Flux.1 - A Midjourney V6.1 competitor from Europe
More topics: Collapses AI on AI-generated training data?, and Tesla filed a patent for a new AI vision system
Hi AI Enthusiasts,
Welcome to this week’s Magic AI News, where we present you the most exciting AI news of the week. This week, we show you an image generator from Europe that can compete with Midjourney V6.1. You can try the image generator for free!
This week’s Magic AI tool is a GPT-4 competitor from a European startup. It’s free and for unlimited use. Stay curious! 😎
Let’s explore this week’s AI news together. 👇🏽
Top AI news of the week
🌅 Flux.1: A Midjourney V6.1 competitor from Europe
The AI startup Black Forest Labs, founded by former Stability.ai developers, has released the image generator Flux.1. The mission of Black Forest Labs is to develop state-of-the-art generative AI models for media such as images and videos.
The details:
- Three variants of Flux.1: pro, dev and schnell
- FLUX.1 [pro]: Best model, state-of-the-art performance image generation with top visual quality, image detail, and output diversity
- FLUX.1 [dev]: Open-weight, guidance-distilled model for non-commercial applications, distilled from FLUX.1 [pro], more efficient model than FLUX.1 [pro] with similar quality, weights are available on HuggingFace
- FLUX.1 [schnell]: Open and fast model for local development, weights are available on HuggingFace
Take a look at the following example:
This image was generated with Flux.1[schnell] on replicate.com. You can try it for free!
Prompt: “filmic photo of a group of three women on a street downtown, they are holding their hands up the camera”
Our thoughts
Wow, this image looks realistic. You can only see small flaws on the fingers.
In our opinion, the image generator Flux.1 is impressive, and the quality is really good. AI-generated images are almost indistinguishable from real photos today! That’s problematic when you think of deep fakes! In our opinion, AI-generated images should always be labeled!
More information
- Announcing Black Forest Labs - Black Forest Labs Blog
🤖 Collapses AI on AI-generated training data?
In a new study, researchers demonstrate that AI-generated images deteriorate after a few model iterations when they use AI-generated images for training. The problem occurs when the researchers repeat the training with synthetic data. This creates a feedback loop. According to the researchers, even a few iterations damage the model permanently.
Do you want to dive deeper into the topic? If so, we recommend you take a look at the paper.
Our thoughts
The study shows that AI models need real-world data to improve! So, it’s time to label AI images reliably. That is a benefit for users and AI developers.
AI developers need to distinguish between generated and real-world data, keeping future models more powerful. In addition, users have more transparency. That is very important in a world where the virtual world is merging with the real world.
More information
🦾 Tesla filed a patent for a new AI vision system
Tesla has filed a patent for an AI-powered image-processing system for robots. The patent focuses on how autonomous robots can perceive and navigate their environment using only camera inputs.
The details:
- The AI vision system uses a single neural network (end-to-end model) to process camera data and output 3D environmental information. No LiDAR or radar is needed!
- It divides the surrounding space into 3D voxels and makes predictions for each voxel in real-time.
- It is designed to run on the robot’s onboard computer.
Our thoughts
This patent could represent a significant advance in autonomous navigation and environment perception. The technology is perfectly suited for autonomous Tesla vehicles and the Tesla Optimus robot, as it uses only camera data and runs entirely on the onboard hardware.
This technology could be crucial in realizing Tesla’s vision that must be able to operate safely in human environments. We are excited to see what Tesla will present at its Robotaxi event.
More information
- Artificial intelligence modeling techniques for vision-based occupancy determination - Google Patents
- Tesla’s Voxel-Based Vision System: Revolutionizing Autonomous Robots - X Post
Magic AI tool of the week
This week’s Magic AI tool of the week is Mistral Le chat - a free ChatGPT competitor.
Mistral AI recently published their new flagship model Mistral Large 2. It’s an open-source LLM with a similar performance to GPT-4. And best of all, it is available for free and unlimited use.
Step-by-Step Guide:
- Sign up for free at mistral.ai
- Choose an AI model of your choice (Mistral Large 2, Codestral, Mistral Nemo)
- Enter a question. The model returns an answer! That’s it! 🎉🎉
👉🏽 Try Mistral Le chat for FREE
Articles of the week
- Our Journey to Become a Data Scientist
- Mistral’s Codestral: Create a local AI Coding Assistant for VSCode
- Build a Local Chatbot in Minutes with Chainlit
- Struggling with Linear Regression? - Theory and Practice Clearly explained
💡 Do you enjoy our content and want to read super-detailed articles about AI? If so, subscribe to our blog and get our popular data science cheat sheets for FREE.
Thanks for reading, and see you next time.
- Tinz Twins
Leave a comment