fal Live. Never-ending shows.
AI television with an audience in the director’s chair.
fal.live (opens in a new tab) is fal’s experimental home for continuously generated shows. Its channel list includes Anime, Chaos, Sitcom, Soap Opera, and Wildlife World. Viewers can vote on what happens next, giving each stream a direction as it unfolds.
Behind it is H3 Max Director (opens in a new tab): a model that accepts new instructions during a continuous session. That opens up a different kind of entertainment—one where the next scene can respond to the room.
Open a channel and watch how its story changes. fal.live requires visitors to confirm they are 18 or older and accept its terms before entering.
Watch fal Live (opens in a new tab)See the fal Live entry screen
Meet the H3 Max family.
Fast enough to make trying another idea feel easy.
H3 Max is fal’s post-trained version of MiniMax H3, tuned for prompt following and visual quality. It generates video with synchronized audio. In fal’s September 8 benchmark, a five-second 768p clip took 2.46 seconds on H3 Max and 1.54 seconds on Turbo. Those are inference timings; your total wait also includes other processing and delivery.
| Model / workflow | Start here when you want to… | Official link |
|---|---|---|
| H3 Max | Make a clip from a prompt or animate a still. | Text to video (opens in a new tab)Image to video (opens in a new tab) |
| H3 Max Turbo | Explore more takes with a faster, lower-cost variant. | Try Turbo (opens in a new tab) |
| Reference to Video | Guide a new shot with reference assets. | Open model (opens in a new tab) |
| H3 Max Director | Steer a continuous video stream while it generates. | Start a session (opens in a new tab) |
| H3 Max Lip Sync | Turn an image and an audio file into a talking video. | Explore Lip Sync (opens in a new tab) |
| 3D to Video | Turn a rough Blender animation into finished-looking footage. | Explore the workflow (opens in a new tab) |
At 768p, fal lists H3 Max at $0.08/second and Turbo at $0.04/second—$0.40 and $0.20 for a five-second clip. The 75% launch discount ended September 14. The H3 Max landing page also advertises five free daily Sandbox generations for signed-in users. Check the chosen endpoint before generating; variants have different prices and limits.
Current H3 Max details (opens in a new tab)A five-second close-up of a glass teapot on a dark wooden table. Warm morning light catches the steam. The camera slowly pushes in as tea pours into a ceramic cup. Natural pouring sound and quiet room ambience. One continuous shot, no text, no music.
Our starter prompt is an idea to try, not a claim about a tested result. Standard H3 Max clips run 5–15 seconds at 480p, 768p, or 1080p; Director’s continuous sessions and Lip Sync’s options are separate.
fal Agent. Keep the idea together.
Less model hopping. More time shaping the work.
fal Agent (opens in a new tab) is a creative workspace that chooses models and carries a project across image, video, and 3D. Give it a brief, attach references, and refine the result in conversation.
fal describes project-scoped memory for references, creative decisions, and rejected takes, alongside stackable skills and tools for maintaining characters and styles. That makes it an interesting place to plan a campaign or build a series of related assets.
Help me build a launch campaign for a fictional tea brand called After Hours. Use deep plum, soft silver, and a quiet evening mood. First propose three concepts and a shot list. After I choose a direction, help me create a square hero image and a matching five-second video. Keep the packaging and lighting consistent. Show me the estimated cost before generating.
Agent uses fal credits, with pay-as-you-go and monthly credit plans shown on its page. Its programmatic API/CLI runs and Agent MCP integration are currently labeled “Coming soon.”
Explore fal Agent (opens in a new tab)And what about fal Worlds?
As of September 24, we couldn’t verify a standalone product named “fal Worlds” in fal’s main-site directory. The confirmed offering below is WMA, the World Model Accelerator. It is related world-model infrastructure; we haven’t established that “fal Worlds” is another name for it.
WMA (opens in a new tab) brings together fal’s inference engine, serverless GPUs, real-time WebRTC transport, and model distribution. It is aimed at developers building interactive world-model experiences. The landing page invites builders to request early access.
For a creative experiment you can open now, explore the example worlds in H3 Max Director (opens in a new tab). For the infrastructure behind a custom interactive experience, start with the WMA documentation (opens in a new tab).
Explore WMA (opens in a new tab)