What is fal H3 Max?
fal H3 Max is a family of MiniMax video-generation endpoints hosted by fal. It can generate native audiovisual output from text or an image. The important detail is that “H3 Max” now describes two different workflows:
- Standard text-to-video and image-to-video endpoints return a finished clip.
- H3 Max Director opens a continuous WebRTC stream and accepts new prompts while that stream is running.
That distinction matters. A fast clip is useful for iteration, but it is not a live session. Director changes the interaction from “submit and wait” to “watch and steer.”
H3 Max Preview versus Director
| Capability | Standard H3 Max | H3 Max Director |
|---|---|---|
| Output | Finite clip | Continuous video and audio stream |
| Transport | Queued HTTP generation | Negotiated WebRTC session |
| New prompts during output | No | Yes |
| Continuity | Can continue from an image/final frame | Maintains recent session context natively |
| AIR Live mode | 5s Preview | Live Director |
AIR Live keeps these modes separate. Preview costs 4 credits for a five-second 480p shot. Live Director costs 72 credits for a 60-second session. It never labels a sequence of preview clips as a true live stream.
What makes Director different?
The official fal API page
describes continuous realtime generation with live prompt updates that preserve
characters, settings, and story continuity. Its public AsyncAPI contract uses a
WebRTC server and defines control messages such as configure, prompt, and
stop, along with provider events for prompt state, chunks, metrics, and
errors.
This makes several products possible that ordinary clip generation cannot deliver cleanly:
- a creator verbally directing the next scene;
- an audience voting on what should happen next;
- an ambient world that changes with live data;
- an interactive story or game broadcast;
- a persistent generated camera for an AI agent.
Does it generate audio?
Yes. Director streams video and audio together. That is different from adding a voiceover after generation. Audio can remain part of the evolving scene, though model output is probabilistic and continuity should be tested for each use case.
What does it cost?
When checked on September 4, 2026, fal advertised a two-week launch rate of $0.02 per generated video second and a standard rate of $0.08 per second for Director. fal also applies a 60-second minimum session charge. That implies a provider cost floor of $1.20 during the promotion and $4.80 at standard pricing, before speech, storage, payment fees, and application margin.
Always verify current pricing on the official model page before building unit economics around these numbers.
Try H3 Max without learning the API
Open the AIR Live director to test a low-cost Preview or direct a continuous session by voice. AIR Live is an independent product and is not affiliated with fal or MiniMax.