Stability Style Control

Where the other two control nodes pin down where things go, StabilityControlStyle pins down how it looks. You feed it a reference image that defines the aesthetic - an oil painting, a product photo's lighting, a game's art direction - and a prompt that defines the subject, and it generates an image that matches your prompt in the style of your reference. It's hosted style transfer: the local equivalent is IP-Adapter or Reference-style control, minus the model files and the VRAM.

The input that makes this node tick is fidelity (0–1, default 0.5), which is roughly "how much of the reference's look must the output keep." At 0.5 the style comes through strongly while the prompt runs the show. Crank fidelity up and the output leans hard into the reference - high enough and it starts to resemble the reference's actual content, not just its look. Drop it low and you're back to "I showed you a painting, I guess you absorbed its mood."

The style_preset trap

This node also has a style_preset dropdown (anime, cinematic, photographic, and fifteen more), and it's worth understanding the difference: style_preset is Stability's built-in named look, while your reference image is your style. They're two ways to do the same job, so feeding both is usually redundant - pick your own reference image and leave the preset at none, or skip the reference and use a preset. Also note the extra aspect_ratio input here (nine ratios, default 1:1), which the sketch and structure nodes don't have - you can control output dimensions directly, useful for restyling into a specific canvas.

Inputs and outputs

• image - required. Your style reference.
• prompt - required. What the output depicts.
• negative_prompt - optional.
• fidelity - 0–1, default 0.5. The main dial.
• aspect_ratio - 1:1, 16:9, 9:16, etc.
• seed - set for reproducibility.
• style_preset - 18 named looks; none unless you're skipping the reference.
• output_format - png / jpeg / webp.
• api_key - per-node override.

Output: a single IMAGE.

Installing it

Same pack - aicu-comfyui-stability-ai-api:

cd ComfyUI/custom_nodes
git clone https://github.com/aicuai/aicu-comfyui-stability-ai-api.git
cd aicu-comfyui-stability-ai-api
pip install -r requirements.txt

Or ComfyUI Manager → search "aicu-comfyui-stability-ai-api" → install → restart. You need a paid Stability AI key: node api_key field, STABILITY_API_KEY env var, or the auto-generated config.ini.

Gotchas

Style transfer is the most subjective of the three control nodes, so treat fidelity as something to sweep: run a couple of seeds at 0.3, 0.5, 0.7 and pick. The usual input constraints apply (64px minimum sides, 4,096–~9.4MP pixels, aspect 1:2.5–2.5:1), and everything runs through Stability's credit meter and moderation. If your style reference is something the platform won't reproduce, you'll get a 403 instead of art. For "make my subject look like this aesthetic," though, it's a genuinely low-effort way to get there.