โ๐จ API Image Generator (Stability - Stable Diffusion) ๐จโ
Call Stable Diffusion 3 from inside your graph
- IMAGE
This one skips your GPU entirely. Paste in a Stability AI API key, write a prompt, and it hits Stability's hosted Stable Diffusion 3 endpoint and hands you back an IMAGE - no checkpoint loaded locally, no VRAM spent. It's one of five "bring your own key" cloud generator nodes in the Bjornulf pack (the others cover FAL, CivitAI, Black Forest Labs/Flux, and OpenAI's gpt-image-1), and this is the Stability flavor.
Why you'd bother, given ComfyUI is built to run models locally: it's genuinely handy for comparing a cloud model against your local setup without leaving the canvas, or for running on a machine with no GPU at all - a laptop, a cheap VPS - where the actual pixel-pushing happens on Stability's servers. It's a convenience node, not a workflow centerpiece.
Worth knowing before you reach for it: Stability's own community standing took a real hit with the SD3 launch - a broken 2B model, a restrictive license, a CivitAI ban within days. Flux (built by ex-Stability researchers) absorbed most of the goodwill the community had left for Stability afterward. None of that makes the API itself unusable - it's a different product from the local weights - but don't assume "Stability" carries the same reputation here that SD 1.5 built.
Inputs and outputs
Two required fields: api_key (your Stability account key, pasted in as plain text) and prompt. Everything else is optional: negative_prompt, aspect_ratio (a dropdown of nine ratios from 21:9 to 9:21), seed, output_format (jpeg/png/webp), and strength (0โ1, default 1). One thing worth flagging honestly: there's no image input on this node at all - it's pure text-to-image - so strength doesn't have anything to apply itself against here. It's likely a passthrough field for the underlying Stability API call rather than something this particular node puts to use; don't spend time tuning it expecting a visible effect.
The single output is IMAGE, and it wires into anything downstream that expects one - a Save Image node, a Preview, or one of the pack's own comparison nodes.
Installing it
Through ComfyUI Manager: search Bjornulf_custom_nodes, install, restart. By hand:
cd ComfyUI/custom_nodes
git clone https://github.com/justUmen/Bjornulf_custom_nodes
Then, inside that folder, install the pack's Python dependencies - pip install -r requirements.txt (or the embedded-python equivalent on Windows: python.exe -m pip install ollama pydub opencv-python ffmpeg-python civitai-py fal_client sounddevice langdetect spacy textblob dill). None of the video/audio/TTS-heavy dependencies matter for this one node specifically, but they install together as one pack - that's the price of admission for 170 nodes sharing one requirements.txt.
Where people get burned
The obvious one: your API key sits in the node's widget as plain text and gets saved into the workflow JSON. Share that workflow file - post it, upload it, drop it in a Discord - and you've shared your key too. Clear it before you export anything, or keep API-key nodes on a workflow you never distribute.
The second: this is the author's own warning about the sibling CivitAI generator node, but it applies to any hosted API in this family - cloud generation costs real money per call, and hosted endpoints can be slow or flaky compared to a local run. Treat these as a convenience layer for quick comparisons, not your primary render path if reliability matters. And because it's a genuinely active pack (justUmen ships breaking changes regularly across versions), it's worth re-checking this node after a pack update if Stability changes their API on their end - third-party API wrappers are the first thing to break when a vendor tweaks a parameter name.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| api_key | STRING | โ | |
| prompt | STRING | A beautiful landscape | โ |
| negative_promptopt | STRING | โ | |
| aspect_ratioopt | COMBO | 1:1 | 9 options: 16:9, 1:1, 21:9, 2:3, 3:2, 4:5, +3 |
| seedopt | INT | 00โ4294967294 | โ |
| output_formatopt | COMBO | png | 3 options: jpeg, png, webp |
| strengthopt | FLOAT | 1.000โ1 | โ |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | โ |