UNI-1

Luma's UNI-1 is a text-to-image model with one feature nobody else in the pack has: a settings.search toggle that makes it search the web for visual references before it generates. Turn it on, prompt a specific landmark or product, and instead of hallucinating a plausible-looking wrong version, the model grounds itself in what's actually out there. For "generate a poster of the Sydney Opera House" work, that's a genuinely useful difference.

The catch is the same one that greets every Luma model in this community: it's closed. When a UNI-1 announcement hit the local-generation subreddit, the top reply was basically "not local, not open source, not relevant." Fair. This node doesn't change that - it's the hosted model rented by the generation through Runware's cloud, part of the ComfyUI-Runware pack. You get Luma's quality and its search grounding, and in exchange you take its alignment and its meter.

How it works

The node sends model AIR luma:uni@1 with an imageInference task over REST via the runware-sdk, and the image returns as a native IMAGE output. The inputs worth knowing:

• positivePrompt - required.
• seedImage - an IMAGE socket for image-guided generation; referenceImages for additional conditioning.
• settings.search - the headline toggle: web search for visual references before generating. Good for real-world subjects, landmarks, products.
• settings.style - a style preset, default auto.
• width / height - 1024 default; numberResults 1–4.
• watermark socket for watermarked output.

Output is one image socket, type IMAGE, into Preview Image or Save Image. Default format JPG at quality 95; PNG for lossless.

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Restart, set your key via ComfyUI Settings → Runware API key, RUNWARE_API_KEY, or runware auth login. No model files - only the SDK, pillow, and soundfile.

Where people get burned

Expecting local behavior from a hosted model. settings.search is a network call inside a generation - it makes the queue slower and possibly pricier, so keep it off for abstract or fictional subjects where there's nothing to search. And don't assume UNI-1's look matches Luma's public demo reels exactly; hosted image tiers drift from marketing. If a prompt comes back softer than the hype, that's the alignment/compression reality of renting a closed model, and the Max tier is the pack's upgrade path.