Two H3 tiers output 2560×1440. One costs 4 credits a second, the other 10.
Both lanes return 2560x1440, but 1440P bills 4 credits per second and 2K bills 10. The premium does not buy pixels: it buys video-to-video, six aspect rati…
Read the guide →All guides
MiniMax H3
Everything we have written about MiniMax H3: what each tier actually outputs, what a second costs, the limits the model enforces, and the shape of the API that submits the job. Every number is measured on a real render or read from the live quote.
960P is the H3 tier nobody documents — and it runs 50% longer than standard 1080P for the same rate
960P outputs 1728x960 at 3 credits per second, the same rate as 1080P, with a fifteen-second ceiling instead of standard 1080P's ten.
Read the guide →Six style presets, one prompt: what drama staging and action staging actually change in MiniMax H3
minimax_h3_style_preset takes six values across two axes, staging (drama vs action) and treatment (live action vs animation, with an optional photoreal pas…
Read the guide →Conditioning a video model with audio: three reference clips, and where lip sync actually stops
H3 accepts up to three reference audio clips (MP3 or WAV, 15 MB each) on image-to-video lanes from 480P to 2K, and never on 2K text-to-video.
Read the guide →The commerce variant: product shots that stay on-model instead of drifting
The commerce style is image-to-video only, runs on 480P and 768P, and refuses any job without a reference image.
Read the guide →Where H3 money actually goes: five free references, then a per-image fee
The per-second tier rate, 10 to 100 credits for ten seconds, is only the first term. The bill also depends on reference images past the fifth (three credit…
Read the guide →A 30-second spot for about $1.50: draft at one credit a second, finish at three
Draft at 480P for one credit a second, review in bulk, then finish only the selects at 1080P for three.
Read the guide →First and last frame: how to make H3 land on an exact pose
A last frame is accepted on 480P, 768P and the 2K image-to-video lane, and not on 960P, 1080P or 1440P.
Read the guide →H3 from a script: the request that works, the response you actually get, and the error codes worth handling
Submit a job, poll it, download from content_url. The response is wrapped as {object: "video.job", job: {...}} with snake_case fields, credits_reserved is…
Read the guide →From shot list to episode: where H3 sits inside an AI film pipeline
The pipeline is plan, quote, approve, render, place dialogue, deliver, and only the fourth step is the video model.
Read the guide →The H3 video atlas is a dataset, not a gallery
The H3 video atlas is ten rendered masters, each with the exact request behind it, fifty prompts in total, measured file metadata and dated run logs, publi…
Read the guide →