What are the stages in Stable Cascade?

Stable Cascade has three stages. Stage C is the text-conditioned prior that does most of the creative work in a highly compressed latent space. Stage B decodes that into a larger latent, and Stage A is a lightweight VAE that produces the final pixels. You mainly tune Stage C and Stage B.

How many steps should each stage use?

Stage C typically needs 20 steps for balanced quality (10 for drafts, 30+ for maximum), while Stage B needs far fewer — around 10 steps is plenty because it is only decoding, not generating. More Stage B steps give diminishing returns.

What guidance scale works best?

Stage C responds well to a CFG of around 4, lower than typical SD models because its compressed latent space is very prompt-sensitive. Stage B usually runs at CFG 1.1 or even 0, since it is decoding rather than interpreting the prompt.

Why is Stable Cascade so memory efficient?

It works in a 42x compressed latent (Würstchen architecture), so the expensive Stage C operates on tiny tensors. That lets it generate high-resolution images with less VRAM and fewer steps than pixel- or 8x-latent diffusion models.

What is the Stable Cascade Configuration Guide?

Guide to Stable Cascade (Würstchen v3) settings. Tune steps and guidance for the Stage C prior and Stage B decoder, pick a latent resolution and balance quality versus speed for your available VRAM. It runs free in your browser on Gera Tools, with nothing uploaded.

Stable Cascade Configuration Guide

Name: Stable Cascade Configuration Guide
Creator: Gera Tools
License: https://creativecommons.org/licenses/by/4.0/

Get one useful tool a week

Like this tool? Enter your email and we'll send you one genuinely useful Gera tool a week — plus a link to come back to this one. No spam, one-click unsubscribe any time.

Stable Cascade configuration

Stable Cascade (the Würstchen v3 architecture) splits generation across three stages working in a heavily compressed latent space. Stage C is the text-conditioned prior doing the creative work, Stage B decodes it into a larger latent, and Stage A is a tiny VAE producing pixels. Most of your tuning happens on Stage C and Stage B, and this guide picks sensible values for your quality target and VRAM.

Why Stable Cascade is different

Most diffusion models (SD 1.5, SDXL, Flux) operate in a latent space compressed 8x relative to pixel resolution. Stable Cascade uses a 42x-compressed latent for Stage C. This is the core architectural difference: the prior operates on far smaller tensors, which means:

Fewer steps needed. Stage C produces strong results in 20 steps rather than the 30–50 typical of SDXL.
Lower CFG needed. The compressed latent is very prompt-sensitive; CFG of 4 is approximately equivalent to CFG 7 in a standard 8x model.
Lower VRAM at high resolution. Because Stage C tensors are tiny, you can generate at 1024×1024 with less VRAM than an equivalent SDXL run.

Recommended settings by quality target

Target	Stage C steps	Stage C CFG	Stage B steps	Stage B CFG
Draft (fast)	10	4.0	4	1.1
Balanced	20	4.0	10	1.1
Maximum quality	30–40	4.0–4.5	10	1.1

Stage B step count changes almost nothing beyond 10. All additional quality investment should go into Stage C step count.

VRAM requirements (approximate, bf16)

Configuration	VRAM
Stage C only (lite variant)	~4 GB
Full pipeline 1024×1024	~8 GB
Full pipeline 1536×1536	~12–14 GB

The lite Stage C variant trades some quality for a dramatically smaller footprint. If you are on 6–8 GB, use the lite prior and full Stage B.

Applying the config in ComfyUI

In a ComfyUI Stable Cascade workflow:

StableCascade_StageC_VAEDecode node controls Stage C. Set steps and cfg here.
StableCascade_StageB_VAEDecode node controls Stage B. Use 10 steps, CFG 1.1.
KSampler for Stage C: sampler euler_ancestral, scheduler simple or karras.
Load the Stage C prior and Stage B decoder as separate model checkpoints.

Tips for quality improvement

Add detail to the positive prompt. Unlike SDXL, Stable Cascade responds well to explicit lighting and material descriptors in the positive prompt rather than relying on a style embedding.
Avoid high CFG. Above CFG 5 in Stage C, colour saturation spikes and fine details start to burn out. Stay at 4–4.5.
Use varied seeds. Stage C generation has notable seed variance; run 4–6 seeds and pick the best composition before spending steps on Stage B.
bf16 throughout. Running any stage in fp32 doubles VRAM and provides no visible quality benefit for this architecture.