H3 Max is the speed-and-cost video model for music-video shot iteration — 5–15 second clips, 480P or 720P, first-and-last-frame transitions for lyric cuts. In our measured run a 5-second 480P shot finished in 5.41 seconds. Pick it in SunoMV when you need a full-song preview today, not a 2K hero take.
H3 Max is a shot-level lever: duration, cost, turnaround. Hailuo 03 / MiniMax H3 is a series-level lever: native 2K, lip-sync, multi-reference lock. If you are boarding twenty lyric shots to see whether the video works, H3 Max is the default — our measured 5-second 480P clip came back in 5.41 seconds at $0.125. If you already know the shot and need a singing mouth in 2K, switch models. Do not ask this page to be both.
A music video is a stack of short clips. H3 Max is the model you reach for when you need twenty of those clips before lunch — not when you need one native-2K close-up with lip-sync.
The picker lists about 10 seconds of generation for a clip. In our own measured run on 27 August 2026 — text-to-video, 5 seconds, 480P, 16:9 — submit took 2.62s, remote work about 2.8s, end-to-end 5.41s. Twenty shots is minutes, not hours.
Quota as shown in the picker: 160 credits per 5-second shot at 720P, 100 credits per 5-second shot at 480P. Draft the whole song at 480P, then restage the chorus at 720P — you are not burning a hero-take budget on every try.
Text-to-video, image-to-video, and start-to-end frames are all in the picker. Lock the last frame of one lyric line as the first frame of the next so cuts hold together instead of jumping to a new face every bar.
Three MiniMax names sit next to each other. Mixing them up wastes a generation. This page is only the fast video model.
Hailuo 03 — also called MiniMax H3 — is the slower, heavier video model: native 2K and 4K, audio reference for lip-sync, multi-reference lock. H3 Max does none of that. If you need a 2K hero or a singing mouth, pick Hailuo 03, not this page.
Music 3.0 writes the song. H3 Max draws the picture. Same vendor prefix, different picker row. Generate audio on a music model, then cut the video on H3 Max — or paste a finished Suno link and storyboard the shots here.
No reference-to-video, no audio reference, no lip-sync, no native 2K or 4K. Duration is 5–15 seconds — not 3 or 4. Resolution is 480P or 720P; a 1080P request snaps down to 720P. Those limits are the point: cheaper, faster iteration.
The model is already in the video picker. You do not apply for a key, and you do not leave the music-video flow to use it.
Open the create flow, choose MiniMax H3 Max from the video models, set 5–15 seconds and 480P or 720P. That is the whole setup.
Describe the scene in text, start from a still, or pass a first frame and a last frame so the clip walks from one lyric picture to the next.
Drop each clip on the lyric timeline, preview vertical, square or widescreen, and export when the chorus finally cuts clean. Previewing the assembly is free.
Nothing to install. The model is already in the SunoMV video picker.
Open create, choose MiniMax H3 Max, set 5–15 seconds and 480P (draft) or 720P (keepers). Ignore 1080P — it snaps down to 720P.
Write the scene, start from a still, or set a first frame and a last frame so two lyric pictures connect. Skip lip-sync and 2K here — that is Hailuo 03.
Lay the clips on the lyric timeline. Twenty 5-second shots generate in minutes, not hours. Preview free, then export vertical, square or widescreen.
No video model is universally better — only better fitted. These are the jobs this row is for.
Twenty 5-second lyric shots at 480P is a cheap storyboard of the whole track. Confirm the cut, then restage two or three keepers at 720P — or move those keepers to Hailuo 03.
First-and-last-frame is how you stop the singer's outfit from changing every line. Pin the end of shot A as the start of shot B.
100 credits per 5 seconds at 480P is the iteration price. Burn the bad ideas here so the expensive model only sees shots you already like.
Same vendor prefix. Different picker rows. Click through instead of guessing.
Video. Fast 5–15s shots at 480P/720P, first-and-last-frame, iteration quota.
Video. Native 2K/4K, lip-sync, multi-reference. Slower, higher quota. Not this page.
Audio. Writes the song. Then you still need a video model for the pictures.
MiniMax H3 Max is a text-to-video, image-to-video and first-last-frame video model. Clips last 5–15 seconds at 480P or 720P. It is selected in SunoMV as h3-max and is used to iterate short music-video shots rather than to output native 2K or lip-synced singing.
MiniMax H3, also shipped as Hailuo 03, is a separate video model with native 2K/4K, audio-reference lip-sync and multi-image reference. MiniMax H3 Max drops those capabilities in exchange for shorter turnaround and a lower per-shot quota at 480P and 720P. They are not aliases for each other.
A first-and-last-frame shot is a generated clip whose opening picture and closing picture are supplied by the user. The model interpolates the motion in between, which is how consecutive lyric lines keep the same character, wardrobe and camera side across a cut.
Capability limits on this page match the SunoMV model picker. Timing and API cost are our own measured run on 27 August 2026, not a vendor marketing number.
MiniMax H3 Max is offered as text-to-video and image-to-video, including a start-and-end-frame path, as the selectable row named MiniMax H3 Max in the SunoMV video picker.
SunoMV — video model picker (h3-max) ↗Our measured run, 27 August 2026: text-to-video, 5 seconds, 480P, 16:9. Submit 2.62s; remote about 2.8s; end-to-end 5.41s; API cost $0.125 ($0.025/s × 5s, launch rate). Sample clip stored from that run.
SunoMV — measured run 2026-08-27 ↗Public announcement of MiniMax H3 Max as a fast 5-second-class video model, distinct from the Hailuo 03 / MiniMax H3 line.
MiniMax H3 Max — public announcement ↗FAQ'S
Ask us anything!
5–15 seconds, 480P or 720P, first-and-last-frame when the cut has to hold. Preview the full-song assembly free, then export for TikTok, Reels, Shorts and YouTube.
/icon.png){kind=icon}
