Models

The Luma Agents API has models for two media types: uni-1 and uni-1-max for image generation and editing, and ray-3.2 for video. They share the same endpoint, request envelope, and submit-poll-download flow — the supported type values and model-specific fields differ.

At a glance

Model	Supports	Use for
uni-1	image, image_edit	Default image generation and editing
uni-1-max	image, image_edit	Higher-quality image output than uni-1
ray-3.2	video, video_edit, video_reframe	Text-to-video, image-to-video, video editing, and aspect-ratio reframing

See Pricing for current rates and Provisioned Throughput plans.

Capabilities

Image models — uni-1, uni-1-max

Capability	Supported
Text-to-image (type: "image")	✓
Image editing (type: "image_edit")	✓
image_ref for style or content guidance (up to 9)	✓
source image for editing	✓
web_search grounding	✓
Output formats png / jpeg	✓

uni-1-max produces higher-quality output than uni-1; the two share the same wire format and parameter set.

Video model — ray-3.2

Capability	Supported
Text-to-video (type: "video")	✓
Image-to-video (video.start_frame / video.end_frame)	✓
Multi-keyframe image-to-video (video.keyframes + video.keyframe_indexes, up to 64 anchors)	✓
Video editing (type: "video_edit" with source.generation_id, source.url, or source.data)	✓
Video reframing (type: "video_reframe")	✓
Extend a prior video (video.start_frame.generation_id / video.end_frame.generation_id)	✓
Per-signal edit controls (video.edit.controls)	✓
HDR output (video.hdr, 720p/1080p)	✓
EXR export (video.exr_export, requires hdr: true)	✓
Seamless looping (video.loop, type: "video" only)	✓

Image model strengths

Both uni-1 and uni-1-max accept the same parameter set; the difference is output quality and per-image price.

Capability	Description
Text rendering	Renders readable text on signs, labels, book covers, and other surfaces. Put exact text in quotes in your prompt.
Spatial reasoning	Produces accurate shadows, correct perspective, and physically plausible object layouts.
Cultural styles	Understands visual traditions — manga panels with screentone shading, ukiyo-e woodblock prints, film noir lighting, art house cinema framing, and more.
Reference-guided generation	Accepts up to 9 reference images via image_ref for text-to-image, or up to 8 for image editing (where the source image occupies its own slot).
Image editing	Modifies existing images via source — change backgrounds, swap objects, transfer styles while preserving unmentioned parts.
Web search grounding	When web_search: true, searches the web for visual references before generating.
Multi-panel output	Generates multi-panel sequences (e.g., storyboards) with consistent style when described in the prompt.

Output specifications

Aspect ratios

Aspect ratios are model-and-type-dependent. The shared AspectRatio enum has twelve members, but no single (model, type) pair accepts all twelve. See:

• Image generation — aspect_ratio — uni-1 / uni-1-max accept nine ratios
• Video generation — aspect_ratio — ray-3.2 video accepts six ratios
• Video editing — aspect ratio is derived from the source video and is silently ignored on the request
• Video reframing — aspect ratio is required and names the target shape

Output formats — images

Format	Best for
png	Lossless quality
jpeg	Smaller file size, photographs

When output_format is omitted, the model picks a format based on the prompt. Applies to image models only.

Output formats — video

Video output is delivered as MP4. When video.hdr: true, the MP4 is HDR-encoded; adding video.exr_export: true exports an EXR file alongside it for professional colour-grading workflows.

Presigned URL expiry

Generated images and videos are delivered as presigned URLs that expire after 1 hour. Download promptly or request a fresh URL by polling GET /v1/generations/{id} again.

uni-1

The default image model. Use it for the broad set of image-generation and image-editing tasks — see Image model strengths above. Defaults to model: "uni-1" when omitted from the request.

See Pricing — uni-1 for the per-task table.

uni-1-max

uni-1-max produces higher-quality output than uni-1. Same wire format and parameter set — same prompts, same aspect_ratio, same image_ref semantics. Pass "model": "uni-1-max" on POST /v1/generations.

{ "model": "uni-1-max", "prompt": "A neon-lit Tokyo alley in the rain" }

See Pricing — uni-1-max for the per-task table.

ray-3.2

Ray 3.2 generates video from text or from anchor images, edits existing videos, extends previous generations, and reframes videos to a new aspect ratio:

• Text-to-video — type: "video" with a prompt
• Image-to-video — type: "video" with video.start_frame and/or video.end_frame as ImageRef
• Video extend — type: "video" with exactly one prior generation id in video.start_frame.generation_id or video.end_frame.generation_id
• Video editing — type: "video_edit" with a source video: source.generation_id, a hosted source.url, or inline source.data (the last two with a video/* source.media_type)
• Video reframing — type: "video_reframe" with a source video and target aspect_ratio

Resolutions: 540p, 720p, 1080p. Durations: 5s, 10s. Aspect ratios: 9:16, 3:4, 1:1, 4:3, 16:9, 21:9.

loop is create-only (type: "video" only). Video edits add video.edit for conditioning — auto_controls for a model-derived schedule, or strength and per-signal controls for manual tuning. Video reframing is standard dynamic range only and rejects video.edit, loop, start_frame, and end_frame.

{
  "model": "ray-3.2",
  "type": "video",
  "prompt": "A slow dolly shot through a misty greenhouse at sunrise",
  "video": { "resolution": "720p", "duration": "5s" }
}

See Video generation for full parameters, Video editing for edits, Video reframing for aspect-ratio changes, and Pricing — ray-3.2 per-video pricing for rates.

Next steps

• Image generation — Full parameter reference for type: "image"
• Image editing — Modify existing images
• Video generation — Ray 3.2 text-to-video and image-to-video
• Video editing — Ray 3.2 video editing
• Video reframing — Ray 3.2 aspect-ratio reframing
• Pricing — Pay-as-you-go and Provisioned Throughput plans
• FAQ — Quick answers to common questions
• API Reference — Complete endpoint specifications