Luma Uni-1 – Artlist

Published: August 6, 2026

Image Model Introduction

Luma Uni-1 is available in two variants: Standard (the default tier, lower cost) and Max (higher quality, roughly 2.5× the price).

Luma Uni-1 is a "reasoning-first" image model that generates and edits images from text and reference images.

Text-to-Image (t2i)

Reasoning-First Generation: Studies the prompt and plans composition before generating.
Text Rendering: Great multilingual text rendering.
Accurate Counting & Placement: Handles exact object counts and spatial positions from prompt.
Manga Style Preset: Dedicated manga style mode.

Image Editing (i2i)

Text-Driven Edits: Modify clothing, backgrounds, props, or lighting via natural language.
Composition Preservation: Edits content while keeping the original layout intact.

Multi-Reference Control

Role-Tagged References: Assign style, character, composition, or lighting roles to each reference.
Character Consistency: Lock character identity across outputs using reference images.
Multi-Reference Style Transfer: Combine style, texture, and mood from separate source images.

Inputs: Text-to-Image · Image Edit · Multi-Reference Generation (up to 9 reference images for generation, up to 8 for editing)
Resolution: Automatic, up to 2048 px on the long edge, based on aspect ratio
Aspect Ratios: 3:1 · 2:1 · 16:9 · 3:2 · 1:1 · 2:3 · 9:16 · 1:2 · 1:3

Resolution ceiling: max ~2048 px on the long edge. Larger canvases require tiling or upscaling.

Prompt Formula: [Scene anchor], [main subject], [visual style], [camera & composition], [lighting & mood], [text / layout / constraints]

Be explicit with counts: "exactly three cups on the table" — Uni-1 handles constraint language well.
Assign reference roles in the prompt: "ref 1 for style, ref 2 for character face, ref 3 for lighting."
For editing, speak like a compositor: "Keep the pose and background, but change the outfit to X."
Use subsequent outputs as references to build character/style consistency across a set.
Prompt max is 6,000 characters — use the space. Detailed, structured prompts outperform short ones.
Style preset is "auto" by default. Switch to "manga" for comic/anime panel work.
Enable web search when the prompt references real products, places, or people for better grounding.
Start with Standard tier for iteration, switch to Max for hero-quality finals.