Hacker News Reader

Is there an interpolate mode for Stable Diffusion? (Interpolate between two images without using a text prompt.)

Here are DALL-E 2 interpolations: https://twitter.com/model_mechanic/status/151297688118364569...

One of the first things that came out when the model was released.

https://github.com/schmidtdominik/stablediffusion-interpolat...

Thank you, but this notebook appears to use images generated from text prompts. I was interested in interpolating between two given images, without generating from text.

You would need to run CLIP and generate CLIP embeddings from the images first, then feed them into the samplers. If you give it the starting images and caption them yourself, then it would also work.

Reply on news.ycombinator.com