Luma and Hedra: complementary creative tools and infrastructure
Luma builds creative tools and the Ray video model. Hedra builds training and serving infrastructure.
POST /v3/models/luma-ray-32
export HEDRA_API_KEY="<key_id>:<secret>"
curl -X POST \
"https://api.hedra.com/v3/models/luma-ray-32" \
-H "Authorization: Key $HEDRA_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "input": {
"prompt": "A slow dolly through a rain-lit Tokyo alley at night",
"aspect_ratio": "16:9",
"resolution": "1080p",
"duration_ms": 5000
} }'1080p · 5s · 40¢/second · $2.00 per clip
Run sovereign inference
Hedra's own inference engine serves open-weight models on hardware you operate — sovereign, air-gap ready, at frontier efficiency on any silicon.
Runs onNVIDIA CUDA · AMD ROCm · GOOGLE TPU · AWS TRAINIUM · CUSTOM ASICS
Hedra and Luma: focus at a glance
Luma builds and ships models for creatives. Hedra is an inference lab, which provides pre- and post-training technology as a service
| Dimension | Hedra | Luma |
|---|---|---|
| Primary focus | Managed inference, training and visual research | Video models and creative tools |
| Core technology | A custom inference engine and efficiency research | Ray 3.2 and the Luma model family |
| Model access | Hosts open-weight and proprietary models, Ray 3.2 among them | Builds and ships the Luma models |
| How you connect | API, SDK, CLI, MCP | Luma Agents |
| Where inference runs | Hedra's managed cloud, or your own cluster | Luma's cloud |
| Best fit | Developers, sovereign clouds and physical AI teams | Creators, filmmakers |
Hedra's infrastructure runs open models faster and cheaper, access partner models through our gateway
Ray 3.2 is accessible alongside Hedra managed open source and partner models
| Provider | Model ID | Rate |
|---|---|---|
| MiniMax | minimax-h3 | 8¢/s · 768p |
| ByteDance | seedance-25 | 79.09¢/s · 1080p |
| Kling | kling-v3 | 16.8¢/s · 1080p |
| Black Forest Labs | flux-3 | 29¢/s · 1080p |
| Luma | luma-ray-32 | 40¢/s · 1080p |
| gemini-omni-flash | 12.67¢/s · 720p | |
| Alibaba | wan-3-0 | 20¢/s · 1080p |
Calling Ray 3.2 through Hedra is one line.
Leverage Ray 3, or switch to an open model running on your infrastructure in one line of code
Direct to Luma
POST agents.lumalabs.ai/v1/generations
Authorization: Bearer $LUMA_AGENTS_API_KEY
{ "model": "ray-3.2", "prompt": "A slow dolly
through a rain-lit Tokyo alley at night" }One provider per integration. A second model means a second credential and a second invoice.
Through Hedra
POST api.hedra.com/v3/models/luma-ray-32
Authorization: Key $HEDRA_API_KEY
{ "input": { "prompt": "A slow dolly through
a rain-lit Tokyo alley at night", ... } }Every provider behind one credential. Swap luma-ray-32 for veo-31 or sora-2-pro and nothing else changes.
FAQs
- How do Hedra and Luma fit together?
- They sit at different layers of the same stack. Luma builds the models and the creative products around them. Hedra is the inference, training and model infrastructure teams build on.
- Is this the same Ray 3.2 that Luma ships?
- Yes. Hedra does not have a hosting agreement with Luma, so the model is proxied via Luma's API at the same price. If you have a license for Luma's model and would like us to host it for you, please reach out.
- Are there open alternatives to Ray 3.2 we can host?
- Yes. Hedra offers a large catalog of open-weight models with managed inference that can match the performance of proprietary models.
