Luma UNI-1 Image

Luma UNI-1 Image is the current text-to-image node from Luma, built around the UNI-1 model - and its most notable feature is how little interface it has. Prompt, model, seed. That's it. No aspect-ratio dropdown, no style sliders, no reference inputs cluttering the node. It's a deliberate design choice: UNI-1 is a model-family where the generation parameters (aspect ratio, style, references) live in a dynamic sub-panel that appears after you pick the model, and where the model is expected to decide things like aspect ratio for you.

Partner node, ships with ComfyUI core, no install, runs on Luma's servers through Comfy's proxy and bills per call. The "1–6000 characters" limit on the prompt is a small window into how this thing is meant to be used: it's a natural-language model, so write sentences, not comma-separated tags.

How it works

Your prompt goes to Luma's generation endpoint with type="image"; the node polls until the job completes and downloads the result as a normal IMAGE tensor. The model combo is dynamic: pick uni-1 or uni-1-max and extra inputs materialize under the hood - like an aspect ratio list that defaults to auto (the model picks the canvas from your prompt) and style and reference options. So the node isn't as bare as it looks; the options are just tucked behind the model choice.

The inputs that matter

• prompt - the text description, 1–6000 characters. Full sentences beat tags, and it's fine to describe the layout you want.
• model - uni-1 or uni-1-max. Max is the pricier, heavier variant; plain UNI-1 is the default for a reason. The price badge updates as you switch.
• seed - the standard partner-node caveat: it only controls whether the node re-runs. Results are nondeterministic regardless of the value.

Output is IMAGE, ready for Save Image or anything downstream.

Where people get burned

• Looking for controls that aren't on the node. Aspect ratio is a sub-option that appears when you select the model, and auto is the default - the model picks it. If you need a specific ratio, set it in the model's panel, not by hunting for a node-level dropdown.
• Manga style + landscape ratio. The execute logic enforces portrait ratios (like 2:3, 9:16, 1:2, 1:3) when you pick the manga style. A mismatched combo errors - by design.
• Tag-style prompts. This is a description model; "1girl, white hair, city, night" style tags will underperform compared to a real sentence. Give it prose.
• Cost. uni-1-max roughly doubles the per-image price. Don't leave it on max just because it sounds better.