CONCEPT FRAME · 16:9
CINEMATIC PURSUIT · 5S · 768P
Make speed readable in one continuous move.
Subject, tracking camera, physical spray, engine sound, and the final wide frame each have a separate job.
POST-TRAINED BY FAL RESEARCH
FAL H3 MAX
Video generation caught up with playback.
fal reports a 5-second 768P video in under 3 seconds, with synchronized audio and stronger prompt adherence.
Watch live AI streamsOfficial fal sample streamed from fal.media · not generated or owned by this site
01 / THE SPEED ADVANTAGE
The render finishes before the clip does.
fal reports that FAL H3 Max generates a 5-second 768P video in under 3 seconds of wall time. That makes generation faster than playback—not just faster than the base model.
ILLUSTRATED WALL TIME
00:00.0
H3 Max render< 3 seconds
READY
Finished clip5 seconds
Timing illustration based on fal's published under-3-second result. This is not a live benchmark or generation.
35×
the throughput of the official MiniMax H3 endpoint in fal's comparison
15×
faster on average than models fal classified as comparable in quality
01
co-designed model and inference stack instead of optimizing serving after training
These FAL H3 Max speed figures are fal's published evaluation claims, checked Sep 4, 2026—not measurements produced by this site.
02 / FAL H3 MAX PROMPT BUILDER
One idea.
Seven decisions.
FAL H3 Max can read a detailed brief. The studio separates the parts so you can inspect the subject, motion, camera, light, sound, and final beat before copying anything to fal.
Load a direction
03 / PROMPT RECIPES
Three directions.
No mystery prompt.
Each FAL H3 Max concept frame is paired with the exact creative direction loaded by the button. They are planning references created for this site, not model-output claims.
CONCEPT FRAME · 16:9
CONCEPT FRAME · MACRO
PRODUCT MACRO · 5S · 768P
Let one light move carry the whole shot.
A controlled action, small sound cues, and a precise ending keep the brief focused on material detail.
CONCEPT FRAME · 9:16
VERTICAL REVEAL · 10S · 768P
Design the action for the phone frame.
The subject stays central while the coat, rain, bus, and camera arc create motion at different depths.
04 / MODEL ENVELOPE
Know the frame before you write it.
FAL H3 Max is fal Research's post-trained variant of the open-weight MiniMax H3 base. The official guide also uses the name MiniMax H3 Max. The live endpoints define a compact working envelope for short video.
Duration5–15seconds per request
Resolution480 / 768native output tiers
Text frame06supported aspect ratios
SoundA/Vgenerated in the same workflow
Text to video exposes 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Image to video follows the supplied starting image instead. Model limits were checked against fal's live endpoint schemas on Sep 3, 2026.
OPEN DIRECTORY ↗
H3 MAX IN THE WILD
See where H3 Max became a live medium.
Watch AI Streams is an independent companion directory for finding and comparing live experiments built around the H3 family.
PUBLIC CHAOSInfinite Slop
CHANNEL NETWORKfal.live
AI REALITY SHOWRenoise Live
Compare how each project handles audience control, recurring worlds, model evidence and broadcast status, then open the original experience.
Browse Watch AI Streams Independent directory · website access and broadcast activity verified separately · reviewed Sep 4, 202605 / WORKFLOW
Write. Inspect. Hand off.
-
01
Build the visible event.
Start with one recognizable subject and one action that can be seen. Add place and camera direction only after the event is clear.
-
02
Check the missing direction.
The five-part check flags absent subject, action, scene or camera, sound, and ending. It measures completeness, not predicted quality.
-
03
Copy to the official route.
Copy the prompt or JSON, verify the live fal price, and run it on the matching text-to-video or image-to-video endpoint.
REVISION RULE
Change one instruction per pass.
After a FAL H3 Max result, compare the clip with the brief before rewriting everything. Keep the subject and successful direction stable, then change the camera, action, sound, or final beat that missed. A repeated seed gives you a reference point, but prompt expansion and model variation mean it does not promise an identical frame. Recording the prompt, settings, seed, and one changed instruction makes the next result easier to judge.
06 / CHOOSE THE MODEL
FAL H3 Max or standard MiniMax H3?
Pick by the delivery requirement, not by a leaderboard headline.
Decision
FAL H3 Max
MiniMax H3
Best fit
Fast iteration and adherence-focused 480P or 768P output
Higher-resolution output and instruction-based video editing
Single clip
5–15 seconds at 24 FPS
5–15 seconds at 24 FPS
Resolution
480P or 768P
480P, 768P, plus provider-listed upscale tiers
Input routes
Text, start/end frames, and reference media endpoints
Text, images, references, and video editing workflows
Source boundary: model facts are checked against fal's live endpoint schemas and fal's Sep 1, 2026 guide. Benchmarks and pricing can change; this page does not label either model “best.”
07 / COST CHECK
Price the clip before the click.
The planner uses fal's published standard list rate, not a temporary promotion. Your selected duration and resolution update the estimate inside the studio.
Return to the cost planner480P$0.05per generated second
768P$0.08per generated second
Checked Sep 1, 2026 · verify the live endpoint before a paid request
08 / SOURCE BOUNDARY
The prompt stays here.
The generation stays on fal.
Nothing is uploaded. Your draft stays in this browser. H3 Shot Lab does not collect a fal API key, sell credits, or submit a paid request. When the brief is ready, the official endpoint remains the final source for availability, pricing, and generation behavior.
09 / FAQ
Before you render.
What is FAL H3 Max?
FAL H3 Max is fal Research's post-trained version of the open-weight MiniMax H3 video model. It is available through fal for text-to-video, image-to-video, and reference-to-video workflows. The “Max” name identifies fal's post-trained variant; it is not a separate MiniMax foundation model.
Does this site generate FAL H3 Max videos?
No. H3 Shot Lab prepares a prompt and request payload without collecting an API key. Copy the result, open the official fal endpoint, review its current price, and submit there. That boundary keeps a paid model call out of an unaffiliated planning tool.
What should an H3 Max prompt include?
Start with one recognizable subject and one visible action. Add the scene, camera move, light, physical sound sources, and a final state. For dialogue, quote the line and describe how it is delivered. Ordered beats are usually easier to inspect than a long list of style words.
How accurate is the cost estimate?
The calculator multiplies duration by fal's published standard list rate: $0.05 per generated second at 480P or $0.08 at 768P, checked Sep 1, 2026. fal may run promotions or change pricing, so the live endpoint is the final source before any paid request.
Are the gallery images H3 Max outputs?
No. They are concept frames created for this independent site to make the three prompt directions easier to scan. They are not performance evidence, generated-video samples, or claims about what a specific FAL H3 Max request will return.
Is H3 Shot Lab affiliated with fal or MiniMax?
No. This is an independent educational and planning site. fal, MiniMax, H3, and H3 Max are names used only to identify the relevant model and endpoints.
ONE SHOT · CLEAR INTENT