Stability Ultra Image Generation

Ultra is the expensive one, and in this pack that's also the point. StabilityImageUltra is Stability's flagship hosted generator - /v2beta/stable-image/generate/ultra - the model that gets the company's best detail and photorealism. If you're paying per image anyway, this is where the money is worth spending; Core is the value pick, Ultra is the quality pick. Think of it as the hosted stand-in for a serious local photorealism stack: Flux-class quality without the 24GB of model files and the refinement loop.

The neat part is that it runs in two modes with no mode switch at all. Wire an image in and it silently becomes image-to-image; leave the image input empty and it's plain text-to-image. The node figures out which one you meant from whether the image socket is connected, and swaps which parameters apply - strength (0–1, default 0.7) governs the i2i mode, aspect_ratio (nine ratios) governs the t2i mode.

Inputs and outputs

• prompt - required, always. Ultra rewards good prompting; it's a strong model, so composition and lighting words carry.
• aspect_ratio - required (used in text-to-image mode).
• negative_prompt - optional.
• seed - set for reproducibility.
• style_preset - 18 named looks; none by default.
• output_format - png / jpeg / webp.
• image - optional. Connect it to switch into image-to-image.
• strength - optional (image-to-image only). How much of the source survives; 0.7 is a sensible "refine it" starting point, lower keeps more of the original.
• api_key - per-node override.

Output: a single IMAGE.

How to use it in practice

Two workflows actually shine here. First, text-to-image at 16:9 or 21:9 - Ultra holds composition better than most hosted models at wide ratios, which makes it good for banner/keyart ideation. Second, the i2i refine pass: generate at Core (cheap), then feed that result into Ultra with strength ~0.5–0.7 to get the flagship's detail on top. That two-node pipeline is the budget-conscious way to use a premium endpoint - one cheap pass for ideas, one expensive pass for the final.

Installing it

Part of the aicu-comfyui-stability-ai-api pack:

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. Paid Stability AI key required: the node's api_key field, the STABILITY_API_KEY env var, or the auto-generated config.ini in the pack folder.

Gotchas

Two validation rules to know before you spend. In image-to-image mode the input must be between 64 and 16,384px per side and at least 4,096 total pixels, and strength is required when an image is present - the node demands it rather than silently defaulting. And the platform moderation still applies at the top tier: a 403 means Stability refused the content, and no prompt engineering around it. Beyond that, the main risk is your wallet - Ultra is the priciest call in the pack, so use the Core-then-Ultra pipeline and don't batch fifty of them by accident.