H3 Max is the speed-optimized build of MiniMax H3, the open-weight multimodal video model. It turns a prompt or an image into a 5-second 768p clip in roughly three seconds — faster than the video plays. h3max.info collects the official links, APIs, open weights, guides and live projects so you can start building in minutes.
Three names show up everywhere and they are not the same thing. Here is the short version.
MiniMax's next-generation general video model. Takes text, images, video and audio in one request and returns 5–15 seconds of video at 24 fps with native stereo audio. Announced July 31, 2026 (API), weights published August 3 under the MiniMax H3 Community License.
Jointly released by MiniMax and fal.ai on August 29, 2026. fal post-trained MiniMax H3 for prompt adherence and visual quality and co-designed the inference stack for speed: a 5-second 768p clip in about three seconds, dozens of times the throughput of the official endpoint.
A further-accelerated variant of H3 Max served at minimax/h3-max-turbo. Same 480p/768p outputs and the lowest per-second price in the family. fal has not published architectural details on how it differs from H3 Max.
| MiniMax H3 | H3 Max | H3 Max Turbo | |
|---|---|---|---|
| Maintainer | MiniMax | fal.ai (post-trained on H3) | fal.ai |
| Max resolution | 2K (API) · 768p (open weights) | 768p | 768p |
| Speed | Baseline | ≈3 s for a 5-second clip | Faster than H3 Max |
| Endpoints | T2V, I2V, reference, editing | T2V, I2V, reference-to-video | T2V, I2V |
| Open weights | Yes (H3-Base) | No | No |
| Best for | Max quality, editing, self-hosting | Fast production & interactive apps | Volume and cost |
Figures are compiled from MiniMax and fal.ai announcements as of September 2026 and may change. Corrections welcome — see Submit.
Hand-picked links, grouped by what you are trying to do. Every entry is checked by a human before it goes in.
The launch post: architecture, capabilities, benchmarks and the case for one model across generation, referencing and editing.
minimax.io Open sourceThe open-weight release announcement, license summary and links to the checkpoints.
minimax.io API docsOfficial developer docs covering the H3 and H3 Max endpoints, parameters, input limits and examples.
platform.minimax.io HubMiniMax's own landing page for the open model with tutorials, deployment notes and workflow examples.
design.minimax.io UpdatesRelease notes, demos and pricing changes usually land here first.
x.comThe flagship endpoint. Prompt in, 5–15 second 480p/768p clip with audio out, in seconds. Playground plus REST, Python and JS clients.
fal.ai fal.aiAnimate a still image with a text prompt. Good for product shots, characters and storyboards.
fal.ai fal.aiKeep a subject consistent across shots by passing reference images alongside the prompt.
fal.ai fal.ai · TurboThe fastest and cheapest tier in the family. Ideal for batch jobs and live, chat-driven video streams.
fal.ai fal.ai · TurboTurbo speed for image-conditioned generation.
fal.ai Explainerfal's write-up on what the model does, how H3 Max differs and how to call it.
fal.aiThe official organization page hosting the H3-Base checkpoints (FL2VA and Ref2VA) and model cards.
huggingface.co CodeInference code, reference implementations and issue trackers for MiniMax's open models.
github.com GuideGPU requirements, VRAM math and a step-by-step cloud deployment of the open checkpoints.
runpod.io GuideAnother deployment walkthrough with cost estimates for different GPU classes.
spheron.network Community postA community explainer on the Hugging Face blog covering the release, the license and how it compares to Hailuo 2.
huggingface.coResolutions, durations, frame rates, supported inputs and prompt length limits in one table.
morphic.com GuideA practical overview for developers: which endpoint to pick, license terms and integration tips.
kylon.iofal's own always-on interactive stream: viewers type prompts and H3 Max renders the next scene in real time.
fal.live LivePieter Levels' chat-driven infinite AI video channel, built on H3 Max in an afternoon and viewed millions of times on X.
levels.io IntegrationOne of the first commercial integrations: H3 Max video generation inside ElevenLabs' creative video tooling.
elevenlabs.iofal.ai bills per second of generated video on a prepaid credit balance. A launch promotion runs until September 7, 2026; standard rates apply after that.
| Endpoint | Resolution | Promo (until Sep 7, 2026) | Standard | 15-second clip at standard rate |
|---|---|---|---|---|
| H3 Max Turbo | 480p | $0.00625 / s | $0.025 / s | $0.38 |
| H3 Max Turbo | 768p | $0.01 / s | $0.04 / s | $0.60 |
| H3 Max | 480p | $0.0125 / s | $0.05 / s | $0.75 |
| H3 Max | 768p | $0.02 / s | $0.08 / s | $1.20 |
Prices as listed on fal.ai in early September 2026. Always confirm on the model page before committing a budget. fal enables auto top-up by default — turn it off in Billing if you want a hard cap.
This is version one: a curated directory. Here is what we are building next.
Tools, tutorials, integrations, live channels, papers — if it is about H3 Max or MiniMax H3, we want it listed. Submissions are reviewed by hand.