UNET Stable Cascade (Simplified)版本Workflow v1.0 + Files ✅ (ID: 422406)

Information:

This is a simplified easy to use Workflow for the UNET version of Stable Cascade. The workflow is annotated with informative notes that offer insight into settings and what is happening at different times during the workflow's process.

The downloadable *.zip file contains the workflow and the 4 files you'll need for the UNET version to function.

Why UNET and not the regular Checkpoint version you ask?
- Because I feel that pruning the the UNET files that are +22Gb in size into the 16-ish Gb regular Checkpoints costed the models a bit in style and quality. (If there are requests for a Simplified Standard Checkpoint Workflow I'll see what I can do.)

GPU Warning
- According to StabilityAI you need at least 8Gb of onboard RAM in your GPU in order to reliably use Stable Cascade.

• A reliable source let me know that an RTX3060 Laptop GPU with 6Gb RAM is also capable of running this model.

Installation:

• Install ComfyUI Manager by ltdrdata ( https://github.com/ltdrdata/ComfyUI-Manager )

  • This will make sure you can easily install the missing nodes you'll need for this Workflow to function.

• Download the *.zip file containing the Workflow and all the extra files you'll need.

• Launch ComfyUI and Load the Workflow

• Use Manager to install the Missing Nodes and then restart ComfyUI (there should be no Red Colored Nodes after this step)

• Put the Extra Files into the Correct Folders

  • model.safetensors --> Copy the file into ComfyUI\models\clip

  • stage_b.safetensors
    stage_c.safetensors --> These two files go into ComfyUI\models\unet

  • stange_a.safetensors --> Cope into ComfyUI\models\vae

These 4 files are originally created and trained by StabilityAI: ( https://huggingface.co/stabilityai )

How to Use:

• Fill in the Positive and Negative Prompts

• Adjust the settings in the Empty Latent Image

  • A list of ratios and sizes is provided inside the workflow as this model was trained with specific sizes in mind.

  • As for Compression:

    • It should be between 32 and 64 (the higher the compression the bigger your image size can be)

    • Personally, I use a RTX4080 and I like it around 42

• Adjust the Ksampler settings

  • Steps are all up to you I usually go with 50 & 20

  • CFG shouldn't be too high. 4 for Realism seems good enough for Stage_C and no higher than 1.5 for Stage_B.

  • I found the best results come from Euler (Realism) and Euler_Ancestral (Artistic) samplers.

  • Scheduler should be on Simple

描述:

训练词语: