line of your Midjourney settings, you can switch between public and stealth modes. Stealth mode is available to users with the Pro plan ($60) and higher. It makes images created in this mode invisible to other Midjourney users.

#### Remix, baby! 

The next option is the most important in Midjourney, which is activating a remix. Remix allows you when flipping the entire grid, a single image, zooming, panning, etc., to invoke a modal (pop-up window) in which you can modify the prompt. This is an amazing option thanks to which you can save the existing composition of the image and change small details in it, flip the whole thing to create amazing effects, add absolutely new things using vary (region) / pan/zoom, add references along the way and much more.


![blog-Midjourney guide_prompt1.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guide_prompt1_d951437389.png)

Replacing the female warrior with a bearded Viking - but in a similar place, light and composition - is just a text replacement in the remix modal. And I assure you, this is the weakest remix option :)

More options for single image variations - since version 5.2, two variation options have appeared under images: Subtle and Strong. They are equivalent to Low Variation Mode and High Variation Mode. I haven't been able to find an answer as to the exact values, but judging by the bot's behaviour, Subtle (Low Variation Mode) oscillates between 50 and 100 and Strong (High Variation Mode) between 250 and 350. **Variation Mode is assigned to the remix function** - also, every time the user calls the modal – the appropriate styling parameter is added simultaneously.

In the last line, you will find the Sticky Style option. It automatically allows you to add the previously used tuner style to your prompt. Only turn it on if you have played with the tuner. If you were having fun, don't turn it on either :). Theoretically, it very intuitively replaces each newly generated style from the tuner with the appropriate suffix and adds it to each prompt. If you enjoy constantly using the tuner, this will be the perfect option.

The following three buttons allow you to switch between **turbo, fast, and relax modes**. Turbo generates images up to four times faster, using twice as many minutes. Fast mode is the user's default. Relax mode (available to users from the Standard Plan) allows you to generate images without using minutes. The generation time increases slightly, but the differences are not significant.

At the very end, there is a button that resets the settings to default. So... you can always go back to the beginning :). But now you are starting a fascinating journey! 

Stay tuned for the next part! 

large_blog-Midjourney guider_blog okładka.png

medium_blog-Midjourney guider_blog okładka.png

small_blog-Midjourney guider_blog okładka.png

thumbnail_blog-Midjourney guider_blog okładka.png

blog-Midjourney guider_blog okładka.png

Midjourney guide part 1: Simple thing - settings

Midjourney is one of the (best generative AI services currently on the market (March 2024). I love Midjourney with all my heart and am incredibly happy to explore its possibilities. The ability to convert text (which I like and know how to use) into an image (unfortunately, I have no talent for this) is one of the most exciting possibilities of AI. I tell the Midjourney Bot what I want to see, which generates this image for me - isn't it stunning? I persistently explore how to explain and convey my ideas so that he understands them.

#### What is a prompt?

Translation is literally a "hint" and a suggestion or encouragement. But this word doesn't quite describe its use in Midjourney, so I'll use its English name. A prompt is **a short text phrase that Midjourney Bot interprets to generate an image**. Midjourney Bot splits the words and phrases in the prompt into smaller pieces, called tokens, that can be compared to training data and then used to generate an image. Therefore, **a well-built prompt is the basis for creating graphics in Midjourney**.


#### Construction of a prompt

The primary prompt can be divided into at least three main sections:
- graphic references (previously given image);
- text prompt (description of what MJ should generate);
- parameters.

The only mandatory part of a prompt is the text - everything else can be omitted. And yes, a prompt can be one word, but we often expand it into more.
![blog-_grafika.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_38d5f63ef4.png)

So, the simplest prompt might look like this:
![blog-_grafika-1.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_1_c2a7d68683.png)

And a bit more complicated like this:
![blog-_prompt2grafika.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_prompt2grafika_90f9c885f8.png)

However, the results will be completely different:![blog-_grafika-2.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_2_d91e7742c0.png)


Midjourney creates impressive opportunities to experiment and approach a given topic. In addition to the randomness of the generated image, we can modify the parameters and direct this process, leading to desired and sometimes astonishing results.

In the next blog entry, I will focus on a more detailed analysis of the prompt structure so you can maximize this tool platform's potential.

![blog-_grafika-3.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_3_15ffbff893.png)


large_blog-_cover.png

medium_blog-_cover.png

small_blog-_cover.png

thumbnail_blog-_cover.png

blog-_cover.png

Midjourney guide part 2: Introduction to MJ prompts

The essence of the prompt is to describe your own graphic idea as best as possible. To facilitate manipulation of the text description itself, you can divide it into specific sections:
![blog-_grafika.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_6dc177de06.png)

#### Subject (must be specified)
The most important thing is an essential approach to the central theme of the painting - "sleeping cat", "a couple kissing", "a house on the hill". Think momentarily and capture what will appear in the generated graphics as briefly as possible. Here, attention to precision and conciseness is the most important.

See below - these are the simplest possible prompts that simply generate a given image:

/imagine a house on the hill

/imagine a couple kissing on the beach

/imagine a cat on the roof

![blog-_grafika-1.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_1_2bbec1254d.png)

#### Details (must be added)

Then, there is space to add the details that you consider most important and relevant. These should be details relating to what the character does, what they look like, what the scenery is like or any other information that you consider essential.

/imagine a house on a **haunted** hill at **night and a full moon**

/imagine **the silhouette** of a couple kissing on the beach **during aurora**

/imagine **a sleeping cat** on the roof wearing **a funny hat**

![blog-_grafika-2.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_2_b33b20a76f.png)


I also recommend adding at this step:

**Colours ** – Midjourney selects the dominant colour in the generated graphics based on this. Of course, we don't always have control over where this colour appears, but it can be skillfully incorporated here.

**Mood** - contrary to appearances, determining the mood of an image adds a lot to its overall tone - "golden kamui mask" will look different in the "mystic" version and differently when "creepy" appears in the prompt

#### Style (is above all)


And this is when you add an outline to the generated graphics. As Midjourney grows and feeds you new images, it creates a specific default Midjourney image style that you can manipulate as you wish. Methods are the term:
Type of media – try to determine whether it is photography, illustration, sculpture, or concept art;
Techniques - if an illustration - then like from a children's book? Or watercolour? If the photo: is it an old-school sepia one or a sports shot of movement?
References to other artists - whether through the title of the work -  "Corridors Like from the Alien movie", or names -  "Photography by Ann Leibowitz", or even pop culture figures - "live action 2B".

Adding a style changes the generated images radically:
/imagine** ink style smudge** house on a haunted hill at night and a full moon, watercolor art

/imagine **a black background kirigami** silhouette of a couple kissing on the beach during aurora

/imagine a sleeping cat on the roof wearing a funny hat, **anime style, sweet dreams, pink**

![blog-_grafika-3.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_3_751a9a38e5.png)



#### Where to look for exciting styles?

First, I recommend Andrea Kovalev's library, available here: https://www.midlibrary.io/. Perfectly sorted, with examples and interesting descriptions.

Additionally – mix it up. Mix the strangest styles - fashion designers with book illustrators, photographers with architects. This method creates unique graphics.

When looking for interesting styles, it is worth starting by exploring the entire aesthetic. The definition of aesthetics is something very elusive, but you can try to call it a set of principles underlying the work of a specific artist or artistic movement. You can start your journey on the aesthetic wiki (https://aesthetics.fandom.com/wiki/Aesthetics_Wiki) - exploring them through images, relationships with others or general moodboards. Almost every word you find on this page brings a huge load of guidelines for the MJ bot.

#### Boosters (do they work or not?)

Boosters are specific phrases (entered passionately by Midjourney users) intended to add an additional, stunning effect. Some do not work, take up significant space in the prompt, or generate noise, while others provide cool effects.

- **Specifying the lens type; cameras** - many photographers using Midjourney tested the prompts - sometimes differences emerge - but they have little to do with the equipment used. So what works? Photography or the name of the photographer above all! But determining the light (golden hour, morning sun, studio lightning, etc.) and selecting the camera viewing angle (bird' eye view, blurred motion, portrait photo) can help immensely.
- **Forcing high quality** (photorealistic, 8k, 4k, UltraHD, high res render, –q 5). These types of prompts do not affect the graphics and cannot produce higher-resolution images using them.
- **Power words** - especially those with a powerful dose of aesthetics. You probably know words like steampunk, cyberpunk, retro wave or hardcore. They give an amazing stylistic effect, affecting every aspect of the image - from technique through composition to details - but the style is undeniably recognizable. Try aesthetic combinations by adding -core,-punk, and -wave to the aesthetics. Explore existing ones and try new ones, such as solarpunk, dreamcore, and gothwave.

#### Parameters

Parameters are options that the Midjourney bot should consider when generating graphics. They can change image aspect ratios, switch between versions of the Midjourney model, increase chaos levels, and more. You can find their discussion here or in the Midjourney documentation.

Describing your vision well as a prompt is only half the battle. What you still need to remember is the seed. You will learn about it in the next post. Stay tuned! 

Midjourney guide part 3: Prompt structure

![blog-_grafika1 brzydki hacker.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika1_brzydki_hacker_d21723680e.png)

Every time I think about web application security, OWASP comes to mind. What is OWASP?

![blog-l grafika 2-logo owasp.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_l_grafika_2_logo_owasp_60c12ece3d.png)



OWASP stands for the Open Web Application Security Project. It is a nonprofit organization that focuses on improving the security of software. OWASP provides freely available tools, documentation, and resources for developers, security professionals, and organisations to improve the security of their web applications.

One of the key contributions of OWASP is the OWASP Top 10, which is a regularly updated list of the top ten most critical security risks facing web applications. This list helps developers prioritise their security efforts and address the most pressing vulnerabilities.

Any application built on top of an existing LLM or GenAI is essentially a web application, so OWASP also monitors this aspect. Let's explore the top 10 potential security risks when deploying and managing Large Language Models.

#### Prompt Injection

![blog-_grafika3 strzykawka.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika3_strzykawka_f05b3ac29e.png)

**What is it?**
The Prompt Injection Vulnerability arises when an attacker manipulates a large language model (LLM) using carefully crafted inputs, thereby causing the LLM to carry out the attacker's objectives unwittingly. This manipulation can occur either by directly breaching the system's prompt or indirectly through manipulated external inputs, which may result in various concerns such as data theft, social engineering, and other related issues.
		
**How it’s working?**
An attacker could inject biased language or leading questions into the prompts given to a language model, causing it to generate outputs that promote misinformation, hate speech, or other undesirable content.
    
**Real world example**
Application for HR, which uses an LLM to check if the uploaded CV fit the needs.
	    
In this example, a malicious actor uploads a document containing hidden instructions—a prompt injection—to manipulate a Large Language Model (LLM). The document suggests it's an excellent job application. When processed by an unsuspecting user through the LLM, the injected prompt alters the LLM's output to praise the document's quality falsely. This demonstrates how prompt injections can deceive LLMs into producing biased or manipulated results, posing significant risks to data integrity and decision-making processes.
		
**Prevention**
Prompt injection vulnerabilities exist due to the inherent nature of LLMs, which do not differentiate between instructions and external data. Because LLMs use natural language, they treat both forms of input as equally valid. As a result, there's no fool-proof prevention mechanism within the LLM to mitigate prompt injection risks, but you can lower the impact using strategies like:
- Implement privilege control for LLM access to backend systems
- Incorporate human oversight to enhance functionality
- Separate external content from user prompts
- Regularly monitor LLM input and output manually


#### Insecure Output Handling

![blog-_grafika_4 zatruty dokument.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_4_zatruty_dokument_99b2add539.png)

**What is it?**
Insecure Output Handling involves inadequate validation, sanitization, and management of outputs generated by large language models before they're used elsewhere. It can lead to XSS, CSRF in web browsers, as well as SSRF, privilege escalation, or remote code execution on backend systems.

**How it’s working?**
In the scenario where LLM output is directly inserted into a system shell or a similar function, it can lead to remote code execution. This means that if the LLM generates code or commands that are executed by the system, an attacker could potentially manipulate the LLM to produce malicious code, leading to unauthorized actions on the system, such as accessing sensitive data or compromising its security.
    
**Real world example**
A website which can share generated LLM content between different users
  
In this scenario, a web application utilizes a Large Language Model (LLM) to generate content based on user text prompts without properly sanitizing the output. This lack of output sanitization means that the content produced by the LLM is not checked for potentially malicious code or scripts before being presented to users.

An attacker could exploit this vulnerability by submitting a carefully crafted prompt to the web application, causing the LLM to return an unsanitized JavaScript payload as part of the generated content. When this content is rendered in a victim's browser, the embedded JavaScript payload executes, leading to a Cross-Site Scripting (XSS) attack.

**Prevention**
- Treat the model as any other user
- Follow the other good practices for input validation and sanitization
- Encode model output to mitigate undesired code execution by JS or Markdown


#### Training Data Poisoning

![blog-_grafika5 lejek złych i dobrych danych.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika5_lejek_zlych_i_dobrych_danych_84e01ece02.png)

**What is it?**
The critical aspect of any machine learning method lies in its training data, often referred to as "raw data." This data should encompass various domains, genres, and languages to achieve high capabilities, such as linguistic and worldly knowledge. Large language models utilise deep neural networks to generate outputs based on patterns learned from training data.

Training data poisoning involves manipulating pre-training data or data used in fine-tuning or embedding processes to introduce vulnerabilities, backdoors, or biases that could compromise the model's security, effectiveness, or ethical behavior. Poisoned information may surface to users or create risks like performance degradation, downstream software exploitation, and reputational damage.

**How it’s working?**
In this context, a malicious actor or a competitor brand intentionally creates inaccurate or malicious documents that are included in the dataset used for a model's pre-training, fine-tuning, or embedding processes.
	
This action is intended to manipulate the model's learning process, potentially leading to several adverse outcomes:
- Bias Introduction
- Model Degradation
- Security Risk
- Reputation Damage


**Real world example**
Brand A implemented a chatbot powered by a Large Language Model (LLM), designed to provide quick and efficient customer assistance with their inquiries and technical issues.
	
Competitor Brand B, seeking to undermine Brand A's reputation and gain a competitive edge, devises a nefarious plan. Brand B infiltrates the training data of a machine learning model. This could be done by injecting documents containing misinformation, biased content, or even malicious code.
		
As a result of the injected false training data, Brand A's chatbot begins to exhibit erratic behavior. It provides incorrect answers to customer questions, offers inappropriate solutions to technical problems, and, in some cases, may even respond with offensive or nonsensical replies.
	

**Prevention**
- Verify the supply chain of the training data
- Verify your use-case for the LLM 
- Ensure sufficient sandboxing through network controls is present to prevent the model from scraping unintended data sources
- Use strict vetting or input filters for specific training data
- Testing and Detection


####  "With great power comes great responsibility"

![blog-_grafika6 superhero.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika6_superhero_c5558433a6.png)

As Large Language Models (LLMs) become increasingly integrated into various applications, it's crucial to recognize their potential security vulnerabilities. While LLMs offer immense power and capabilities in natural language processing, they also introduce significant threats to application security.

In this article, we explore the top three security vulnerabilities associated with LLMs. These vulnerabilities can include prompt injections, output handling flaws, and training data poisoning. Each poses unique risks, ranging from unauthorized access to sensitive data to the introduction of biased or malicious outputs.

It's essential to understand that LLMs are not standalone entities but integral parts of application infrastructures. Therefore, addressing their security vulnerabilities is paramount for safeguarding overall system integrity.

To mitigate these risks effectively, companies and development teams are encouraged to stay informed about emerging security issues related to LLMs. Reading reports from organizations like OWASP can provide valuable insights into best practices and strategies for securing LLM implementations.

In conclusion, while LLMs offer transformative capabilities in natural language processing, they also bring significant security challenges. By remaining vigilant and proactive in addressing these vulnerabilities, organizations can harness the power of LLMs while minimizing potential security risks.


Read more:
https://owasp.org/www-project-top-10-for-large-language-model-applications/


Top 3 security vulnerabilities for Large Language Models

In the latest episode of PozTechTalks, Łukasz, Piotr, and Patryk delve into the realm of AI tools for developers. Serving as a continuation, this discussion builds on the insights and evaluations presented in the previous episode and accompanying article, providing a deeper exploration of innovative AI solutions tailored for programming challenges.

[**What the Diff**](https://whatthediff.ai/): This tool, aimed at enhancing pull request descriptions, integrates directly with GitHub. It distinguishes itself by generating comprehensive change summaries. However, its effectiveness varies, with occasional inaccuracies in generated descriptions, highlighting a mismatch between actual changes and the tool's interpretations. The tool also offers code review suggestions, which didn't perform well in tests, suggesting unrelated changes. Its usage is constrained by token limitations, making its free plan somewhat impractical for regular use.

[**Lintrule**](https://www.lintrule.com/): Differing from traditional linters, Lintrul allows for the creation of custom, human-readable rules for code consistency checks. It directly integrates with GitHub, enhancing code quality by enforcing project-specific conventions that might not be covered by standard linters. Despite its innovative approach, the need for precise rule definition and potential costs for extensive usage were noted.

[**ChatGPT**](https://chat.openai.com/auth/login): Not directly a development tool, ChatGPT's versatility is highlighted, showing its ability to analyse dependencies, generate code snippets, and interact with APIs using natural language. Its ability to browse the web and execute plugins extends its utility beyond simple Q&A, proving useful for various development-related tasks. However, reliance on this AI requires careful verification of its outputs due to occasional inaccuracies.


To conclude, the trio's exploration of AI tools reveals the evolving landscape of software development, where AI not only optimises workflow but also introduces new methodologies. Their discussions underscore the necessity of blending technical precision with creative experimentation to leverage AI's potential in programming fully. This insightful journey through AI tools not only broadens understanding but also invites developers to engage with these technologies, promising enhancements in efficiency and innovation.

Check out the full episode (in Polish)!


PozTechTalks 19

Exploring AI Tools for Developers: Insights from PozTechTalks part 2

Maintaining the consistency of the character created is one of the biggest challenges for Midjourney (and for the user). Midjourney is generative AI, which means that every time you prompt it, it creates a new image. Entering the exact string of words will give (more or less significantly) different results the next time - that's the whole idea of Midjourney. If you try to recall the same object, you are going against the rules of Midjourney - but that doesn't mean it's impossible.

#### The basis, i.e. seed


An important option here is understanding how the –seed parameter works. The Midjourney seed value itself has no particular meaning, as it is simply a number used to generate a random but consistent element of "noise" in the process. It sounds a bit strange, but "noise" is the initial step in creating graphics by Midjourney. It is simply a set of elements with specific characteristics (if you are very interested in this, you will find a note about how Midjourney generates its images here).

If you don't define a seed, Midjourney will use a random one each time it generates images. This will cause the generated graphics to be significantly different despite the same prompt.

By specifying a specific seed, you ask Midjourney to use the same starting point (noise) for your prompt. This helps significantly maintain the consistency of subsequent images.

#### How to find an image seed?

Step 1: Click the Discord message reaction button.
Step 2: Select the standard envelope emoji. This will launch its replies from the Midjourney bot.
Step 3: When the bot sends you a private message, open it.
Step 4: Copy the seed number from the message.

![blog-_grafika.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_b376bd6aac.png)

The bot usually responds within a few seconds and sends a message containing the following details:

- Full prompt
- Task ID
- Seed number
- Multimedia files
- Direct link to the news

#### And how to use this seed?

We enter seed as a parameter at the end of our—seed XXXX command (where we enter the number provided by the Midjourney bot). It is also best to leave the original prompt and consider what we need to change.

The most important thing here is to make small changes - rewriting the entire prompt with new things will only generate a similar style and not add elements to the existing image. From such a portrait, we will not get a silhouette of the whole figure riding a horse in the cybernetic desert. But at this stage, we can add elements to the portrait or change the parts that interest us. It's easy to change your facial expression or clothes; in this respect, it's a handy feature.

![blog-_grafika-1.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_1_8fd130c9cc.png)

You can clearly see that small changes remain highly consistent with the initial image. However, the further the change - in style image - the more difficult it is to maintain the same image.

Summary of the relationship **between seed and prompt depending on the MJ version**.

Here are the possible results of combining seed and prompt (for 5.2):

- prompt + no seed = completely different results
- prompt + seed = almost identical results
- prompt with change + seed = similar results, including change.

Here are the possible results of combining seed and prompt (for 5.0 and 5.1):
- prompt + no seed = completely different results
- prompt + seed = identical results
- prompt with change + seed = very similar results, including the change.
- TAGS

Here are the possible results of combining seed and prompt (for 6.0):
prompt + no seed = completely different results
prompt + seed = similar results 
prompt with change + seed = similar results, including the change.

Version 6 was introduced with two additional parameters to maintain image consistency, hence probably the lower convergence in the seed parameter. The sref and cref options allow for more precise recall of a style or character, and I will discuss them in more detail in the next article.

In the next article, I will present in detail the parameters that need to be considered when creating graphics using artificial intelligence. 


Midjourney guide part 4: seed

#### ASPECT RATIO

With its precise control, the –aspect or –ar parameter changes the aspect ratio (not size!) of the generated image. Aspect ratio is the ratio of the width to the height of an image. It is usually expressed as two numbers separated by a colon, such as 7:4 or 4:3. A square image has equal width and height, known as an aspect ratio of 1:1. The image can be 1000px × 1000px or 1500px × 1500px, and the aspect ratio will still be 1:1.

The best illustration showing the difference -ar is straight from the Midjourney documentation:

![blog-_grafika.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika_e577224849.png)


**Important properties**

- the default aspect ratio is 1:1
- since version 5.0, arbitrary aspect ratios are supported
-  –ar must use integers
- aspect ratio has an impact on the shape and composition of the generated image (and a significant one)

#### STYLIZE

The -stylise or -s parameter, a key to unlocking artistic potential, affects how much Midjourney favours artistic colours, compositions, and forms. Low styling values create images that closely match the prompt but are less artistic. High styling values create images that are very artistic but less related to the prompt. Experimenting with the stylised value also greatly changes the generated image. You can also set the default value in /settings.

![blog-walentynki-midjourney_high.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_walentynki_midjourney_high_f6cb630c53.png)

Selecting the stylise option also sets the default value of the parameter:

- RAW Mode 0
- Stylise low 50
- Stylise med 100
- Stylise high 250
- Stylise very high 750

**Important properties**

- The default value of –stylise is 100.
- –style takes the values 0-1000.


#### STYLE 

The –style parameter allows you to customise the aesthetics of some versions of the Midjourney model. Adding a style parameter can help you create photorealistic images, movie scenes, or prettier characters. Model versions 5 and 6 accept –style raw. **The –style raw parameter reduces the impact of Midjourney's default aesthetic** and works well for advanced users who want more control over their images or more photographic images. This means that Midjourney then focuses more on reproducing the text prompt.

The Niji 5 version accepts –style cute –style scenic –style original, or –style expressive.


#### CHAOS 

The -chaos or -c parameter affects the diversity of the initial image meshes. High chaos values produce more unusual and unexpected results and compositions. Lower-chaos values have more reliable, repeatable results.

**Important properties**
- The default value of –chaos is 0.
- –chaos takes values 0-100.


#### WEIRD 

The experimental–weird parameter introduces strange and unusual features into the generated images, resulting in unique and unexpected results.

**Important properties**

- The default value of –weird is 0.
- –weird takes the values 0-3000.
- –weird is not compatible with seed
- when using stylise at the same time - the values should be the same


#### NO

The -no parameter tells Midjourney what not to include in the image. Entering what you don't want into the prompt may cause artefacts and, unfortunately, does not help much in eliminating unnecessary elements. Saving the elements (tokens) you do not want to include in the –no parameter gives you a greater chance of making them disappear.

**Important properties**
- basic parameter for working with exclusions
- –no accepts multiple words separated by commas: –no word1, word2, word3, phrase1, phrase2
- if you want to use a comma in the content -no, precede it with the / sign.


#### TILE 

The –tile parameter generates images that can be used as repeating tiles to create patterns for fabrics, wallpapers, and textures.

#### QUALITY

The –quality or –q parameter changes the amount of time spent generating the image. Higher-quality settings take longer to process and produce more detail. Higher values also mean more GPU minutes are used per task. The quality setting has no effect on the resolution. Currently recommended for creating quick sketches (at reduced quality), but switching to higher quality often changes the image significantly.

**Important properties**
- The default –quality value is 1.
- –quality only accepts values: .25, .5 and 1 for model 4 and above (larger values are rounded down to 1)
- –quality only affects the initial image generation.

Parameters indirectly affecting the image.

### VERSION

The latest model is always the default, but other models can be used by adding the—version or—v parameter or by using the /settings command and selecting the model version. Different types of images distinguish each model.

Important properties
- –v 5.2 is the current default model.
- –version takes the values 1, 2, 3, 4, 5, 5.1,  5.2 and 6.

**Niji Model 6**

The Niji model is a collaboration between Midjourney and Spellbrush, tailored to create anime styles and illustrations with much more knowledge of anime, anime styles, and anime aesthetics. Perfect for dynamic action shots and character-centric compositions. To use this model, add the –niji parameter to the end of the prompt or use the /settings command and select Niji version. By default Midjourney will use the latest version.


#### REMIX 

**Remix mode allows you to change hints, parameters, model versions or aspect ratios between variants**. Remix will take the overall composition of the starting image and use it as part of the new task. Remixing can help you change the positioning or lighting of an image, expand an object, or achieve difficult compositions.

The best thing we can change with a remix is also described in the Midjourney documentation.

![blog-DecodingReactUI_blog_01.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Decoding_React_UI_blog_01_0f660cd570.png)

#### SEED

Midjourney uses a seed number to create a starting point for generating initial image grids. The numbers are randomly generated for each image but can be specified using the –seed parameter. You will get similar final images if you use the same seed value and the same prompt.

**Important properties**
- –seed accepts integers 0-4294967295.
- identical –seed and prompt values using model versions 1, 2, 3, test and testp will produce images with similar composition, colour and detail.
- Identical –seed and prompt values using model versions 4, 5 and niji will produce almost similar images.

Seed numbers are not static and should not be relied upon between sessions.


#### REPEAT

The –repeat or –r parameter runs the task repeatedly. Combine –repeat with other parameters increases the pace of visual exploration and speeds up the generation of several meshes (Because we don't give up after one /imagine).

**Important properties**
- The –repeat parameter is available to Standard and Pro subscribers
- –repeat accepts values 2-10 for Standard subscribers.
- –repeat accepts values 2-40 for Pro subscribers.

The –repeat parameter can only be used in Fast mode.


#### STOP 

The –stop parameter terminates the task mid-process. Stopping the task earlier may result in a blurry and less detailed image.

**Important properties**
- –stop takes the values 10-100.
- the default value of –stop is 100.
- –stop has no effect when scaling up.


#### VIDEO
Use the –video parameter to create a short video of the initial image mesh generation. React to your completed work with an envelope ✉️ emoji to have Midjourney Bot send the video link to your direct messages.

–video only works for image grids, not upscaling.
–video works with model versions 5, 5.1, 5.2 and niji 5 and older model versions 1, 2, 3, test and testp.


Midjourney guide part 5: Parameters

In our latest newsletter, we've gathered the most fascinating updates from the tech world, sprinkled with intriguing tidbits from its many orbits. Our subscribers were the first to get their hands on these cutting-edge insights, delivered straight to their inboxes. 

From the groundbreaking launch of OpenAI’s GPT-4o to the thrilling countdown to the Paris 2024 Olympics, we’ve covered it all. Stay ahead of the curve with Bitnoise and explore the innovations shaping our future. Check out the highlights now and see why it's worth subscribing! 

Explore the latest from OpenAI, countdown to the Olympics, and meet the Polish Enigma code breaker!

#### GPT-4o and additional tools are now available to free users of ChatGPT

GPT-4o, the newest flagship model from OpenAI, offers GPT-4-level intelligence but operates faster and enhances capabilities in text, voice, and vision.
To broaden AI's global accessibility and utility, GPT-4o has improved language functionalities in terms of quality and speed, supporting over 50 languages for various user interactions. Are you curious how it might work in practice? Be sure to check out [this moment](https://www.youtube.com/watch?v=DQacCB9tDaw&t=1349s&ab_channel=OpenAI) from
yesterday's presentation.

[Read more about it!](https://openai.com/index/gpt-4o-and-more-tools-to-chatgpt-free/)


#### Tool of the month

![stackradar_tool of the month.png](https://btnw-strapi-prod.dev.btndev.space/uploads/stackradar_tool_of_the_month_ef4a30257d.png)

Stackradar offers a curated selection of the best tech tools available online, designed to help individuals enhance and expand their work. At Stackradar, a tech stack is seen as a set of business operators, each playing unique roles and interacting with other departments, offering cost efficiency compared to traditional departments.

Check what others use, get inspired and find your own tool of the month!

[Check it out!](https://www.stackradar.co/)

#### Trending

[1. The latest major update to ChatGPT-4o now supports audio-video interactions with an AI chatbot that can express “emotions”. Check out more news from the presentation.](https://arstechnica.com/information-technology/2024/05/chatgpt-4o-lets-you-have-real-time-audio-video-conversations-with-emotional-chatbot/)
[2. Bloomberg reports that Apple is close to finalizing a deal with OpenAI to integrate ChatGPT features into iOS 18.](https://9to5mac.com/2024/05/10/ios-18-chatgpt-features-apple-openai/)
[3. Paris 2024 is preparing to tackle an unprecedented level of cybersecurity threats.](https://oebmgr.clicks.mlsend.com/tj/cl/eyJ2Ijoie1wiYVwiOjg0NzIxMixcImxcIjoxMjEzODE3NzYxNTQxNjcyNzgsXCJyXCI6MTIxMzgxNzc3Mjc4MjM5OTgzfSIsInMiOiI3ZWRiNDA4MzAzMGFlNGI5In0)

![nagłówek newsletter.png](https://btnw-strapi-prod.dev.btndev.space/uploads/naglowek_newsletter_f1f3c5940f.png)

#### Around the Web

🍎 Did you know that Alan Turing wasn't the first to decipher the Enigma code? [Tune into these interviews](https://www.flyingpenguin.com/?p=56989) to uncover more about this fascinating piece of history!

⚖️ TikTok and ByteDance are [taking legal action against the US government](https://www.theguardian.com/technology/article/2024/may/07/tiktok-sues-us-ban?CMP=share_btn_url&utm_source=substack&utm_medium=email) to prevent the
enforcement of a law that mandates the sale of the app.

🤖 Kids are forming [friendships with AI chatbots](https://www.theverge.com/2024/5/4/24144763/ai-chatbot-friends-character-teens).

🎤 Deciphering in pop culture. [Switzerland wins Eurovision with song 'The Code’.](https://eurovision.tv/story/switzerland-wins-eurovision-2024)

#### What's new in Bitnoise

![blog-Midjourney guider_blog okładka.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_Midjourney_guider_blog_okladka_feac2e3ddd.png)

Are you ready to begin your journey with MidJourney? Kaja, our Project Manager, has crafted a comprehensive series of articles that serve as a complete guide!

[Check it out!](lhttps://www.bitnoise.pl/blog/midjourney-guide-part-1-simple-thing-settings)

large_blog-newsInsights_cover_high.png

medium_blog-newsInsights_cover_high.png

small_blog-newsInsights_cover_high.png

thumbnail_blog-newsInsights_cover_high.png

blog-newsInsights_cover_high.png

Unveiling the Latest Tech Insights: May Newsletter Highlights


Maintaining the consistency of the generated image - to apply it in later ones and create a coherent story - is a huge challenge for Midjourney. The puppet method is tedious and rigorous, but it allows for very satisfactory results - especially for individual characters.

The puppet method uses a very specific prompt template and allows for relatively efficient operation of a previously generated character.

The template looks like this:
![blog-_grafika1.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika1_f3d2858794.png)

So let's discuss all aspects one by one:

Puppet template:

**1. Character image**

**2. Background description**

**3. Character description**

**4. Description of activities – character animation**

**5. Extras – atmosphere and parameters**


You should use this template with all keywords - then they act as a reusable puppet (actor, character). You can change the setting, activity, atmosphere, media type, etc., around her to direct this character. Later, once you've mastered the template, you can experiment with removing or rearranging keywords.

![blog-_grafika2_2.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika2_2_d3bf85d54f.png)


The template is based on weighted phrases, strong keywords and a specific sequence. As you use it, it becomes easier and easier to modify. To complete it you need:

#### Character image

The character reference image should roughly reflect the character's appearance when Midjourney integrates it into the final scene. You can use a ready-made photo or a previously generated image—you need a link (to the image on Discord or online), which will be a model of the character. It is important that the character "fits" the scene.

If you provide a reference in the form of a cropped head-to-shoulder portrait, remember that Midjourney will have difficulty coming up with a body for the character and will likely ignore attempts to show the entire body. And if you provide an image in which the character's face is only a few pixels in size, remember that Midjourney will have a hard time coming up with a face, and the results showing the character's face may be completely different. Therefore, it is essential to roughly match the reference image to its appearance in the final image. The character image must have a simple or blank background to help Midjourney focus on the character.

**REMEMBER**! Of course, there is no guarantee that the character will look like the image provided. Reference graphics only have influence, so you won't be able to recreate an exact portrait of yourself or your loved ones, but you have a chance to get a really similar result. You can also increase the weight of the reference image using multiprompts.

##### EXAMPLE

Using the prompt: photoshoot of blue-hair pixie-cut hipster-girl, full body, white background, I generated Lizzy. The portrait below has a light background a clear face, and is cropped in half - this suits me because I don't want to create the entire silhouette and a blurred face, but rather place it in various situations in this close-up. So – this is Lizzy.

![blog-_grafika3.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika3_6ddc2a4210.png)

#### Background description

The background description begins with "Setting is" (case sensitive!) and can be supplemented with both text and an image. 

The important thing here is to describe the scenery as precisely as possible. These should be short, characteristic phrases that clearly define Midjourney, where our puppet is to be. We do not use verbs then - we try to provide a static description of natural circumstances. You can also use a reference image for the background - then remember that your prompt should match it - that is, describe what is actually in the scenery. 

In this case, we set the scenery to a night vision of the city and its rooftops.
![blog-_grafika4.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika4_11e801ffed.png)

Pretty good, right? It's definitely Lizzy - our chosen character. The last two pictures clearly show the city at night. Lizzy is reading a book, as we indicated in the prompt. Let's move it to some other, more appropriate place…
![blog-_grafika5.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika5_98bb9c9d66.png)

Lizzy is still reading the book - but the environment has changed. He's not on the roof anymore, just... that's right. Not very library-y, right? It was supposed to be an old-school library, but there were a few books. So, I will generate and add a background reference image:
![blog-_grafika6.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika6_446f10dfea.png)

Oh, a very nice bookcase, a different atmosphere of the place, so I'll add both a link and a description to the prompt:
![blog-_grafika7.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika7_eec74e1768.png)

And this is much closer to the effect we wanted to achieve.

The entire prompt remains unchanged—it only adds a link to the background. It is also important to maintain a similar (or preferably the same) background description—"vintage library with bookshelves"—as when generating it. If you use a ready-made photo, describing what is in it is important. The discrepancy between the reference image and its description may lead to interesting effects, but it will certainly add artefacts and weaken control over the character. 

#### Character description

The text description should always start with the Name is (case sensitive) and consist of the character's characteristic features. There is also a part here where you give the character a name - it should be as neutral as possible. If you choose the name Mulan and then create a blonde girl - you may find that she doesn't look exactly like a Swedish girl. However, remember - archetypes are on your side and choosing a super-characteristic name can also have its advantages.

But be careful! This is not the place to describe what the character does - only the most characteristic features that define him as a character. You can dress up the character and make a puppet out of him in a moment - not now. Now focus on a short and focused description

If you generated a character using Midjourney - it is worth using the description that created it. If it is long, use the /shorten command to see which keywords Midjourney considers the most important. In the examples given, I constantly use the exact description of Lizzy that I used to create her.  

####  Description of activities – character animation

A text description of a character doing something in a scene or setting must begin with Setting contains (case sensitive). Now you can "move" your puppet. But don't go crazy - the activity description should be limited to two or three elements.

It's best and easiest to change things one at a time, such as:
- character wardrobe
- character emotions
- key activity


So, in the image below, we dress Lizzy up in a scarf and jacket.
![blog-_grafika8.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika8_03d373620f.png)


While elements such as clothing or expressed emotions are easy to change here, and Midjourney usually has no problem, assigning the action performed is more complicated. Everything the character does should be matched to the character's reference image. Do you have a full figure? Forget biting your lip. Do you have a portrait painting? Jumping over obstacles won't be good. 

#### Extras – atmosphere and parameters

The last space is left for adding a few keywords and parameters. Try not only to maintain style but to enhance it. Notice that in the previous two examples, when Lizzy is in New York wearing a scarf and jacket - the prompt is - photo by Ann Leibowitz. I chose an emotional, well-known artist to suggest to Midjourney that I wanted a piece of street photography. In the example of escaping from a dinosaur, I added an "action movie cinematic photo" to get a dynamic image like an action scene, not a nostalgic portrait. You can also set all other parameters of the generated graphics here - version, aspect ratio, style - just like in any other.

![blog-_grafika9.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_grafika9_a76f67463b.png)


BONUS TIPS

Multiprompts divide the template into appropriate sequences ending with two colons. Before you start modifying them, make sure you know the rules of multiprompts. Increasing the weight of any multiprompt sequence (background, animation, etc.) can strengthen its impact on the generated graphics.


Midjourney guide part 6: The puppet method

In our latest newsletter, we've gathered the most fascinating updates from the tech world, sprinkled with intriguing tidbits from its many orbits. Our subscribers were the first to get their hands on these cutting-edge insights, delivered straight to their inboxes. 

From the AI clones that will be able to participate in video calls to the the world's thinnest lens, we’ve covered it all. Stay ahead of the curve with Bitnoise and explore the innovations shaping our future. Check out the highlights now and see why it's worth subscribing! 


### Digital avatar clones, the world's thinnest lens, virtual IKEA and META has some trouble, again


#### AI clones will be able to participate in online meetings on your behalf

Eric Yuan, CEO of Zoom, aims to expand the company beyond videoconferencing to compete with Microsoft and Google by offering a suite of enterprise tools, including email, documents, and chat. A central element of this vision is the use of AI, especially in creating "digital twins"—AI avatars capable of attending meetings and making decisions for users.

He acknowledged the technological and cultural shifts required to achieve this, emphasizing the need for personalized AI and ongoing advancements in AI and AR technologies. Eric believes AI can significantly reduce manual work, potentially leading to shorter workweeks.


[Read more about it!](https://www.theverge.com/2024/6/3/24168733/zoom-ceo-ai-clones-digital-twins-videoconferencing-decoder-interview)

#### Tool of the month

![blog-AI Tools for Developers_high (1).png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_AI_Tools_for_Developers_high_1_503d656e1e.png)

Elevenlabs has created a tool that generates any sound effect from a text prompt, streamlining the audio production process. Users can quickly produce a variety of sound effects simply by describing their needs. The tool offers precise control, allowing customization of sounds such as "footsteps" by specifying details like shoe type and environment. All paid subscriptions come with a comprehensive commercial license, guaranteeing royalty-free use.

[Check it out!](https://elevenlabs.io/sound-effects)

#### Trending

1. [On June 6, privacy advocacy group NOYB lodged complaints in 11 European countries, alleging that Meta is using personal data from its platform users to train artificial intelligence models.](https://www.euronews.com/next/2024/06/06/meta-faces-complaints-in-11-countries-over-data-use-for-ai)
2. [Elections to the European Parliament.](https://www.reuters.com/world/europe/european-parliament-poised-rightward-shift-after-final-voting-2024-06-09/)
3. [Apple's Worldwide Developers Conference (WWDC) 2024: 10-14 June.](https://developer.apple.com/wwdc24/)

#### Around the Web

👁️ New "Recall" function from Microsoft AI - [security issues](https://www.malwarebytes.com/blog/news/2024/05/microsoft-ai-recall-feature-records-everything-secures-far-less?fbclid=IwZXh0bgNhZW0CMTAAAR15R8YOt-VaXVYViW-R3TCtuYmzCnA92uZouriI0GSF2Hlxsiud5BKOwDE_aem_AZCm_r1z6t3S3F6xt7nAFBShHCSk1jOKUJH7uHds1CCZgSj1tOSwjtpQr474-2IpbPV-BmZpBgfoD2ogCm_UUIG5).

🚨 Adobe intends to [train its AI models using user-generated content?](https://alternativeto.net/news/2024/6/adobe-faces-backlash-over-new-photoshop-tos-granting-unlimited-access-to-user-projects/?fbclid=IwZXh0bgNhZW0CMTAAAR2uNHcXDm7x--FSpGN4h_9SQsLgx_SYW5j1ulrLb_KvW1hIkitWLK7eLHA_aem_AZA9tsiaerAcPR-li0nPTSFLp97JFdSJgeJw7ohwNjZXmaOldTzFyCUNeAIleRqi6sggsJgc8hgR7Yvv2G31aPiC)

🔥 Scientists have developed [the world's thinnest lens](https://scitechdaily.com/just-three-atoms-thick-scientists-have-developed-the-worlds-thinnest-lens/), just three atoms thick.

🎮 IKEA wants to pay people to [work in its new store inside Roblox game.](https://time.com/6985642/ikea-hiring-workers-roblox-game/)

#### What's new in Bitnoise

![blog-_cover.png](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_cover_1187589fac.png)

Kaja, our Project Manager, has crafted a comprehensive series of articles that serve as a complete guide!

**Midjourney guide part 6: The puppet method**
Maintaining the consistency of the generated image - to apply it in later ones and create a coherent story - is a huge challenge for Midjourney. The puppet method is tedious and rigorous, but it allows for very satisfactory results - especially for individual characters.

[Check it out!](https://www.bitnoise.pl/blog/midjourney-guide-part-6-the-puppet-method)

large_blog-newsInsights_cover_high (1).png

medium_blog-newsInsights_cover_high (1).png

small_blog-newsInsights_cover_high (1).png

thumbnail_blog-newsInsights_cover_high (1).png

blog-newsInsights_cover_high (1).png

Unveiling the Latest Tech Insights: June Newsletter Highlights

As technology continues to advance, so does the functionality of AI-driven tools. Chat GPT extensions have been developed to cater to various needs and preferences. Planty, Tattoo GPT, and Wine Sommelier stand out for their unique functionalities. This article delves into these three extensions' basic functions, advantages, and drawbacks.

![Planty.jpg](https://btnw-strapi-prod.dev.btndev.space/uploads/Planty_32ebe84ead.jpg)

#### Planty


##### Basic Functions:
Planty is designed to provide detailed care instructions for various plants. Whether you are a novice gardener or an experienced botanist, Planty offers tailored advice on watering schedules, sunlight requirements, soil types, and pest control.

##### Advantages:

1. Comprehensive Database: Planty has a vast database of plant species, making it a reliable source of information for plant care.
2. User-Friendly: The interface is intuitive, making it easy for users to input their plant’s information and receive detailed care instructions.
3. Personalisation: Planty can customise advice based on specific environmental factors such as local climate and indoor conditions.

##### Drawbacks:
1. Limited Interaction: While Planty provides excellent static advice, its interactivity is limited. It needs to work on follow-up questions and nuanced problems.
2. General Advice: Sometimes, the advice can be too generic, lacking the specific touch needed for unique plant issues.


![Tatto GPT.jpg](https://btnw-strapi-prod.dev.btndev.space/uploads/Tatto_GPT_0a56061c52.jpg)

#### Tattoo GPT


##### Basic Functions:
Tattoo GPT is an extension for designing tattoos. It allows users to input design preferences and receive custom tattoo designs in return.

##### Advantages:
1. Customisation: Users can request personalised tattoo designs based on their specific preferences.
2. Visual Input: The extension allows users to upload images as inspiration for their tattoo designs.

##### Drawbacks:
1. Command Sensitivity: Tattoo GPT often struggles with understanding complex commands. For instance, if asked to change one element of a design while leaving another intact, it might alter the entire design instead.
2. Monotony: The generated designs can be somewhat repetitive and lack variety, which is disappointing for users seeking unique inspirations.
3. Intellectual Property Issues: When users request designs featuring elements from well-known franchises like Harry Potter or Lord of the Rings, copyright restrictions prevent the generation of such graphics. Although switching to more universal motifs like moons and stars speeds up the process, the output often lacks a strong connection to the original inspiration.
4. Inconsistent Output: Even when given clear instructions, the results can sometimes be off-mark, as seen with designs that merely stay within a grayscale palette without much creativity.


![Wine Sommelier.jpg](https://btnw-strapi-prod.dev.btndev.space/uploads/Wine_Sommelier_a57b3cee2e.jpg)

#### Wine Sommelier


##### Basic Functions:
Wine Sommelier aims to provide wine recommendations tailored to the user’s taste preferences. It suggests specific wines, pairing options, and purchasing locations.

##### Advantages:
1. Tailored Recommendations: When given information about personal preferences, such as a user’s favourite type of whiskey, the Wine Sommelier can suggest wines that align with those tastes.
2. Brand Suggestions: Upon request, it can provide specific wine brands and where to purchase them, including options available in local supermarkets.
3. Practical Advice: The extension’s ability to recommend both premium and supermarket wines makes it versatile for different budgets.

##### Drawbacks:
1. General Suggestions: Initial recommendations tend to be broad, focusing on wine varietals rather than specific labels or regions unless explicitly asked.
2. Lack of Proactive Queries: The Wine Sommelier does not inherently ask detailed questions about user preferences, requiring users to volunteer additional information to refine the recommendations.


##### Close, but not there yet

Each of these GPT extensions brings something valuable to the table, catering to specific needs and preferences. Planty is ideal for plant enthusiasts seeking detailed care instructions but might leave those with unique issues wanting more. Tattoo GPT offers the potential for personalised tattoo designs but requires clear and straightforward commands for best results and struggles with creative and legal constraints. Wine Sommelier is a handy tool for wine recommendations, though it shines brightest when users provide detailed information about their tastes.

Despite their drawbacks, these extensions highlight AI's incredible potential for personalised user experiences. As these tools continue to develop, their capabilities will likely expand, offering even more refined and tailored solutions for users worldwide.


Exploring GPT-Extensions: Planty, Tattoo GPT, and Wine Sommelier

### AI at the Olympics, two halves of an apple and Google makes boredom go away

In our latest newsletter, we've gathered the most fascinating updates from the tech world, sprinkled with intriguing tidbits from its many orbits. Our subscribers were the first to get their hands on these cutting-edge insights, delivered straight to their inboxes. 



#### Google Meet will soon be able to take notes during your calls automatically.

![news GOOGLE MEET.jpg](https://btnw-strapi-prod.dev.btndev.space/uploads/news_GOOGLE_MEET_f6e270da41.jpg)

Have you ever wished your Google Meet calls could take notes for you? Well, that wish is about to come true! Google is rolling out an exciting new feature that automatically generates notes during your meetings, saving you time and helping you stay focused on the conversation. This new addition is perfect for busy professionals who juggle multiple meetings and want to ensure they get all the details. Want to know more about how this game-changing feature works? 

[Check out](https://www.techradar.com/pro/good-news-your-google-meet-call-will-soon-be-able-to-take-notes-for-you?_bhlid=576f4a4de257c6aa0a3886a6cbb5ce34d6294bd1) the full article to get the inside scoop!


#### Tool of the month

![news FLUX.jpg](https://btnw-strapi-prod.dev.btndev.space/uploads/news_FLUX_f633650130.jpg)

Flux is an innovative AI model designed for image generation, functioning similarly to the widely known Midjourney. It utilises deep neural networks, specifically diffusion models, to convert text descriptions into visual content. The process begins when a user inputs a text description of the desired image. The AI model then interprets this text through its neural network, "understanding" the content and producing an appropriate image based on the description.

What sets Flux apart? It's open source. Interested? [Give it a try!](https://blackforestlabs.ai/#get-flux)


#### Trending 

1. Artificial intelligence is revolutionising the sports industry. [The Olympic AI Agenda](https://olympics.com/ioc/olympic-ai-agenda), the third in a series of strategic documents spearheaded by IOC President Thomas Bach, outlines the potential transformative effects AI could have on the world of sports.
2. [Intel is on a lengthy path to recovery](https://www.theverge.com/2024/8/1/24210656/intel-is-laying-off-over-10000-employees-and-will-cut-10-billion-in-costs), and over 15,000 employees will not continue the journey. The chipmaker has just revealed plans to reduce its workforce by more than 15 percent as part of a new $10 billion cost-saving initiative for 2025. This reduction translates to a potential loss of over 15,000 jobs, with layoffs possibly reaching up to 19,000 people. 
3. OpenAI’s popular ChatGPT AI has begun responding to voice commands and conducting conversations with users. The latest feature, called [Advanced Voice Mode (AVM)](https://arstechnica.com/information-technology/2024/07/when-counting-quickly-openais-new-voice-mode-stops-to-catch-its-breath/), is impressive in its effects but also a bit scary.


#### Around the web 

⚖️ Judge Rules ['Google Is a Monopolist'](https://www.nytimes.com/2024/08/05/technology/google-antitrust-ruling.html) in Landmark Antitrust Case
The decision on Google's search dominance marks the first antitrust ruling of the modern internet era against a major technology company.

🧜‍♀️ Changing of the guard at [Starbucks](https://www.geekwire.com/2024/brian-niccol-who-led-chipotles-digital-push-named-new-ceo-at-starbucks/). What impact will a new CEO have on the coffee giant?

🍏 Foldable Apple? Why not! New iPhone [coming soon](https://www.reuters.com/technology/apple-likely-release-foldable-iphone-early-2026-information-reports-2024-07-23/). 

🕶 [Gemini ](https://blog.google/products/gemini/made-by-google-gemini-ai-updates/?_bhlid=38eb5834b5ee043160c764f22bc38870316d39ef)transforms your mobile device into a powerful AI assistant.


#### What's new at Bitnoise

![blog_cover_high.jpg](https://btnw-strapi-prod.dev.btndev.space/uploads/blog_cover_high_2db7c5d85d.jpg)

Excited to dive into the first part of our new blog series: "Backend Architecture on Frontend" by Łukasz! 

Discover:
🤖 how frameworks like Next.js and Remix are bridging the gap between frontend and backend development, 
🤖 why architectural discipline is crucial, 
🤖 and how adopting principles like GRASP can bring order to your codebase. 

Perfect for developers looking to streamline complex interfaces with simple APIs. Stay tuned for practical tips on creating a robust architecture that meets your actual needs.

[Check on your own! ](https://www.bitnoise.pl/blog/backend-architecture-on-frontend-part-1)


large_blog-newsInsights_cover_high.jpg

medium_blog-newsInsights_cover_high.jpg

small_blog-newsInsights_cover_high.jpg

thumbnail_blog-newsInsights_cover_high.jpg

blog-newsInsights_cover_high.jpg

Unveiling the Latest Tech Insights: August Newsletter Highlights

#### Why do we blindly trust AI, what Elon came up with, and yes, we saw the new iPhone


In our latest newsletter, we've gathered the most fascinating updates from the tech world, sprinkled with intriguing tidbits from its many orbits. Our subscribers were the first to get their hands on these cutting-edge insights, delivered straight to their inboxes.

#### [Telegram](https://www.bbc.com/news/articles/cvglp0xny3eo) will now provide some user data to authorities

![newsletter#7 MAIN TOPIC.jpg](https://bitnoise-cms.btn.fyi/uploads/newsletter_7_MAIN_TOPIC_8061c9ab8c.jpg)

The messaging app Telegram said it will hand over users' IP addresses and phone numbers to authorities with search warrants or other valid legal requests.

Telegram has historically been known for its strong stance on user privacy, encryption, and resistance to government interference. However, in recent years, there have been instances where the platform has complied with legal authorities under certain conditions, particularly in cases involving serious crimes, such as terrorism or child exploitation.

[Read more](https://www.bbc.com/news/articles/cvglp0xny3eo). 



#### Tool of the month 

![notebookLM_tool of the month.jpg](https://bitnoise-cms.btn.fyi/uploads/notebook_LM_tool_of_the_month_e12ad5e4f2.jpg)

NotebookLM is an experimental AI-based tool from Google that uses the Google Gemini model. As the authors say - It helps you “think your best” by reading, taking notes, asking questions, and organising your ideas. In other words, NotebookLM is like an AI-based research assistant that helps users better understand, summarise, brainstorm, and generate ideas based on their submitted sources. NotebookLM excels in workflows that involve synthesising information spread across multiple documents. Some of the main features of this tool are uploading multiple sources, using the Notebook Guide, asking questions, taking notes, changing articles into a podcast, and suggesting actions.

[Check it out](https://notebooklm.google/). 

#### Trending

1. [It’s all about the money](https://www.reuters.com/technology/artificial-intelligence/openai-chair-says-board-has-discussed-equity-compensation-ceo-sam-altman-2024-09-26/). The chair of OpenAI mentioned that the board has considered offering equity compensation for CEO Sam Altman.
2. [You have been blocked.](https://www.engadget.com/social-media/x-is-nerfing-the-block-button-blocked-users-will-be-able-to-see-your-posts-214928396.html) But not really. X is weakening the block button: Blocked users will still be able to view your posts.
3. [iPhone 16 is here.](https://www.techradar.com/phones/iphone/iphone-16-review) But is it really worth the upgrade? 

#### Around the Web

🏰 We are all part of one big [Disney](https://www.techradar.com/streaming/disney-plus/disney-plus-unveils-password-sharing-pricing-ahead-of-its-impending-price-hike-and-now-i-m-not-sure-i-ll-be-able-to-finish-agatha-all-along) family. Disney Plus’ password-sharing fees are official—here’s how much they cost. 

🧭 [The CTO is leaving Open AI](https://www.engadget.com/ai/cto-mira-murati-is-the-latest-leader-to-leave-openai-200230104.html) because “she wants to create the time and space to do her own exploration”.

❌ [Advanced AI chatbots](https://www.engadget.com/ai/advanced-ai-chatbots-are-less-likely-to-admit-they-dont-have-all-the-answers-172012958.html) are less inclined to acknowledge that they don't have all the answers. The study also revealed that people tend to accept incorrect answers from bots too quickly.

#### What's new at Bitnoise

🚀 The first part of our series on the #InklusionDigitalSaga is live! 

In the first article, Luiza Kwiatek explores why the Discovery and Design Phase is critical to product success. 

Learn how taking time to thoroughly research, plan, and design before development can lead to more innovative and user-friendly solutions and save much money! If you want to understand how strong foundations lead to great products, this one’s for you.

[Check it out](https://www.bitnoise.pl/blog/the-discovery-and-design-phase-as-the-main-component-of-product-success-the-inklusion-digital-saga-part-1)! 


large_blog-newsInsights7_cover_high.jpg

medium_blog-newsInsights7_cover_high.jpg

small_blog-newsInsights7_cover_high.jpg

thumbnail_blog-newsInsights7_cover_high.jpg

blog-newsInsights7_cover_high.jpg

Unveiling the Latest Tech Insights: October Newsletter Highlights

#### OpenAI's new direction: Creating a sustainable AI ecosystem

OpenAI’s DevDay 2024 signifies a strategic shift, focusing on ecosystem development rather than high-profile product announcements.

Though this strategy may seem less thrilling to the broader audience, it reflects a deeper, more mature grasp of the AI industry’s present challenges and opportunities.

What new things can we expect?

#### [Prompt caching](https://platform.openai.com/docs/guides/prompt-caching): A potential game-changer for reducing developer costs

One key announcement is the launch of Prompt Caching, designed to cut developers' costs and latency by offering a 50% discount on recently processed input tokens. This will lead to significant savings for apps that reuse context.

At a press conference, OpenAI's head of product, Olivier Godement, highlighted the company's progress, noting GPT-3 costs have been slashed by nearly 1000x in just two years. This reduction opens new opportunities for startups and businesses to pursue previously unaffordable applications.

#### Vision fine-tuning: Exploring a new horizon in visual AI technology

OpenAI also introduced vision fine-tuning for [GPT-4o](https://openai.com/index/hello-gpt-4o/), enabling developers to enhance the model's visual understanding with images and text. This update could significantly impact areas like autonomous vehicles, medical imaging, and visual search.

#### [Realtime API](https://platform.openai.com/docs/api-reference/introduction): Closing the divide in conversational AI

OpenAI launched its Realtime API in public beta, allowing developers to build low-latency, multimodal experiences like speech-to-speech applications. This enables ChatGPT’s voice controls in apps. To showcase its potential, OpenAI demoed an updated version of [Wanderlust](https://github.com/widgetti/wanderlust), where users can naturally converse with the app to plan trips, even allowing mid-sentence interruptions for a more human-like interaction.

#### Model distillation: Advancing toward more accessible AI solutions

The most transformative announcement was Model Distillation, a workflow allowing developers to use outputs from advanced models like [o1-preview](https://openai.com/index/introducing-openai-o1-preview) or GPT-4o to enhance the performance of smaller, efficient models like [GPT-4o mini](https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/). This method helps smaller companies access advanced capabilities without the high computational costs, bridging the gap between cutting-edge models and more accessible alternatives.

You can find more information about OpenAI DevDay 2024 [here](https://openai.com/devday/). 


Tech News March 2025 - Newsletter Highlights

Dominika

Beyond language borders with HeyGen: crafting videos that speak to the world!

Dominika

Harnessing AI for Love: a Valentine’s Day Guide

How AI can supercharge your business in 2024 on multiple levels of your cost and effort

Exploring AI Tools for Developers: Insights from PozTechTalks

Midjourney guide part 1: Simple thing - settings

Kaja

Midjourney guide part 2: Introduction to MJ prompts

Kaja

Midjourney guide part 3: Prompt structure

Kaja

Top 3 security vulnerabilities for Large Language Models

Exploring AI Tools for Developers: Insights from PozTechTalks part 2

Midjourney guide part 4: seed

Kaja

Midjourney guide part 5: Parameters

Kaja

Unveiling the Latest Tech Insights: May Newsletter Highlights

Midjourney guide part 6: The puppet method

Kaja

Unveiling the Latest Tech Insights: June Newsletter Highlights

Dominika

Exploring GPT-Extensions: Planty, Tattoo GPT, and Wine Sommelier

Unveiling the Latest Tech Insights: August Newsletter Highlights

Unveiling the Latest Tech Insights: October Newsletter Highlights

Everything you need to know about OpenAi DevDay 2024 -Newsletter Highlights

How did AI help me deliver the best client demo possible? - the Inklusion Digital Saga part 4

Unveiling the Latest Tech Insights: November Newsletter Highlights

Dominika

Unveiling the Latest Tech Insights: December Newsletter Highlights

Dominika

Tech News January 2025 - Newsletter Highlights

Dominika

Your New Work Buddy - Cursor Review

Maciej

Tech News February 2025 - Newsletter Highlights

Dominika

GET IN TOUCH