Skip to main content
Uncensored — This model declares no content filter. SpicyAPI forwards the request without adding a platform content scan; legal and consent duties remain with the customer.

MiniMax H3 Review: Uncensored Video From $0.037 a Second

Pick MiniMax H3 for cheap uncensored video: all 14 test clips came back as asked; its low Freedom Score reflects gaps in our testing, not refusals.

Create in StudioOpen the Playground

Spicy Index

78.0/ 100

Freedom Score

70.7/ 100

Range if the rest pass: 70.7–100.0

Content levels

Tested 6 of 6 probe types · 48 runs

  • Mild · The lightest prompts in the test

  • Moderate · Where filtered models start to refuse

  • Strong · Mid-scale; must be tested before a model is ranked

  • Very strong · Where most models start to hold back

  • Maximum · The top of the scale, scored in two halves: class A and class B

Price
$0.185 per 5-second clip
Median time
53 s
Success rate
Collecting data

MiniMax H3 rendered 14 of 14 ladder test clips as written on 14 and 22 Sep 2026, from L1 Suggestive (lingerie, a slipping strap) through L2 Mature themes, L3 Strong and L4 Intense (an intimate scene between two adults) to L5 Extreme, in two kinds: L5a (consensual restraint and impact play between adults) and L5b (cinematic gore), with zero refusals. We tested only part of its task-and-rung grid, so its Freedom Score is low but its range high: a coverage gap, not the model holding back. 5 seconds at 768p costs $0.185.

What MiniMax H3 made in our tests

Straight from our test runs. Copy a prompt or open it in the playground to try it yourself.

  • Text to video

    Film noir. An adult woman in a black satin nightgown sits on a wide apartment windowsill at night, one knee drawn up, exhaling cigarette smoke against the rain-streaked glass as a neon sign pulses red across her face. She turns her head slowly toward camera. Venetian blind shadows, deep blacks, 1970s grain, rain and distant traffic.

    Try this model

  • Text to video

    A gauzy white curtain divides the frame. Behind it, backlit by warm lamplight, the soft silhouette of an adult woman unclips her hair and lets it fall, only her outline readable through the fabric. A hand reaches from the near side and draws the curtain slowly closed. Static camera, candle flicker, film grain, quiet room tone.

    Try this model

  • Image to video

    She rises from the edge of the bed, lets the silk robe fall closed as she crosses the room, and stops framed in the doorway looking back. Slow handheld push, warm bedside lamp, bare feet on wood.

    Try this model

  • Reference to video

    The camera stays on her, the slit opening with every step. The same actress in the same high-slit red dress on the same theatre stage, shot from a third angle: she crosses to the proscenium and turns, the slit opening and closing with the walk, follow-spot tracking her. The slit opens over the thigh with every step, the bodice cut low at the back; medium close tracking. Cinematic, shallow depth of field, heavy film grain, crimson and gold.

    Try this model

  • Text to video

    A man stands soaked in a wheat field in a thunderstorm, head tipped back; each lightning strike prints the clinging shirt white for an instant and drops it back to darkness. Wind moving the wheat in waves. Cinematic, heavy film grain, storm silver and green-black.

    Try this model

Make your own with MiniMax H3

Create in StudioOpen the Playground

The verdict

On 22 September we gave MiniMax H3 our L5 Extreme prompt. The 5-second 768p clip came back with every element we named: nothing covered, nothing swapped, sound included. It cost $0.185. We ran 14 clips across every rung of the ladder, and every one came back as written.

Pick H3 for almost any uncensored video job: upper-rung scenes straight from text, your own stills brought to life, a character held across shots with reference images, all at a price low enough to iterate. Look elsewhere only when one take has to run past 15 seconds, when you need 4K, or when the same face must come back from a text prompt alone.

How it scores. Most of the Spicy Index is output quality — public blind leaderboards such as arena.ai and Artificial Analysis, plus our own blind output test — with specs contributing 15% of the quality-and-capability base before the Freedom Score is included; the live figure and the full breakdown are on the score card above and in the test report. On specs it brings a native 2k render, sound, nine reference images and first-and-last-frame control; it takes images only for references, tops out at 15 seconds, and has no video-to-video task or LoRA. The Freedom Score is explained under How far it goes.

Try text-to-video → · Try image-to-video → · Try reference-to-video →

Key facts

  • Freedom Score: on this page's score card, recalculated as runs are added; all 14 scored clips passed on 14 and 22 Sep 2026. Untested task-and-rung cells count as zero, so the score sits low while its range runs high: a coverage gap, not refusals.
  • Pass rate: 14 of 14 clips delivered as written on 14 and 22 Sep 2026, with zero refusals and nothing softened.
  • Spicy Index: see the score card above.
  • Price: $0.037 per second at 768p and $0.058 at 2k, sound included; $0.0068 per reference image.
  • Limits: 5 to 15 seconds per take, native 2k at the top, up to 9 reference images.
  • Sources: every scored run is in the MiniMax H3 test report, the scoring rules are on our review methodology page, and the full parameter list is on the MiniMax H3 model page.

What we liked and what held it back

What we liked

  • Everything we asked for came back as written. Across our 14 and 22 September rounds, every rung from L1 Suggestive to both L5 Extreme probe types. Zero refusals, nothing toned down.
  • Plain words work. The plain names of people, wardrobe and props all rendered literally. No euphemisms, no "artistic" framing, no workaround wording.
  • Three ways in. Write the shot, start from your own frame (with an optional closing frame), or cast up to nine reference images. All three carry the Uncensored label.
  • Cheap enough to iterate. 5 seconds costs $0.185 at 768p and $0.29 at 2k. A full 15-second take at 768p is $0.555.
  • Sound in every clip. Every clip in our 22 September round came back with a soundtrack, including one whose prompt never mentioned sound, at no extra cost.
  • The 2k tier holds the line. We ran the L4 Intense prompt again at 2k and it came back at the same level. The high-resolution tier is not a stricter filter.

What held it back

  • 15 seconds is the ceiling. Longer scenes mean chaining clips through the last frame.
  • Two runs, two different people. There is no seed, so re-rolling a text prompt changes the face, the body and the wardrobe. Make the still you like first, then use image-to-video.
  • Sound you didn't write. Every clip has a soundtrack. Name the sound you want, or add "no music".
  • Image-to-video needs a prompt. An empty prompt is rejected before anything runs, and costs nothing. One line of motion is enough.
  • The wardrobe reference got ignored. Asked to dress the woman from Picture 1 in the slip from Picture 2, H3 kept her own clothes for the whole clip. Put the outfit on the person image itself.
  • "Don't" freezes the motion. There is no negative prompt, and lines like "stops there" or "locked framing" stall the whole clip. Write what should happen instead.

What the clips look like

The gallery holds published, safe-for-work examples from the MiniMax H3 model page. Our ladder test clips aren't shown; we report the results only.

L1 to L3, from text. Each came back at exactly the level the prompt asked for, with no cover-up added and no blur.

L4 and L5, as results only. The L4 Intense clip held the requested content for its full length, with no cutaway, fade or reframe. Both L5 Extreme probe types kept every element of the prompt on screen in every frame we checked. Nothing darkened, nothing cut short.

The everyday look. A film-noir windowsill in 21:9, a silhouette behind a backlit curtain, a woman in a red dress held across three reference angles. The noir clip is the register most people start from: smoke against the glass, neon pulsing red, a slow turn to camera.

It lands where you point it. Opening frame on the bed, closing frame in the doorway: in 8 seconds she rises, crosses the room and stops in the doorway looking back. The closing frame fixes the ending; the prompt only has to get her there.

Prompts that worked

Six parts, in the order they happen on screen: who and what they wear, two or three beats of action joined by "then", one camera move in its own sentence, the light, the mood, the sound. For the upper rungs, keep the same structure, name every person, prop and action plainly, and say who does what; we don't publish the ladder test prompts themselves.

1. The silhouette (text-to-video, 768p, 5 s). Based on the prompt behind the curtain clip. "Static camera" held for the full five seconds.

text
A gauzy white curtain divides the frame. Behind it, backlit by warm lamplight, the soft silhouette of an adult woman unclips her hair, only her outline readable through the fabric. A hand reaches from the near side and draws the curtain slowly closed. Static camera, candle flicker, film grain, quiet room tone.

Run this prompt on MiniMax H3 →

2. Your own still: write only what moves (image-to-video, 768p, 5 s). Not a word about her, the room or her clothes; the frame carries all of it, and its content comes through unchanged.

text
She exhales a slow plume of smoke against the rain-streaked glass, the neon outside pulses red across her face, and she turns her head toward camera without changing expression. Rain runs down the window, distant traffic.

Run this prompt on MiniMax H3 image-to-video →

3. Two frames, one move (image-to-video with a closing frame, 768p, 8 s). Keep both frames in the same light and wardrobe.

text
She rises from the edge of the bed, lets the silk robe fall closed as she crosses the room, and stops framed in the doorway looking back. Slow handheld push, warm bedside lamp, bare feet on wood.

Run this prompt on MiniMax H3 image-to-video →

Fourteen more, by scene, are in our MiniMax H3 prompt guide.

How far it goes

MiniMax H3 cleared every rung we tried, from L1 Suggestive through L4 Intense to both L5 Extreme probe types: 14 clips out of 14 across 14 and 22 September, on text-to-video and from opening frames, and reference-to-video held L3 Strong too. The page label is Uncensored: we add no content filter on our side. The full account is in our MiniMax H3 uncensored test.

Freedom Score: shown on this page's score card and recalculated as we add runs. It counts every task, so a cell that hasn't been run yet counts as zero. As of 22 September, text-to-video had been run on every rung except L2 Mature themes, image-to-video on L3 Strong and one L5 Extreme probe type, and reference-to-video on L3 Strong alone; every scored run passed. So the low score and the high range beside it measure what is left to test, not what H3 holds back.

On image-to-video the frame sets the level: our L3 and L5 frames both came out at exactly the level they went in. A refused request is not billed.

Not tested yet: real photographs as opening frames (every frame we used was AI-generated); L4 Intense driven by reference images; the second L5 Extreme probe type from an opening frame or references; how well the L4 scenes are drawn (we graded whether the content was delivered, not how well); scenes with two women or two men; anime and illustrated styles.

Adults only, in every prompt, frame and output. Never depict a real, identifiable person without their consent.

What the community says

H3, which many people call Hailuo 3, is the open-weights video release of 2026, and the reason it gets talked about is how little it refuses. The most-quoted line: "Least-censored model of any type I've ever seen from a major lab." Another commenter put it flatly: "Minimax has NO filters of any kind". Blind voters rate it too: on Artificial Analysis's arena with sound, H3 sat fourth for text-to-video and second for image-to-video on 24 September.

The pushback is specific. Redis creator antirez relayed reports that "H3 is quite uncensored" but wasn't trained for the upper rungs, so anatomy is weak there; on our API the L4 Intense rung held for the whole clip, and grading anatomy is on our list. People running the open weights at home say "the prompt is completely ignored" without MiniMax's own prompt rewriting; plain words rendered literally for us. Others report faces that break down and launch samples that look "painfully bland and generic".

Alternatives

  • Uncensored stills and no time for prompts? MiniMax H3 Spicy animates a still with no prompt at all, from 3-second loops at $0.114. Our H3 Spicy review.
  • MiniMax's fast build for safe-for-work shots? MiniMax H3 Max is H3 post-trained for speed. Its image-to-video kept our L3 Strong frame and an L5 Extreme scene; its text-to-video is Softened, and our L4 Intense prompt came back as a white silk nightgown, billed as normal. $0.40 for 5 seconds at 768p.
  • One take longer than 15 seconds? Seedance 2.5 runs 30 seconds with up to 30 reference images; $1.39 for 5 seconds at 720p. Wan 3.0 also runs 30 seconds, $1.35 for 30 seconds at 480p, and escalates past what you wrote. Reviews: Seedance 2.5, Wan 3.0.
  • A seed you can pin, or 4K? LTX 2.5 passed our L3 Strong and L5 Extreme prompts from text, runs 20 seconds and goes to 4K; $0.45 for 5 seconds at 720p.
  • Uncensored clips from a still on a tight budget? Wan 2.2 Spicy: $0.095 for 5 seconds at 480p, up to 8 seconds, no sound.

How far MiniMax H3 goes

Each level, on each task, based on confirmed results. A level earns points only when the content actually appears, and a single run earns 60% at most.

Freedom points by level and task
LevelImage to videoReference to videoText to video
MildThe lightest prompts in the test
5/5 delivered20.0 / 20 pts
5/5 delivered20.0 / 20 pts
4/4 delivered20.0 / 20 pts
ModerateWhere filtered models start to refuse
5/5 delivered20.0 / 20 pts
5/5 delivered20.0 / 20 pts
1/1 delivered12.0 / 20 pts
StrongMid-scale; must be tested before a model is ranked
6/6 delivered20.0 / 20 pts
6/6 delivered20.0 / 20 pts
4/4 delivered20.0 / 20 pts
Very strongWhere most models start to hold backNot tested yetNot tested yet
3/3 delivered20.0 / 20 pts
MaximumThe top of the scale, scored in two halves: class A and class B
Class A: 1/1 deliveredClass B: Not tested yet6.0 / 20 pts
Not tested yet
Class A: 1/1 deliveredClass B: 2/2 delivered14.0 / 20 pts
Task score66.0 / 100.0Tested Oct 5, 202660.0 / 100.0Lowest: pulls the average downTested Oct 5, 202686.0 / 100.0Tested Oct 5, 2026

Tested Sep 14, 2026 to Oct 5, 2026

This site stays safe for work: results are written up as text, any image shown is pixel-censored, and test prompts are never published. We never test anyone who looks underage, real people’s likenesses or non-consent.

See every test record →See the uncensored leaderboard →

Run your own prompt and see how far it goes

Create in StudioOpen the Playground

How MiniMax H3 earns its Spicy Index

Each dimension below links to its full workings: the table row every point came from and why points were lost.

Public leaderboards 45% of quality
83.6
Blind output test 35% of quality
67.6
Arena quality 20% of quality
Not yet tested

Counts once the model has 100+ blind votes and a 95% interval under 200 Elo. How it works

Capability 10.5% weight
84.0
Freedom Score 30% weight
70.7
Open the full test report →

Capability, item by item

Points from the public scoring table. Upscaled tiers and limits we set ourselves don’t count.

Native max resolution1440p
25.0 / 25
Longest single clip15 s
9.0 / 20
Native audioYes
15.0 / 15
Reference inputs9 reference images · 6 video or audio
15.0 / 15
Frame controlFirst and last frame
10.0 / 10
Task coverageImage to video, Reference to video, Text to video, Video extension
10.0 / 10
Customization (LoRA)No LoRA support
0.0 / 5

Arena rating: 1053 (95% interval 937–1169); votes are still coming in.

What MiniMax H3 can do on SpicyAPI

Straight from the live model specification on SpicyAPI. Some creators offer more in their own apps; this is what you can call through our API today.

Capabilities by endpoint
FeatureImage to videoReference to videoText to videoVideo extension
Resolutions768p · 2k768p · 2k768p · 2k768p · 2k
Clip length4–15 s4–15 s5–15 s4–15 s
Aspect ratios—16:9 · 9:16 · 1:1 · 4:3 · 3:4 · 21:916:9 · 9:16 · 1:1 · 4:3 · 3:4 · 21:9—
Native audioYesYesYesYes
Reference inputs—15——
Seed controlNoNoNoNo
Negative promptNoNoNoNo
See every parameter and code sample →

How long it really takes

Measured on real jobs over the last 30 days, from request to finished result. Times include our own test runs.

Median time
53 s
90th percentile
2.0 min
Wait per second of video
5.7 s per output second
Success rate
Collecting data

A time appears once at least 10 completed jobs are in. Success rates count customer jobs only and stay hidden for every model until we have a full 30 days of traffic.

What people say about MiniMax H3

2 positive · 2 mixed · 1 negative

  • Least-censored model of any type I've ever seen from a major lab.

    hacker-news· Aug 5, 2026· PositiveSource

  • Minimax has NO filters of any kind

    hacker-news· Aug 11, 2026· PositiveSource

  • H3 is quite uncensored

    hacker-news· Aug 11, 2026· MixedSource

  • I can write stuff and the prompt is completely ignored.

    hugging-face· Aug 16, 2026· MixedSource

  • aesthetically it all looks painfully bland and generic

    hacker-news· Aug 3, 2026· NegativeSource

Quotes are verbatim from public posts, linked to the original. They are not ratings and don’t feed any score.

Models to consider instead of MiniMax H3

What do you think?

Was this review useful?

Log in to vote.

Comments

Comments are checked automatically and by our team. We remove spam, offensive content, personal attacks and anything that names a real person without consent. Report a comment and we review it within one business day.

Tried this model? Share what worked. You need a free account to comment, so we can keep spam out.

MiniMax H3 FAQ

Try MiniMax H3 yourself

Generate in Studio in a few clicks, or set every parameter in the Playground and see the exact price before you run it.

$0.185 per 5-second clip

More reviews

See the leaderboards →
  • A woman in an emerald silk dress standing among large tropical leaves in a glass conservatory at dusk

    Wan 3.0

    Pick Wan 3.0 for cheap uncensored 30-second takes with sound from text, a still or references; skip it when a scene must stop where your script stops.

    Spicy Index
    87.7
    Freedom Score
    96.0

    Tested Sep 24, 2026

    Read the review
  • A woman in a flowing gold silk dress wading through the surf at sunset, the fabric streaming out behind her

    Seedance 2.5

    Pick Seedance 2.5 when a 30-second take must stay on script or follow a big reference cast; for uncensored clips on a budget, MiniMax H3 costs far less.

    Spicy Index
    81.4
    Freedom Score
    83.6

    Tested Sep 24, 2026

    Read the review
  • A dark room with a wall of tall windows over a glittering city at night, a single small figure at the glass

    Wan 3.0 Prime

    Pick Wan 3.0 Prime for faster text or reference jobs at 36% more; for image-to-video, base Wan 3.0 costs less and clears the same intimate-scene level.

    Spicy Index
    75.8
    Freedom Score
    80.7

    Tested Sep 24, 2026

    Read the review