MiniMax H3 Max renders 15-second 720p clips in about 16 seconds

• One 15-second test produced the sharpest timing comparison: ozansihay's timed test logged 5.17 seconds and $0.7125 at 480p, versus 17.32 seconds and $0.855 at 720p.
• The 720p result showed up again in two separate clips: ozansihay's follow-up reported 15.45 seconds, while ozansihay's second 720p test reported 16.24 seconds at $0.855.
• Multi-shot prompt sheets are doing real directorial work: Magnific's creature test specifies three shots, a reveal, sound design and a whispered line in one generation.
• Magnific launched H3 Max as a fal-tuned MiniMax H3 variant, with unlimited 480p generations for Premium+ and Pro users for three days and credit-paid 768p and 2K upgrades, according to Magnific's launch post.

A Magnific octopus clip puts footage, narration, room-tone-like ambience and score into a 12-second prompt. A Magnific runway clip asks for heels to land on an industrial bass track, while ozansihay's multi-clip test takes the opposite route, building four shots for a later Premiere and Topaz finish.

These are individual creator timings, not a controlled benchmark. The useful pattern is that multiple 15-second 720p examples clustered near the length of the finished clip.

The original side-by-side test lists:

• 480p: 15 seconds of video in 5.17 seconds, billed at $0.7125.
• 720p: 15 seconds of video in 17.32 seconds, billed at $0.855.

Other posted runs put more numbers around that result:

• A second 15-second 720p output took 15.45 seconds in ozansihay's follow-up.
• A historical battle sequence took 16.24 seconds at $0.855 in ozansihay's second 720p test.
• chrisfirst's 9-second test took 6.75 seconds.
• kaigani's 5-second 720p test took 2.6 seconds.

Before the model was identified, bennash anticipated a service that could make a five-second clip in roughly four seconds in a pre-launch post. fal's H3 Max model page now claims a five-second video in under three seconds.

Resolution labels need care. The creator posts call these outputs 720p, while the current text-to-video API lists native 480P and 768P outputs, with durations from five to 15 seconds. Its listed promotional rates, $0.025 per second at 480p and $0.04 at 768p, also do not exactly match the two billed totals in ozansihay's test; the documentation says those rates double after September 1.

The examples treat 15 seconds as a timed sequence of shots, not a single scene description.

Magnific's stop-motion thread reduced a handmade look to four levers:

1. Name physical materials, including felt, wool and wire armature.
2. Ask for 12fps judder rather than fluid motion.
3. Specify flaws such as fingerprints, seams and loose fibres.
4. Keep the set compact and light it with practical sources.

Its freeze experiments lock one subject while the crowd, water, or surroundings keep moving. The creature prompt in Magnific's creature test instead breaks a ten-second scene into three timed shots: shell crack, scientists' reaction, then the limb reveal with an alarm and whisper.

That format is concrete enough to contain camera direction, action, cut points and audio within a small number of seconds.

fal's model page advertises synchronized native audio at up to 15 seconds and 768p. The creator prompts write sound as part of the scene plan rather than adding it as a later instruction.

• An underwater octopus sequence combines a narrated line with bubble hiss, soft strings and a tracking shot in Magnific's octopus clip.
• A lab reveal pairs a ceramic crack, wet tearing, an alarm and the line "Don't move" in Magnific's creature test.
• The fashion piece calls for industrial bass, synchronized heels, shutter clicks and a blackout in Magnific's runway clip.
• A wingsuit POV moves from wind roar and breathing to a canopy crack, then birdsong, in Magnific's wingsuit clip.
• A miniature coastal-city timelapse assigns narration and piano across morning, dusk and night in Magnific's timelapse clip.
• A lighthouse monologue pairs a close voice with wind, foghorn and chair creak in Magnific's lighthouse clip.

ozansihay used H3 Max as a shot generator inside a conventional finishing workflow.

The creator generated four 15-second image-to-video clips with start and end frames, connected them in Premiere Pro with Morph Cut, then added color, halation and grain before a Topaz upres to 1080p at 60fps. The post says every clip was a first attempt.

A Magnific sequence took a more contained path: its start-and-end-frame workflow says the frames were set before generation and every scene change came from one render, with no editor opened. A Reddit music-video workflow stitched 14 15-second clips and proposed seven-second clips for higher-resolution output and faster generation.

Magnific gave Premium+ and Pro users three days of unlimited 480p output in its launch post and a follow-up. Its speed example framed 480p as a five-second clip made in under five seconds; an access explainer said 768p and 2K used credits at full speed. Magnific also credited fal with stronger prompt adherence and visual quality in its prompt-adherence post.

venturetwins' access note said both text-to-video and image-to-video were available, and fal maintains a separate image-to-video API. Magnific confirmed availability in a reply, while one plan post, another plan post and a third plan post identify Premium+ and Pro as the eligible tiers.

The current API only documents 480P and 768P as native resolutions, and does not explain the route behind Magnific's 2K upgrade. Separate from the subscription promotion, fal's sandbox page says signed-in users receive five free generations on a rolling 24-hour window, up to 15 seconds each at 768p with native audio.