Text to video · Image to video · Native audio
Create a 5–15 second AI video from a prompt or reference image with MiniMax H3 Max. fal Research post-trained the model for stronger prompt following and visual quality, then optimized it for fast 480P and 768P generation with synchronized dialogue, ambience and sound effects in the same result.Create a 5–15 second AI video from a prompt or reference image with MiniMax H3 Max.
Describe the video you want, or upload an image to guide the subject and composition. You can also add video and audio references. Choose the length, resolution and format; the exact credit cost appears before you generate.
Sample result gallery
These are our own five-second, 768P MiniMax H3 Max generations. The first four animate a reference image; the next seven start from text. Each clip comes directly from the same endpoints used by the generator above, with no editing, upscaling or separate audio pass. Hover to play and use the speaker button for sound.
The reference locks the model, silver pleating, travertine set and editorial light while H3 Max adds camera movement and physically weighted fabric motion.
Source: Our H3 Max image-to-video run · 768P · 5 seconds · no post-production
The bottle facets and black-sand composition stay fixed while a restrained macro arc, moving highlight and wind-driven grains create the commercial motion.
Source: Our H3 Max image-to-video run · 768P · 5 seconds · no post-production
The room geometry and furniture remain intact as the camera advances, rain moves across the glass and the warm fireplace breathes against the cool forest.
Source: Our H3 Max image-to-video run · 768P · 5 seconds · no post-production
The folded fox, layered hills and visible paper fibers carry from the illustration into tactile stop-motion movement and a small narrative beat.
Source: Our H3 Max image-to-video run · 768P · 5 seconds · no post-production
A continuous low tracking shot has to land three ordered beats: arrival, four simultaneous wheel changes, then launch through tyre smoke.
Source: Our H3 Max render · 768P · 5 seconds · no post-production
The vertical phone-style take holds the face, spoken line, wind noise and launch action in one coherent piece of UGC footage.
Source: Our H3 Max render · 768P · 5 seconds · no post-production
A ruined arcology rises through ash, the picture falls to black, and the requested title lands on the final impact.
Source: Our H3 Max render · 768P · 5 seconds · no post-production
A square commercial macro shot keeps the bottle sharp while the splash, rim light, camera orbit and impact sound happen together.
Source: Our H3 Max render · 768P · 5 seconds · no post-production
The courier runs, jumps a robot, ducks a drone and collects a parcel while the parallax, chiptune and game effects stay in sync.
Source: Our H3 Max render · 768P · 5 seconds · no post-production
Visible fingerprints, handmade set, stop-motion cadence and comic audio all survive a complete action with a beginning and payoff.
Source: Our H3 Max render · 768P · 5 seconds · no post-production
The waitress delivers a lip-synced line, hears something outside and turns toward the window while the push-in and room tone continue.
Source: Our H3 Max render · 768P · 5 seconds · no post-production
Features
fal reports about three seconds of model inference for a five-second 768P clip. That timing covers the model run itself; upload, queue and file-delivery time can make the full request take longer.
fal added new training data and reinforcement-learning work to improve ordered actions, requested visual styles and specific camera direction.
Dialogue, ambience, foley and music are created alongside the picture instead of added in a separate synchronization pass.
Start with a text prompt, animate an image or add video and audio references for motion, identity and sound. Refer to each file in the prompt as Image 1, Video 1 or Audio 1.
Choose 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16 for cinematic shots, social feeds and vertical video without a separate reframing step.
The result includes backend inference timing, while Soku keeps the request saved through upload, queueing and delivery so you can leave the page and return later.
Name the subject, action, camera movement and lighting. Put events in order when timing matters, and describe dialogue, ambience or sound effects in the same prompt.
Upload the product, character, composition, movement or voice you want to preserve. Leave references empty when you want the model to invent the shot from text.
Select 5–15 seconds, 480P or 768P, and an aspect ratio. Review the credit charge before submitting, then open the finished video in Creatives to edit or resize it.
Why Soku
Soku turns MiniMax H3 Max into more than a model demo. Sign in, pay, generate and move the result into your creative workflow without downloading and rebuilding every take by hand.
Sign-in, credit check, payment and generation all happen here, so you do not have to move to a disconnected model playground.
Open a finished clip on the canvas to trim, caption, resize or create another variation without starting over.
Generate several hooks, compare the outputs and keep developing the direction that deserves the next pass.
Use cases
Performance creative teams
Explore openings, camera moves and visual styles before approval, then spend editing time only on the strongest take.
Product and growth teams
Fast model inference makes live creative tools, rapid previews and high-volume generation workflows more practical.
Social teams
Generate each placement in its native shape, with dialogue and ambience already synchronized, instead of stretching one master into every feed.
Specifications
These are the controls the live endpoint accepts today; the studio is generated from the same server-owned specification.
Alternatives
They share a foundation but are different production choices. Pick the one whose ceiling matches the job.
Best for
Resolution ceiling
5s 768P inference
Native audio
Video editing
768P fal list price
Use H3 Max when latency changes the workflow. Use standard H3 when resolution or instruction-based editing changes the deliverable.
Pricing
New users can create an account without leaving the page. Signed-in users see their live credit balance and the exact cost of the selected length and resolution before the job is submitted.
Billed per generated second; 480P and 768P are priced separately
$0/ month
No card required
$19/ month
7,000 credits / month
$49/ month
20,000 credits / month
If the balance is too low, the page opens the plan and top-up flow. A provider-side generation failure refunds the committed credits. See all plans and the credit calculator.
Write a prompt or upload a reference, choose your format and generate with native audio. Sign-in, credits and the handoff into Creatives all happen here.
Start with my asset
Soku AI
