Stability Fast Upscaler

This is the simplest node in the whole pack, and the name is the whole feature list: an image goes in, a bigger image comes out, and your GPU never gets involved. StabilityUpscaleFast calls Stability AI's hosted upscale endpoint (/v2beta/stable-image/upscale/fast), waits a few seconds, and hands you back a 4x upscale as a normal ComfyUI IMAGE tensor.

The catch, because there always is one: it's a paid API. Every call burns Stability credits, so before you reach for it, ask whether the local alternative actually fails you. If your only goal is more pixels from an already-sharp source, a free ESRGAN model (4x-UltraSharp, Remacri) is instant, runs offline, and cannot invent anything - that's the correct default for most people. Where this node earns its keep is when you're working on a machine with no room for an upscaler model at all, or you want the dead-simple "feed image, get 2048px" path with zero setup.

How it works

The node validates your image against Stability's constraints (32–1536px per side, total pixels between 1,024 and 1,048,576 - an out-of-range source raises an error before anything is spent), uploads it as multipart form data, and requests the fast upscale. The API replies with the finished image directly - no polling, no waiting on a generation ID. That's the whole mechanism, and it's why this node feels snappier than its siblings in the pack.

Inputs and outputs

There's just one required input:

• image - the IMAGE tensor you want upscaled. Wire it straight from a Load Image node (or anything that outputs IMAGE, like a prior generation).

Everything else is optional and, honestly, barely matters:

• api_key - override the pack-level key if you want to charge this call to a different account.
• output_format - png (default), jpeg, or webp. Pick jpeg if you're piping into something size-sensitive.

Output is a single IMAGE, which you route to a Save Image / Preview Image node like any other tensor.

Installing it

It ships in the aicu-comfyui-stability-ai-api pack, which you install once and it unlocks all the Stability nodes:

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 just search "aicu-comfyui-stability-ai-api" in ComfyUI Manager and hit install. Then restart ComfyUI.

You'll also need a Stability AI API key from platform.stability.ai - the account is credit-based and paid. The pack reads it from the node's api_key field, the STABILITY_API_KEY environment variable, or a config.ini that gets auto-generated in the pack folder on first run (drop your key under [stability]).

Where people get burned

The input-size limit is the one that bites: feed it a 2048px-wide render and it errors out before you've spent anything - downscale first. And remember this upscaler only enlarges. It doesn't restore a soft or compressed image - if the source is mushy, the output is a bigger mushy image, and you want the pack's conservative or creative upscaler instead. For a thin utility node that's the whole story: fast, dumb, hosted, and genuinely handy when your VRAM is the bottleneck.