Table of contents
English
English
Stable Cascade is a text-generated image (txt2img) model released by Stability AI in February 2024. The model adopts an innovative three-stage approach, setting new standards in quality, flexibility, fine-tuning and efficiency, while reducing hardware requirements. Stability AI also released training and inference code on its official GitHub to facilitate users to customize models and their output. Stable Cascade is compatible with the diffusers library and can be used for inference tasks.
This article will introduce how to use Stable Cascade in ComfyUI, covering text-generating images (txt2img), image-generating images (img2img), image variants (image variants), control networks (ControlNet), and internal drawing (Inpainting) workflows.
Using Stable Cascade in ComfyUI requires downloading the following model files:
Stable Cascade checkpoint (stored in ComfyUI/models/checkpoints)
ControlNet and Inpainting models (stored in ComfyUI/models/controlnet)
stable_cascade_canny.safetensors (ControlNet line draft model)
stable_cascade_inpainting.safetensors (Inpainting internal drawing model)
Stable Cascade adopts a three-stage process :
Stage C : The diffusion model generates low-resolution latent images.
Stage B : The diffusion model enlarges the latent image.
Phase A : VAE converts the enlarged latent image into visible pixel space.
This process can be used to generate completely new images from text prompt words.
Workflow
Download and load the HY_stable_cascade__text_to_image.json workflow.
Example Prompt
Evening sunset scenery blue sky nature, glass bottle with a fizzy ice cold freezing rainbow liquid in it
Generate effect:
This process generates new variants based on the input image while referring to text prompt words.
Workflow
Download HY_stable_cascade__image_to_image.json and load it into ComfyUI.
Example Prompt
Evening sunset scenery blue sky nature, glass bottle with a fizzy ice cold freezing rainbow liquid in it
Generate effect:
Stable Cascade supports the use of CLIP visual models to generate variants of input images.
Workflow
Download HY_stable_cascade__image_remixing.json and load it into ComfyUI.
Example Prompt
Evening sunset scene blue sky nature, glass bottle with a fizzy ice cold freezing liquid in it
Generate effect:
Multiple images can be entered simultaneously and fused to generate new images.
Workflow
Download HY_stable_cascade__image_remixing_multiple.json .
Example Prompt
Evening sunset scene blue sky nature, glass bottle with a fizzy ice cold freezing liquid in it
Generate effect:
Use ControlNet to control the composition generated by the image, such as line drawing coloring, depth map control, etc.
Workflow
Download HY_stable_cascade__canny_controlnet.json and make sure the stable_cascade_canny.safetensors file is stored in the ComfyUI/models/controlnet folder.
Example Prompt
Evening sunset scenery blue sky nature, glass bottle with a fizzy ice cold freezing rainbow liquid in it
Generate effect:
Used to repair or modify specific areas in the image.
Workflow
Download HY_stable_cascade__inpaint_controlnet.json and make sure that stable_cascade_inpainting.safetensors are stored in ComfyUI/models/controlnet .
The entered sample image can be downloaded from here: yosemite_inpaint_example.png .
Right-click the image of the LoadImage node
Use the mask tool to select the area to modify
Enter a new Prompt to fill
Example Prompt
Evening sunset scenery blue sky nature, glass bottle with a fizzy ice cold freezing rainbow liquid in it
Generate effect:
Stable Cascade is a new generation of text-generated image model launched by Stability AI, with the characteristics of being efficient, flexible and fine-tuning . Through ComfyUI, users can realize literary and artistic images, image variations, ControlNet constraint generation and Inpainting details repair , greatly improving the creative freedom of AI painting.
If you want to experience more powerful image generation, it is recommended to download ComfyUI and try the Stable Cascade workflow!
