H3 Max

submit to result wait

A 5-second 768P product turntable completed in one attempt. Voyager reports no reroll or retouch in the common nine-clip fixture; the listed cost is an estimate from the published promotional rate.

Reported metric: 6 seconds

Reported result: 5s clip; one attempt; estimated cost $0.20 at the published promotional rate.

Prompt: A clean product shot that holds its shape through a full slow rotation. (Full prompt preserved at source.)

Settings: endpoint: minimax/h3-max/text-to-video · resolution: 768P · duration seconds: 5 · aspect ratio: 16:9 · prompt expansion mode: disabled

submit to result wait

A busy 768P text-to-video scene retained distinct moving people and objects in Voyager's test, while submit-to-result wait was materially longer than the backend inference examples.

Reported metric: 26 seconds

Reported result: Distinct moving people and objects without merging; estimated cost $0.20.

Prompt: A crowded Saturday farmers market seen from slightly above, shoppers with tote bags, a dog on a leash, a busker with a guitar, bunting swaying, late morning light, gentle camera drift.

Settings: endpoint: minimax/h3-max/text-to-video · resolution: 768P · duration seconds: 5 · aspect ratio: 16:9 · prompt expansion mode: disabled

submit to result wait

An image-to-video test at 768P kept the source silhouette recognizable, but Voyager reports that the dripper filled like a cup rather than following the requested physical action.

Reported metric: 6 seconds

Reported result: Output returned at 768x768; source silhouette remained recognizable but the dripper filled like a cup.

Prompt: The coffee dripper slowly fills as coffee drips through it, steam rising, the camera drifting gently around it, the dripper itself unchanged.

Settings: endpoint: minimax/h3-max/image-to-video · resolution: 768P · duration seconds: 5 · prompt expansion mode: disabled

submit to result wait

The camera-control test kept the product recognizable; Voyager cautions that precise orbit accuracy is difficult to establish from the near-symmetric object used.

Reported metric: 6 seconds

Reported result: Product remained recognizable; precise 3D reconstruction was not established.

Prompt: The coffee dripper is rigid and motionless. Only the camera moves. Preserve its shape and materials and the white background.

Settings: endpoint: minimax/h3-max/camera-controls · resolution: 768P · duration seconds: 5 · camera trajectory: [{"time":0,"azimuth":0,"elevation":0,"distance":1},{"time":1,"azimuth":45,"elevation":10,"distance":1}]

submit to result wait

A lip-sync adapter test retained the supplied audio and changed mouth shape. Voyager explicitly did not score synchronization quality by ear.

Reported metric: 11 seconds

Reported result: Estimated cost $0.6895; supplied audio was retained and mouth shape changed.

Prompt: Animate the generated portrait to the supplied synthetic ad-read audio.

Settings: endpoint: minimax/h3-max/lip-sync/image-to-video · resolution: 480P · duration seconds: 14

reported result

In a 768P mechanical-transformation test, the main assembly was readable and the design survived into wing beating, but the requested fly-past was missed by the late frame.

Reported result: Main assembly was readable and design survived into wing beating, but the requested fly-past was missed by the late frame.

Prompt: One continuous 5-second macro shot in a sunlit watchmaker's workshop... loose brass gears and cobalt-blue plates rapidly snap together into one mechanical hummingbird... then launches forward before arcing past the camera. (Full prompt preserved at source.)

Settings: resolution: 768P · duration seconds: 5 · aspect ratio: 16:9 · passes: 1 · prompt expansion mode: balanced · safety checker: true

submit to result wait

A 768P vertical action test measured 6.97 seconds submit-to-result wait while fal inference was reported as 2.79 seconds. Subject and motion remained coherent in sampled frames; fine wheel geometry softened at peak speed.

Reported metric: 6.97 seconds

Reported result: fal inference 2.79s; measured wait 6.97s. Subject and motion remained coherent; fine wheel geometry softened at peak speed.

Prompt: One continuous 5-second vertical documentary shot on a rainy city side street at blue hour... bicycle messenger ... sharp corner ... puddle ... accelerates out of frame. (Full prompt preserved at source.)

Settings: resolution: 768P · duration seconds: 5 · aspect ratio: 9:16 · passes: 1 · prompt expansion mode: balanced · safety checker: true

submit to result wait

A 768P square stop-motion paper-craft test measured 7.29 seconds submit-to-result wait while fal inference was reported as 1.28 seconds. The broad transformation succeeded and NORTH was legible, but the opening form was not exact.

Reported metric: 7.29 seconds

Reported result: fal inference 1.28s; measured wait 7.29s. Broad transformation succeeded and NORTH was legible; opening form was not exact.

Prompt: One continuous 5-second stop-motion paper-craft shot... folded cobalt-and-coral transit map opens ... six miniature buildings unfold ... sign with the single word NORTH. (Full prompt preserved at source.)

Settings: resolution: 768P · duration seconds: 5 · aspect ratio: 1:1 · passes: 1 · prompt expansion mode: balanced · safety checker: true