Stability Control Sketch
Turn a scribble into a scene, cloud-style
- image
- IMAGE
Control Sketch is the "rough lines in, finished image out" node of the pack. Draw a sloppy sketch - or feed it line art produced by a local scribble/lineart preprocessor - add a prompt, and Stability's servers render a full scene that respects your lines. It's the hosted equivalent of a ControlNet scribble model: same idea, none of the setup, and none of the VRAM.
Locally, scribble control is genuinely fun and genuinely fiddly: you need a ControlNet model that matches your checkpoint's architecture, a preprocessor, a weight, and start/end step tuning. This node collapses all of that into image + prompt + one slider. It's part of the official "Stability API nodes for ComfyUI" pack, it calls stable-image/control/sketch on Stability's cloud, and it costs credits per call with a Stability API key. The community has been unimpressed with the whole pack - "not local, not interested" has been the standing reply - and if you already run a local ControlNet stack you have more control than this gives you. This is the low-effort path, or the path for machines that can't run the local one.
How it works
The server takes your image, reads it as a sketch/edge condition, and regenerates content that follows it, steered by prompt. Your sketch doesn't need to be pretty - scribbles are the point, not polished line art. The inputs that matter:
- image - the sketch. Hand-drawn in an image editor, or generated locally with a lineart/scribble preprocessor and fed in.
- prompt - what the sketch becomes. This is where subject, style, and lighting go.
- control_strength (0.01–1.0, default 0.7) - how tightly the output follows your lines. High = the sketch rules the composition; low = the model interprets and improvises.
- negative_prompt and seed - steering and reproducibility.
Then the pack's shared options: style + style_preset (the preset is ignored unless style is on - the recurring quirk), output_format for png/webp/jpeg. One IMAGE out, into Save Image or downstream.
Installing it
Same pack as every node here - ComfyUI Manager (search "Stability API nodes for ComfyUI"), or:
cd ComfyUI/custom_nodes
git clone https://github.com/Stability-AI/ComfyUI-SAI_API
# restart ComfyUI
No models download; requests is the only dependency beyond ComfyUI's own stack. Key via SAI_API_KEY, a sai_platform_key.txt file, or the per-node api_key_override (which travels in the workflow, so don't share those).
One thing worth knowing before you blame the node: misspelled workflows sometimes reference this as "Stability Control Skech." Same node, typo'd name. Fix the class name and move on.
Common issues
- Output ignores your sketch -
control_strengthtoo low, or the sketch is so faint it reads as noise. Crank the strength and make sure your lines are clearly visible against a white background. - Everything looks over-detailed or muddy - feed a cleaner sketch. The server doesn't clean up your scribbles the way a local preprocessor would.
- Wrong node - if you want the layout/pose of a photo preserved rather than a hand-drawn sketch followed, that's Control Structure.
For "I doodled a composition and I want to see it rendered," this is the fastest route in the pack. Just keep the credits topped up.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| prompt | STRING | — | |
| control_strengthopt | FLOAT | 0.700.01–1 | — |
| negative_promptopt | STRING | — | |
| seedopt | INT | 00–4294967294 | — |
| styleopt | BOOLEAN | false | — |
| style_presetopt | COMBO | 17 options: 3d-model, analog-film, anime, cinematic, comic-book, digital-art, +11 | |
| output_formatopt | COMBO | 3 options: png, webp, jpeg | |
| api_key_overrideopt | STRING | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |