How to prompt Luma Uni-1

How to structure your prompt

Create Image mode only. Direct natural language describing the subject, setting, action, composition, visual style and light. Use specific, named aesthetics (for example '1970s Italian giallo film poster, high-contrast color blocking') and concrete technical terms ('golden hour, 85mm lens, shallow depth of field') instead of vague praise words. Aim for 80-250 words. When the user has uploaded references, label each one and give it a single role. Choose the aspect ratio and seed in Create Image; use Modify Image separately for changes to an existing image.

Length and syntax

Up to 6,000 characters. Longer prompts are not better if the extra words do not change what appears on screen.

Reference labels: 'Use IMAGE1 ([short description]) as a [ROLE] reference.' Roles: style, character, composition, color palette, lighting, texture, mood. One role per image; with several references, end with a line stating that each reference controls only its assigned layer. No parameter flags.

Prompt skeleton

[Image type and named aesthetic] of [subject with specific details] [action] in [setting]. [Composition and framing]. [Lighting, lens and depth of field]. [Color and mood]. [If references: Use IMAGE1 ([description]) as a [ROLE] reference.]

What it’s best at

• niche and specific visual styles
• visual composition
• creative exploration
• natural language direction
• up to 9 role-labeled reference images
• seed reproducibility

Known limitations

• editing an existing image is a different workflow
• reference images need to be uploaded separately
• unlabeled references are guessed and unreliable
• conflicting instructions can cause parts of the prompt to be ignored

Avoid in your prompt

• pretending references exist when none are uploaded
• unrequested lettering
• unsupported flags
• vague terms such as 'beautiful' or 'amazing'
• redundant or conflicting instructions
• references without a stated role