Video generated faster than it plays.
Four CG dancers on a cyan-lit stage, synchronized footwork
345 frames - 14.375 s - 13.5 s to generate
Traceur vaults a rooftop ledge, tracking shot alongside
345 frames - 14.375 s - 13.5 s to generate
Stage performance, front-loaded prompt variant
345 frames - 14.375 s - 13.5 s to generate
Unedited output from the deployed configuration, transcoded for the web. The API returns 1344x768 mp4 with the audio track the model generates in the same pass.
13.506 s
to generate 14.375 s of video
0.94x
realtime, end to end
1344x768
24 fps, native synced audio
8xH100
sequence-parallel, one node
An asynchronous job, not a long request
POST returns immediately with a job id. Poll it, then fetch the mp4. A warm worker finishes in about 13.5 seconds; a cold one spends several minutes loading 144 GB of weights and warming compiled kernels, and you can tell the two apart instead of watching a request hang.
Every generation is priced on the video it produces — $0.02 per second of output, $0.29 for a full 14.375 s clip — so the bill follows the request, not our GPU luck. A job that fails is refunded automatically.
Credits are prepaid. You cannot be surprised by an invoice, and a leaked key cannot run up a bill larger than the balance behind it.
curl -X POST https://api.highlander.sh/v1/generations \
-H "Authorization: Bearer $HIGHLANDER_KEY" \
-H "Content-Type: application/json" \
-d '{"prompt": "A traceur sprints across a rooftop and vaults a
concrete ledge, tucking into a roll. Late afternoon sun."}'
# -> 202 {"id":"...","status":"queued","poll":"/v1/generations/..."}
# poll until succeeded, then:
curl -L https://api.highlander.sh/v1/generations/$ID/video \
-H "Authorization: Bearer $HIGHLANDER_KEY" -o out.mp4Start with $1.00 of credit
Enough for a full-length clip. Add credit when you want more; nothing recurring, no seat minimums, no sales call.
Open the playground