WORKS WITH MINIMAXWorks with
Running series on MiniMax H3
On your own GPU or through fal.ai, rendered from the same shot list.
WORKS WITH01—IN KONTEEngines in Konte
Engines that run MiniMax
Named as the app's settings screen names them. Whichever you pick, the shot list stays as it is.
- Local GPU
MiniMax H3 on a local GPU
- Resolution
- Rendered at 480P, brought up to 1080p
- Time
- About 139 s for a 4-second shot (measured), one at a time
- References it takes
- Reference images of the cast, sets and props, plus voice actors' samples
- Usage fees
- No API fees
- Cloud
MiniMax H3 via the fal.ai API
- Length per shot
- 5–15 s
- Resolution
- 768P by default (no quantization)
- References it takes
- One start-frame image
- Usage fees
- Billed to the fal.ai key you registered
- Cloud
MiniMax Hailuo via the fal.ai API
- Length per shot
- 6 or 10 s
- References it takes
- One start-frame image
- Usage fees
- Billed to the fal.ai key you registered
02—WHAT KONTE ADDSThrough Konte
What MiniMax alone doesn't keep
Faces, lines, voices, captions, the history of every retake: the ledger for shooting on.
Shooting with the model alone
Shooting through Konte
- Face and wardrobe
- Shooting with the model alone
Re-attach the reference image every time
- Shooting through Konte
Pick the cast and their reference images go into the shot
- Lines
- Shooting with the model alone
Write who says what into every prompt
- Shooting through Konte
Choose the speaker on the shot list and it's written into the prompt
- Voice
- Shooting with the model alone
The voice tends to change from shot to shot
- Shooting through Konte
Pass the voice actor's sample so each speaker keeps their voice (MiniMax H3 on your own GPU)
- Captions
- Shooting with the model alone
Have the model draw it and the letters tend to break
- Shooting through Konte
Burned in at finishing, so the letters never break
- Retakes
- Shooting with the model alone
Keep track yourself of which prompt made which clip
- Shooting through Konte
Each shot keeps its history, and you can go back to an earlier take
- Assembly
- Shooting with the model alone
Line the clips up one by one in an editor
- Shooting through Konte
Joined into one video in shot-list order, with a subtitle file
03—FITGood fit, poor fit
What it's good at, and what to watch
GOOD AT
- Generates picture and sound together
- On your own GPU, retake as often as you like with no API fees
- Takes the cast's reference images and the voice actor's sample in the same shot (on your own GPU)
WATCH OUT FOR
- On your own GPU, shots render one at a time: about 139 s for 4 seconds
- Through fal.ai, references are limited to one start-frame image
- Text inside the picture tends to break, so captions are burned in at finishing
04—PROOFShot with this model
MADE WITH KONTE—CASTING
Same direction, only the cast swapped
Four scenes. In each, the prompt is identical word for word; the only change is the cast.
Eye level · Static · 3 s
NANAMIWoman, 30s
MIOWoman, 20s
AOIWoman, 20s
HINAWoman, 20s
REIWoman, 50s
YUMAMan, 30s
6 cast members × 4 scenes · the 6 clips of a scene share the same shot direction and camera · MiniMax H3 on a local GPU
MADE WITH KONTE—SET & LOCATION
Same cast, a morning and a night version
A 15-second ad, same shots. Only the place and time of day change, for versions A and B.
Version A: Morning kitchen14.8S
Version B: Living room at night15.2S
Fictional skincare ad · Cast: NANAMI · 3 s × 5 shots · Version A shot 03 trimmed by 0.4 s at the end · MiniMax H3 on a local GPU
05—ALSO WORKS WITHOther models
Same shot list, other engines
GET STARTEDStart here
Your film studio, starting today
Start by creating a single cast member. We're happy to talk it through.