Luma AI

Agents That Research, Generate, and Refine

Luma Agents research, generate, and refine across video, image, audio, and text, with your brand holding steady through all of it. The platform is positioned as a creative partner: you have the idea, Luma helps make it real.

Ten directions before the morning standup. Fifty variants by end of day.

Ray 3.2: Direct Every Frame, Finish Every Cut

Ray 3.2 lets you direct video the way you would direct a shoot, putting every frame under your control, holding continuity across every cut, and finishing in cinematic quality. It supports Text-to-Video, Image-to-Video, Video-to-Video, and Reframe across Draft, 540p, 720p, and 1080p, with SDR base, HDR (2x), and HDR+EXR (3x) options.

Frame-level control with continuity across cuts, finished in cinematic quality.

Uni-1: Brand Intelligence at the Model Level

Uni-1 understands your brand as deeply as it generates for it, carrying style references, character references, and visual grounding into every asset. It is available for image Create/Modify at 30 credits per image, and a Uni-1.1 API was introduced for directing intelligence and shipping aesthetics.

Work that is right for your brand, your audience, and this exact moment.

Frontier Research Ships Straight to Creatives

Speed like this doesn't come from a wrapper: Luma builds the models underneath Luma, and frontier research ships straight into the hands of working creatives. Published engineering work includes Terminal Velocity Matching (Nov 26, 2025), the Ray3 Evaluation Report (Oct 14, 2025), and Inductive Moment Matching (Mar 11, 2025).

Model advances reach working creatives rather than staying in the lab.

The Open Physical AI Lab

An open science effort to solve generalization in physical AI, built in the open for the benefit of all humanity. It sits alongside Luma's stated mission to build unified general intelligence that can generate, understand, and operate in the physical world.

Open, shared progress toward generalization in physical AI.

Luma Skills: Build a Workflow Once, Run It Forever

Introduced June 16, 2026, Luma Skills lets you build a creative workflow once and run it forever. It complements a model and tool lineup that spans video, image, audio, and utilities such as background removal, image blending, reframing, upscaling, and layers.

Repeatable creative workflows instead of one-off generations.

• Frontier research ships directly into the productLuma builds the models underneath its platform, so frontier research ships straight into the hands of working creatives. Recent releases include Ray 3.2, Uni-1.1 API, and Luma Skills, all announced on the site.
• Transparent credit-based pricing with clear per-model costsThe pricing page lists exact credit costs for video, image, and audio generation across models like Ray3.2, Kling, Veo, Seedance, and ElevenLabs, with no hidden fees. Plans start at $30/month for Plus, $90/month for Pro, and $300/month for Ultra.
• Multiple third-party models available in one platformLuma offers access to models including Ray3.2, Ray3.14, Seedance 2.0, Kling Omni, Kling 3.0, Kling 2.6, Veo 3.1, Veo 3, Minimax H3, Flux 3, GPT Image 2, Nano Banana, Seedream, and ElevenLabs, all with listed pricing.
• Enterprise and team plans with collaboration featuresTeam plans include member management, projects, team-wide sharing, usage analytics, shared credits, and SSO. Enterprise adds commitments, dedicated education and training, and custom fine-tuning.

• Credit costs vary significantly by model and resolutionFor example, Ray3.2 text-to-video costs 20 credits/5 sec at draft, but 1200 credits/10 sec at 1080p HDR+EXR. Seedance 2.0 4K costs 959 credits/sec. Always check the pricing table before generating to avoid surprise credit usage.
• Audio generation costs are per unit, not per secondElevenLabs v3 text-to-speech costs 21 credits per 1000 characters, SFX v2 costs 25 credits per minute, and Music v1 costs 98 credits per minute. Utilities like background removal cost 1 credit per image.
• Some models have no audio supportThe pricing table shows that Ray3.2, Ray3.14, and Kling 2.6 do not list audio support, while Seedance 2.0, Kling Omni, Kling 3.0, Veo 3.1, Veo 3, Minimax H3, and Flux 3 do. Verify audio needs before choosing a model.
• Yearly billing is required for the displayed monthly pricesThe Plus, Pro, and Ultra plans show monthly prices but note they are billed yearly (e.g., $300 billed yearly for Plus). Monthly billing may differ, and no monthly-only pricing is shown on the page.

Video Production

Luma's Ray 3.2 model gives you direct control over every frame, holding continuity across cuts and finishing in cinematic quality. You can generate text-to-video, image-to-video, and video-to-video content at resolutions up to 1080p, with options for HDR and EXR. The platform also offers utilities like upscaling to 4K and reframing, making it a comprehensive tool for professional video workflows.

Image Generation

Luma's Uni-1 model understands your brand at the model level, carrying style references and visual grounding into every asset. You can create or modify images with models like GPT Image 2, Nano Banana, and Seedream, with quality options from low to high and resolutions up to 4K. This enables you to produce on-brand visuals quickly and consistently.

Audio Creation

Luma supports audio generation including text-to-speech with ElevenLabs v3, sound effects with ElevenLabs SFX v2, and music generation with ElevenLabs Music v1. You can also isolate vocals from existing audio. This makes it easy to add voiceovers, sound effects, and background music to your video or image projects.

Creative Workflow Automation

Luma Agents research, generate, and refine across video, image, audio, and text, allowing you to explore ten directions before your morning standup and produce fifty variants by end of day. With Luma Skills, you can build a creative workflow once and run it forever, streamlining repetitive tasks and accelerating your creative process.

Define the exact 3D Tools workflow Luma AI should support.

Compare it with closely related AI tools in the same category before committing.

Set review rules for accuracy, privacy, brand voice, compliance, and final approval.

Connect useful outputs to the wider stack instead of leaving them inside the AI tool.