Luma UNI-1 Image Edit

Luma UNI-1 Image Edit is the edit sibling in Luma's newer pair of nodes: hand it an image, tell it in plain English what to change, and it edits the photo in place. No mask, no ControlNet, no inpainting dance - it's the modern "describe the edit" workflow that the model-first labs all converged on. It's a ComfyUI core partner node, so it runs entirely on Luma's servers through Comfy's proxy and bills against your Comfy account credits.

The interesting bit for Comfy people is what this does to a graph. Because the input is a plain IMAGE and the output is a plain IMAGE, you can wire this mid-pipeline - generate something locally, hand it to UNI-1 for a fix-up, take the result back for local post-processing. Cloud edit nodes are usually a dead end; this one slots in like any other node.

How it works

The source image gets uploaded, your prompt goes to Luma's generation endpoint with type="image_edit", the node polls until the job lands, then downloads the result as an IMAGE tensor. The model dropdown is a dynamic combo that carries extra settings depending on which model you pick - so you might see more knobs appear under the hood (like aspect ratio or style) once you select uni-1 or uni-1-max. uni-1-max is the bigger, pricier variant.

The inputs that matter

• source - the image to edit. This is required, unlike some sibling nodes.
• prompt - the edit instruction, 1–6000 characters. Be descriptive: "replace the car with a red bicycle, keep the lighting" beats "change car." This is where all the quality lives.
• model - uni-1 or uni-1-max, with the dynamic sub-options.
• seed - the family-standard disclaimer: it only controls whether the node re-runs; the actual result is nondeterministic regardless.

Output is IMAGE, ready for Save Image or further nodes.

Where people get burned

• Vague prompts. UNI-1 is genuinely good at following edit instructions, but "fix it" gets you nothing useful. Say what changed and what stayed.
• Seed won't reproduce a result. Same as every partner node - don't queue for reproducibility.
• Cost scales with the model. uni-1-max costs more per edit; for most fixes the plain uni-1 is enough. The price badge on the node tells you which you've selected.
• No fine-grained region control. This isn't a mask-based editor - Ideogram V3 or a local inpaint node is the right tool if you need pixel-precise edits. UNI-1 is for whole-scene, prompt-driven changes where describing the outcome is natural.