Stable Cascade is here! [Colab] txt2img/img2img

UPDATE: Now in ComfyUI - click here if you want the workflow

A100 Pro Gated Notebook: 2048x1152 native in 20 secs (keep scrolling for the T4)

https://colab.research.google.com/github/MushroomFleet/StableCascade-text2image/blob/main/240217_Pro_gated_colab_of_Stable_Cascade.ipynb

UPDATE: for anyone having trouble with the GRADIO T4 version, i have made this fork available that allows txt2img with T4 but without high ram, just stay under 1024x512
T4 Notebook: https://colab.research.google.com/github/MushroomFleet/StableCascade-text2image/blob/main/Johnsons_Fork_240216_T4_free_colab_of_Stable_Cascade.ipynb

Colab Free Tier (T4) Gradio Notebook by mkshing.
https://colab.research.google.com/github/mkshing/notebooks/blob/main/stable_cascade.ipynb

~

my day one notebook:
https://colab.research.google.com/github/MushroomFleet/StableCascade-text2image
which runs at full speed with all parameters exposed, this is great to learn how the new system is working as you can see all of the code needed to run inference, but requires A100 to run. This is probably more than most people will need, so while the GRADIO version at the top might be slower, it is running on T4.

Official info here - Stable Cascade has arrived. https://github.com/Stability-AI/StableCascade
This revolutionary 3 stage model generates 1024x1024 images faster than SDXL

There is a huggingface space for simple inference, however i put together a colab (requires A100) if people want to play with all the parameter in each stage. You can use a text prompt with txt2img, all four parameters are exposed for the B and C stages. there is an image variation mode, which only needs an image URL with no prompt. Finally there is an img2img mode, which can take an image URL and a prompt, with some parameters automatically scaling, which you can see in the code.



each video has a walkthrough if you need it :)

another step forward for Stability.