The AI Canvas

Inspiration

We are inspired by the possibilities of AI for expressing visual ideas through language, and by the lack of good interfaces for using this technology effectively.

What it does

It consists of an infinite canvas that contains 20 different fine-tuned Stable Diffusion models in combination with image inpainting, outpainting, image variations, and even GPT-3.

How we built it

We built it using Svelte and fabric.js on the front-end, and pytorch/diffusers on the backend, all connected through a Redis queue.

Challenges we ran into

A TON. How do we enable users to see a timeline of their generations? How do we make that the AI server stops crashing? Should we make it collaborative or just run it in our computer? We had a fun night of hacking.

Accomplishments that we're proud of

It works, we have a timeline, the AI server is stable, and we were close to get a feature to paint directly in the canvas.

What we learned

We were WAY slower than expected. It was super hard to find some bugs, especially with storing the history of each generation, and with problems that were spontaneously happening in the AI server.

What's next for The AI Canvas

We will enable users to train their own models.

Built With

• cloudflare
• pytorch
• redis
• svelte