Hi folks! 👋🏻 It's been a good week for AI makers so far - we got new GPT3 model that is turning out to be much more creative and better at taking instructions and Emad from Stability.ai is teasing big news for Stable Diffusion today.

Let's see what else is going on 👇🏻

Don't confuse the AI 😵‍💫

This new model "Conffusion" can take a damaged image and make an educated guess about what the true pixel values should be. It does this by creating a range around each pixel, with a certain probability that the true value is inside the range.

This can be extremely valuable for image-to-image AI models such as inpainting, where the model has to be certain of what it needs to "fill" in the image.

Free-form prompt scene control 🏞

Current AI models are very good at generating images from prompts. But, it's hard to control the shapes of different regions/objects or their layout.

SpaText is a new AI model that in addition to a text prompt that describes the entire scene, the user can provide a segmentation map and annotate each region with a prompt.

📚 Educational corner

• Best practices for prompt engineering with OpenAI's GPT3 models

• Learn what prompts work with Stable Diffusion 2.0 with this demo (Prompts from SD 1.5 don't really work with 2.0, so this might be a good exercise to find out what works)

• A Twitter thread that compares OpenAI's newest model (003) to the older one (002) on different tasks

🛼 Makers corner

• AI freestyle rapping assistant dropping some heat

• Woolitize an image with this DreamBooth model

• GPT3-based browser automation in PHP - you can search something like "Buy Harry Styles ticket in Amsterdam" and this will run on autopilot to complete the task

• Storytimes.ai - create the perfect children's book with AI

🤓 Tech corner

• NeuralLift-360 - transform a 2D image into 3D with 360° views (a new framework that outperforms all previous models)

• ReCo - a novel framework that is more precise about what the AI draws on a specific location of the photo, something that LLMs had issues with in the past.

• DATID-3D - a novel model that can create 3D models in various domains (such as Pixar, Sci-fi, Robots) without using additional training images & data

🎨 Creativity corner

🧱 What are you building?

If you're using AI in your work or projects, I'd love to hear about it! Please reply to this email and let us know what you're up to. We may feature your work in our newsletter.

❤️ If you like The Prompt, and want to support my work:

• Share The Prompt with a friend, and invite them to subscribe here.

• Book an ad in The Prompt ( reply to this email if you’re interested)

Thank you for reading! ✌🏼