
These are both fal models, not MiniMax ones. fal post-trained MiniMax's open-weight H3 into H3 Max, then distilled H3 Max into H3 Max Turbo. Same weights underneath, so the same 768p ceiling, the same native stereo audio and the same 5 to 15 second range. That leaves two real questions, what the distillation costs and what the speed buys, so we ran the same six prompts through both and timed every one.
H3 Max vs H3 Max Turbo on the same prompt
Six prompts, each testing something different, each run through both models at 768p and 16:9 with prompt expansion disabled. The prompt is printed under each pair so you can run it yourself. Every clip carries its own audio, so unmute one to hear it.
The style reel: range in one take
Fifteen seconds, twelve shots, a hard cut roughly every 1.25 seconds, each shot a different subject, scale, palette and technique. This is the one that shows the most in the least time.
H3 Max
H3 Max Turbo
Both hold twelve legible shots and both render the word in shot eleven correctly. H3 Max resolves more fine detail inside each shot, most visibly in the match flame and the watch bezel; Turbo softens those but keeps the same look and the same cutting rhythm.
A rapid-cut style reel: 12 distinct shots, each held about 1.25 seconds, hard cuts between them, no dissolves, no captions except where stated. Every shot changes subject, scale, palette and technique so the reel reads as a range demo. 1) Extreme macro, a match head strikes and blooms into sulfur flame, warm orange against near black. 2) Aerial top-down, cracked white salt flats in hexagon plates, one small red truck crossing frame. 3) Hand-drawn anime cel, a girl turns and her hair whips across frame, flat colour and clean linework. 4) Underwater, sunlight shafts through kelp as a shoal of fish turns at once like a silver sheet. 5) Stop-motion claymation, a thumbprinted clay bird tilts its head, visible frame judder. 6) Studio product shot, a chrome watch case rotates slowly, hard key light raking the bevel. 7) Handheld street documentary, rain on a night market awning, steam rolling off a griddle. 8) Isometric 3D, a paper city folds up out of a flat sheet, towers extruding. 9) Close portrait, an old woman lit by one window, skin and eye detail, one slow blink. 10) Two-colour risograph print, a cyclist mid-pedal, coarse dot texture and deliberate misregistration. 11) Motion graphics on black, a white scribble resolves into the clean word SIGNAL. 12) Wide landscape, a storm front over a wheat field, wind combing the crop, last light. Lighting must be motivated in every shot, no neon.
Prompt adherence: countable, placeable detail
The brief names exactly three tools in a row, a blue enamel mug behind the wheel, and slip on the forearms. This is what fal's post-training targeted, so it is where the two should separate.
H3 Max
H3 Max Turbo
Both place the mug and the sponge. H3 Max lays the three tools out as a readable row and keeps focus on the hands; Turbo scatters them and drifts wider off the hands.
One continuous take, no cuts. A ceramicist's hands centre a grey clay cylinder on a spinning wheel. Slow push-in from three-quarter left. Late afternoon window light from camera right, warm, with a soft bounce filling the shadow side. Exactly three tools rest on the bench in a row, left to right: a wooden rib, a wire cutter, a small sponge. A blue enamel mug sits on the bench behind the wheel, half full. Water beads run down the clay as it turns. The potter's sleeves are rolled to the elbow and the forearms are flecked with dried slip. Shallow depth of field, focus stays on the hands throughout, the bench behind falls soft.
Human performance: a face and a beat
The brief asks for a specific piece of acting: she finishes the last line, stops, swallows once, then lifts her eyes past the lens.
H3 Max
H3 Max Turbo
H3 Max completes the beat and lands the look up. Turbo keeps her reading and only half lifts, so the shot ends before the moment does. Hands and fingers are correct in both.
One continuous take, no cuts, no dialogue. Close on a woman in her thirties reading a handwritten letter by a window at dusk. She reaches the last line, her eyes stop moving, she swallows once, then lifts her gaze just past the lens. Light is from the window only, cool and soft, falling off quickly toward the right of frame. Handheld with a very slight breathing drift. Skin texture, fine hair and the moisture on her eyes are visible. Both hands hold the paper steady at the lower edge of frame, fingers relaxed and correctly formed. Focus holds on her eyes throughout.
Physics: viscosity and surface tension
Backlit honey off a dipper into a half-full jar, coiling on the surface before it sinks.
H3 Max
H3 Max Turbo
Both read as real liquid rather than blur. H3 Max keeps the stream unbroken and the caustics on the slab; Turbo adds a jar lid nobody asked for.
One continuous take, macro, no cuts. Backlit honey pours from a wooden dipper into a clear glass jar already half full. The stream is thin and unbroken at first, then coils onto the surface and folds over itself before sinking. Late sun rakes through from behind, so the honey glows amber and throws a warm caustic onto the marble slab under the jar. A single drip clings to the dipper, stretches, and falls. 120fps slow motion feel. Surface tension, viscosity and the way the coil settles must read as real liquid, not as a soft blur.
Typography: legible in-frame text
The hardest thing on this page. The sign has to read NORTHFIELD BAKERY on two lines with EST 1954 beneath, in gold leaf, while reflections move across the glass.
H3 Max
H3 Max Turbo
Both spell it correctly, hold the two-line layout, and keep the letters stable under moving reflections. This is the pair where the distillation costs least: H3 Max is a little crisper on the drop shadow, and that is the whole difference.
One continuous take, no cuts. A sign painter finishes a shop window in gold leaf and black outline. The camera holds square to the glass as the brush completes the final letter. The finished sign reads exactly: NORTHFIELD BAKERY, on two lines, NORTHFIELD above BAKERY, in a classic serif with a thin drop shadow. Below it, smaller, in plain capitals: EST 1954. Warm afternoon light from the street side, reflections of passing traffic drifting across the glass. Every letter must be correctly spelled, evenly spaced and stay stable, with no warping or flicker as the reflections move.
Camera and 3D coherence
A twelve-storey fall down the middle of a spiral library stairwell, settling above a compass rose. Parallax has to hold or the railings bend.
H3 Max
H3 Max Turbo
Both hold the fall and land on the compass rose without the geometry sliding. H3 Max carries more shelf and book detail on the way down.
One continuous take, no cuts. The camera falls straight down the open centre of a spiral library stairwell, twelve storeys deep, accelerating. Brass railings spiral past, each floor lined with dark oak shelves and small green reading lamps that streak into rings as the fall speeds up. A glass dome high above throws hard shafts of light down the shaft, and dust turns slowly in them. At the bottom the fall decelerates and settles a few feet above a black marble floor inlaid with a pale compass rose. Parallax must hold true all the way down, with the railings and floors staying rigid and correctly spaced rather than sliding or bending.
H3 Max vs H3 Max Turbo specs, side by side
Measured on the runs above; specs from fal.
| Feature | H3 Max | H3 Max Turbo |
|---|---|---|
| Made by | fal, post-trained from MiniMax's open-weight H3 | fal, distilled from H3 Max |
| Released | August 2026 | September 2026 |
| 15s clip, measured | 44.9s | 19.1s |
| Relative speed, measured | Baseline | 2.3x faster |
| Quality, on fal's evaluations | The higher tier | 97th percentile of H3 Max |
| Maximum resolution | 768p | 768p |
| Output, measured | 1344 x 768, 24fps | 1344 x 768, 24fps |
| Clip length | 5 to 15s | 5 to 15s |
| Audio, measured | AAC stereo, 32 kHz | AAC stereo, 32 kHz |
| Text and image to video | Yes, including a first and last frame | Yes, including a first and last frame |
| Reference to video | Up to 12 files | The one mode Turbo drops |
| Detail and instruction following | Holds fine texture, completes the beat | Softens detail, sometimes stops short |
| Best for | Maximum detail in the frame | The same look, roughly twice as fast |