Overview

UNI-1 is a unified image model in the UNI-1 family, built for both image creation and precision image editing.

It is best suited to workflows that need more than prompt-only generation: structured image modification, multi-reference control, seed-based iteration, niche style guidance, and stronger reasoning over composition, space, and plausibility.

Strengths

Create and Modify in One Model

UNI-1 supports two core operating modes: creating new images from prompts and references, and modifying an existing source image while preserving structure unless explicitly changed. This makes it useful for workflows that move from ideation into controlled refinement.

Balanced General-Purpose Variant

Within the UNI-1 family, this variant is the broader day-to-day choice for image generation and editing tasks that need flexibility across prompts, references, and controlled modifications.

Multi-Reference Control

The model supports up to 9 reference images and responds best when each reference is assigned a role such as character, style, composition, lighting, color palette, texture, or mood. This gives more direct control than loosely guided reference workflows.

Reasoning-Informed Image Editing

UNI-1 is built to decompose complex edit instructions, resolve constraints, and preserve scene coherence while transforming an image. It is particularly useful when edits require spatial, causal, temporal, or logical consistency rather than only stylistic change.

Strong Directed Editing

In Modify mode, UNI-1 is designed for precise, source-grounded edits. It works well for changing time of day, weather, materials, atmosphere, style, or scene details while keeping the parts of the image that should remain unchanged.

Style and Culture Awareness

UNI-1 is tuned to handle culturally specific visual languages, meme formats, manga aesthetics, and narrow niche styles while preserving identity and composition across variants.

Seeded Iteration and Aspect Ratio Range

The model supports seed control for reproducibility and offers a wide aspect-ratio range for different use cases, including square, portrait, widescreen, panoramic, and ultra-tall formats.

Capabilities

Text-to-Image

UNI-1 supports prompt-driven image generation for brand-new compositions, with optional references to guide style, identity, composition, and other visual layers.

Image-to-Image

UNI-1 supports precision editing of an existing source image, preserving composition and structure unless the prompt explicitly changes them.

Multi-Reference Generation

The model can use multiple references in Create mode and benefits from explicit role assignment for each reference.

Multi-Turn Refinement

UNI-1 is designed for iterative refinement workflows where prompts, references, and seeds are adjusted deliberately over several generations.

Input and Output

• AIR ID: luma:uni@1
• Input: text prompts with optional reference images, or a source image plus edit instructions
• Output: generated or modified images
• Reference images: up to 9
• Seed support: supported

Best Fit

• Controlled image editing and refinement
• Multi-reference art direction
• Character consistency workflows
• Style transfer with preserved structure
• Concepting that moves into precise visual iteration