Stability Image Ultra

Image Ultra is the flagship generator of the official Stability pack - the node you pick when you want Stability's best-looking text-to-image result and you don't have the hardware (or the patience) to run a comparable model locally. It calls stable-image/generate/ultra, and it's the one node in the pack that doubles as both a text-to-image and an image-to-image generator depending on whether you feed it a starting image.

The same framing applies here as everywhere in this pack: it's a paid cloud route, needs a Stability API key, burns credits per call, and uses zero VRAM. And the same community sentiment applies: "not local, not interested" has been the pack's reception since launch, and if you've got a tuned local Flux or SDXL setup, you already have something comparable for free. Where Ultra earns its keep is the machines that can't - and honestly, for "I need a genuinely good image in one call with no setup," it's the strongest pure-generation node in the pack.

How it works

prompt is the only required input. From there:

• image (optional) - here's the clever bit. Leave it empty and this is a straight text-to-image generator with an aspect_ratio picker (nine presets, 1:1 through 21:9). Feed an image in and it silently switches to image-to-image mode: aspect_ratio is dropped and strength (0.0–1.0, default 0.7) controls how much of the source survives. Same node, two jobs, one optional socket.
• strength - only relevant when an image is provided. Low keeps the source nearly intact; high lets the prompt take over.
• negative_prompt, seed - the usual pair.
• style / style_preset - the pack's standard list (anime, cinematic, photographic, and friends). Remember the pack quirk: the preset does nothing unless style is flipped on.
• output_format - png, webp, or jpeg.

Output is a single IMAGE tensor.

Installing it

From the "Stability API nodes for ComfyUI" pack - 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 model downloads; requests is the only dependency beyond ComfyUI's own stack. Key via SAI_API_KEY, a sai_platform_key.txt file beside the node, or the per-node api_key_override (which rides in the workflow JSON, so don't share those files).

Common issues

• "payment_required" - out of credits; the trial bucket evaporates fast here.
• Aspect ratio seems ignored - you've got an image connected. With an input image the node is in image-to-image mode and the ratio is forced by the source; disconnect the image if you want to pick a shape.
• Style preset does nothing - style is false. Pack-wide behavior, not a bug.

It's the closest thing in the pack to "just make me a good image." If you've got a GPU and strong opinions about local control, you won't use it. If you don't, this is the one generation node worth knowing by name.