Luma AI
Listed at https://lumalabs.ai
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | Credit-based subscription. Plans: Plus $30/mo (10,000 credits), Pro $90/mo (40,000 credits), Ultra $300/mo (150,000 credits). Ray 3.2 video: 1080p text-to-video 400 credits/5s, 720p 100 credits/5s, Draft 20 credits/5s; Seedance 2.0 1080p 240 credits/sec, 4K 959 credits/sec. | per million tokens |
| Output | Image credits: Uni-1 30 credits/image, Seedream 1-3 credits/image, GPT Image 2 from 3 (Low-1K) to 255 (High-4K) credits. Audio: ElevenLabs v3 TTS 21 credits/1,000 chars. Utilities: background removal 1 credit/image, reframe video 32 credits/sec, upscale to 4K 17 credits/sec. | per million tokens |
🤖 Supported Models (12)
✨ Pros
- ✓Ray 3.2 flagship video API: 1080p output + up to 16 keyframes (Multi-Keyframe) for frame-by-frame shot control
- ✓Single async REST surface (POST /v1/generations → poll → presigned URL download) covering image + video + audio
- ✓Uni-1 / Uni-1.1 / Uni-1-Max image models: reasoning + generation in one call, high first-pass success, claimed half price / half latency
- ✓Aggregated third-party frontier models: Veo 3.1, Kling 3.0, Seedance 2.0, GPT Image 2, Nano Banana in one API
- ✓V2V (video-to-video) up to 20s + HDR generation and 16-bit EXR export, compositable into Resolve/Nuke pipelines
- ✓Official Python / TypeScript / Go SDKs (luma_agents package), turnkey async generation workflow
- ✓Evolved from creative platform to API-first: one brand system across markets and formats, aimed at e-commerce and content pipelines
⚠️ Cons
- ×No permanent free API tier; entry at Plus $30/mo
- ×China access requires proxy (no mainland direct edge nodes), higher latency
- ×Credit-based billing across many model/resolution combinations makes cost estimation complex
- ×High-res long video (e.g. Seedance 4K) burns nearly a thousand credits per clip, so budgets grow fast
- ×Video generation is async polling; more keyframes / longer clips mean longer waits
- ×No function calling / tool use — pure generation API, not suited for agent orchestration
- ×No affiliate program
🎯 Best For
Teams needing production-grade video generation (e-commerce ads, brand marketing, game CG trailers); multimodal content pipelines (image + video + audio in one API); developers who want Veo / Kling / Seedance frontier models without separate integrations; post-production (HDR / EXR / V2V compositing workflows)
💰 Pricing & Plans
| Plan | Monthly Price | Credits / month | Best For |
|---|---|---|---|
| Plus | $30 | 10,000 | Hobbyists, small creator teams, API evaluation |
| Pro | $90 | 40,000 | Freelancers, agencies, active production (4x Luma Agents usage) |
| Ultra | $300 | 150,000 | Studios, high-volume pipelines (15x Luma Agents usage) |
| Team | Custom | Shared credits | Teams with members, analytics, SSO |
| Enterprise | Custom | Custom | Committed usage, fine-tuning, training |
🔧 API & Developer Experience
- •API Surface: Luma Agents API is a single async REST surface at api.lumalabs.ai (docs at docs.agents.lumalabs.ai). Workflow: POST /v1/generations to submit, GET /v1/generations/{generation_id} to poll, then download output from presigned URLs once state reaches 'completed'. No OpenAI-compatible chat endpoint — it is a generation API, not a chat API.
- •Models: Image: uni-1 and uni-1-max (reasoning + generation in one endpoint). Video: Ray 3.2 (flagship, 1080p, multi-keyframe up to 16 frames, V2V up to 20s). Third-party models are also served: Veo 3.1, Kling 3.0 / Kling Omni / Kling 2.6, Seedance 2.0, MiniMax H3, Flux 3, GPT Image 2, Nano Banana, Seedream.
- •Generation Modes: Text-to-video and image-to-video (Ray 3.2), video-to-video (V2V, up to 20s on Ray 3.2), image creation/edit (Uni-1), reframing, and utilities (background removal, upscale). Multi-Keyframe lets you define up to 16 keyframes inside a single clip for frame-by-frame shot control.
- •Output & Formats: Video output at Draft / 540p / 720p / 1080p (and up to 4K for Seedance). Supports native HDR generation and 16-bit EXR export so AI work composites alongside live-action plates in Resolve or Nuke. Images at 1K / 2K / 4K depending on model.
- •SDKs & Tooling: Official Python SDK (luma_agents package, pip installable), TypeScript, and Go SDKs. The Python SDK wraps the async submit-poll-download flow: client.generations.create(prompt=..., aspect_ratio=...) then poll for completion and download. cURL examples are in the docs.
- •Rate Limits & Errors: Rate limits are tiered by plan (Plus lower, Ultra/Premier higher per-minute ceilings). Errors are returned in REST JSON format; the docs list common error messages. The API is designed for asynchronous batch workflows rather than low-latency chat.
- •Authentication: API key obtained from the Luma API Platform (platform.lumalabs.ai) or via the Luma Agents API key (set as LUMA_AGENTS_API_KEY). The SDK picks it up from the environment variable automatically.
🎬 Video Generation Capabilities
Luma's defining capability is cinematic video generation through the Ray 3.2 API: 1080p output with Multi-Keyframe control that lets you set up to 16 keyframes inside a single clip, directing the cut frame by frame rather than hoping the model gets the sequence right. Video-to-video (V2V) runs up to 20 seconds, and native HDR generation plus 16-bit EXR export means AI composites sit alongside live-action plates in Resolve or Nuke without a tone-mapping bottleneck. Reframe handles format variants from one master shot, and upscaling pushes output to higher resolutions.
🎬 Video Generation Capabilities (cont.)
Beyond first-party Ray models, Luma aggregates the industry's strongest third-party video models behind one API: Veo 3.1, Kling 3.0, Kling Omni, Kling 2.6, Seedance 2.0 (up to 4K), MiniMax H3, and Flux 3. This turns Luma into a multi-model gateway for video the way OpenRouter is for text — one async interface, several model families, credit-based metering. For brand work, a single prompt set can scale across markets and formats while retaining visual consistency, which is why e-commerce and ad-production teams use it for variant generation at volume.
🌐 Regional Availability & Latency
Luma AI is a US-based company (San Francisco, CA) serving the API from global AWS/GCP edge infrastructure. There is no mainland China regional endpoint or official China access program, so production access from China requires a stable overseas proxy or an aggregator that fronts the Luma API. Generation is inherently asynchronous — submission acknowledgment is fast (sub-second over a good connection), but a 5-second 1080p Ray 3.2 clip typically completes in 30-120 seconds depending on plan tier and queue depth.
🌐 Regional Availability & Latency (cont.)
For production workloads serving China-based end users, two paths exist: (1) a proxy-based integration with Luma directly, accepting higher latency and async wait times; or (2) routing through a domestic alternative such as ByteDance Doubao video models or MiniMax, which offer mainland-direct endpoints and lower latency for short-video-format content. Luma does not publish a regional endpoint map, so teams targeting Chinese users should budget for proxy infrastructure and evaluate a domestic video-generation fallback for latency-sensitive workloads.