Video generation with sound is live on OpenAlchemy

Video generation is now live on OpenAlchemy. It runs the open-weights MiniMax H3 model on Grid compute — community GPUs — and every shot comes with sound: native stereo audio is generated together with the picture, not dubbed on afterwards.

What a request produces

Each request renders one shot of 4–15 seconds at 768P, 24 fps, with stereo audio. As a reference point, a 5-second 768P shot takes roughly 2.5 minutes on an RTX 5090-class GPU (measured).

Three modes

Text-to-video (t2v) — describe the shot in a prompt.

Image-to-video (i2v) — give a first frame, a last frame, or both. Chaining shots through the previous shot's last frame lets you build longer films one shot at a time.

Reference-to-video (ref) — guide the shot with up to 9 reference images.

The API

Video is asynchronous. POST /v1/videos returns 202 with a job id; poll GET /v1/videos/{id} until the status is succeeded, then download the file from GET /v1/videos/{id}/content:

# 1. Submit a shot — returns 202 with a job id
curl https://api.openalchemy.io/v1/videos   -H "Authorization: Bearer $OPENALCHEMY_API_KEY"   -H "Content-Type: application/json"   -d '{"model":"minimax-h3","mode":"t2v","prompt":"...","duration":5,"resolution":"768P","aspect_ratio":"16:9"}'

# 2. Poll until status is "succeeded"
curl https://api.openalchemy.io/v1/videos/$VIDEO_ID   -H "Authorization: Bearer $OPENALCHEMY_API_KEY"

# 3. Download the finished shot
curl -o shot.mp4 https://api.openalchemy.io/v1/videos/$VIDEO_ID/content   -H "Authorization: Bearer $OPENALCHEMY_API_KEY"

Billing is per output second. The estimate is held while the shot renders and refunded if it fails.

Prefer to try it without code? Open the Grid Playground → Video.

For GPU providers

GPU providers with a 32 GB card can serve video jobs. There is also a new headless Linux worker — CLI only, no GUI — on the Download page. Pair it with a code from the Grid dashboard and it receives jobs over the network.