How to use Stable Cascade

In ComfyUI, using Stable Cascade is easy to set up. Here are the steps:

Update Comfy UI

Ensure ComfyUI is up to date. Open Comfy UI, click “Manager” then “Update ComfyUI”.

A photo-realistic digital render of an abstract, futuristic cityscape at sunset. Skyscrapers of various shapes and sizes, some with curved surfaces, others with angular edges, reach towards the sky, their glass facades reflecting the brilliant hues of the setting sun. Monorails weave through the city, their trails of light creating a dynamic rhythm amidst the architectural marvels

Download the Stable Cascade Models

You’ll need to download the models and place them in the correct ComfyUI folders.

Download the following models from this site https://huggingface.co/stabilityai/stable-cascade/tree/main

Stage_A.safetensors
Stage_B.safetensors
Stage_C.safetensors

Alternatively, you may wish to download the _bf16 versions of the Stage B and Stage C files, which will be easier to run on some machines, while providing slightly degraded results.

Download this model from here: https://huggingface.co/stabilityai/stable-cascade/blob/main/text_encoder/model.safetensors

model.safetensors

After the downloads are complete, we’ll move the files into the comfy UI folders as described below.

Place Stage A in the following folder:

ComfyUI/models/vae

Place Stage B and Stage C in this folder:

ComfyUI/models/unet

And then place the model.safetensors file here:

ComfyUI/models/clip/

Start Generating Stable Cascade Images

Watercolor painting of a snug log cabin, as seen from the inside. The inviting morning light pours through the window, casting elongated shadows and illuminating the quaint furniture. A placid lake, reflecting the soft pastel colors of the morning sky, is visible through the window. Tropical blue water and breathtaking sunrise clouds

You can download the image above and drag it into your ComfyUI workspace to setup all the required nodes and start generating Stable Cascade images right away.

Otherwise, here’s what you need to get started:

The general flow of Stable Cascade is Clip -> Stage C -> Stage B -> Stage A.

Screenshot of the following setup: Setup the first KSampler, connect a UNETLoader with the Stage C model loaded. Add two CLIP text encoders for the positive and negative prompt, with the Model.safetensors connected to them. Then add a StableCascade Empty Latent Image, with dimensions set to 1024x1024.

Setup the first KSampler, connect a UNETLoader with the Stage C model loaded. Add two CLIP text encoders for the positive and negative prompt, with the Model.safetensors connected to them. Then add a StableCascade Empty Latent Image, with dimensions set to 1024×1024.

Then add the second KSampler, connect another UNETLoader with the Stage B model loaded. Add a StableCascade Stage B conditioning node with a Conditioning Zero Out node connecting that, which is connected back to the positive Clip Text Encoder. Connect the Latent Image back to the Empty Latent image from before, on the Stage B node.