[Feature Request]: Add Kandinsky 2.0/2.1 support [Image Mixing]

[DenkingOfficial]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What would your feature do ?

Kandinsky 2.1 inherits best practicies from Dall-E 2 and Latent diffusion, while introducing some new ideas.

Repo: https://github.com/ai-forever/Kandinsky-2

• 2.0 model can do multilingual txt2img, inpainting and img2img
• 2.1 model can do multilingual txt2img and image mixing (like MJ)

Also, 2.1 can do image mixing with 2+ images and prompts

Example input:

images_texts = ['red cat', Image.open('img1.jpg'), Image.open('img2.jpg'), 'a man']
weights = [0.25, 0.25, 0.25, 0.25]

Web demo (txt2img and img2img): https://fusionbrain.ai/diffusion

Model architecture (2.1):

Image mixing example (2.1):

Inpainting (2.0):

There was a similar feature request https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/8939, but it was only about 2.0 model.

I guess image mixing is a wanted feature which is present in 2.1 model. It will be awesome if it is possible to make webui support this.

Proposed workflow

1. Set model to Kandinsky 2.1/2.0
2. Type a prompt / Set images (+prompts if needed)
3. Get txt2img/img2img/image mixing result

Additional information

No response

[0-NiK-0]

Kandinsky 2.1 https://github.com/ai-forever/Kandinsky-2 https://huggingface.co/ai-forever/Kandinsky_2.1 https://huggingface.co/spaces/ai-forever/Kandinsky2.1

Demo https://fusionbrain.ai/ https://t.me/kandinsky21_bot https://rudalle.ru/kandinsky2

Info https://habr.com/ru/companies/sberbank/articles/725282/ https://www.reddit.com/r/StableDiffusion/comments/12bf5k2/kandinsky_21_beats_stable_diffusion_and_allows/

[user425846]

Is there any update on this?

[catboxanon]

Duplicate of https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/8939