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 X Review: Uncensored Video With Sound, Lower Price

Pick MiniMax H3 X for low-cost uncensored video with spoken lines; start from an image or references when the scene is a single person or male subjects.

Open the Playground

Spicy Index

62.2/ 100

Preliminary

Freedom Score

46.0/ 100

Range if the rest pass: 46.0–100.0

Content levels

Tested 5 of 6 probe types · 13 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
$0.16 per 5-second clip
Median time
1.8 min
Success rate
Collecting data

MiniMax H3 X is a tuned build of MiniMax H3 that runs only on SpicyAPI. It is not an official MiniMax release. The trainers describe the tuning as a round of reinforcement-learning post-training aimed at prompt following, picture quality and stable bodies and objects; we have not run a side-by-side quality comparison against standard H3, so we report what it did on our own clips. It starts at $0.022 a second at 480p, with sound on every clip.

What MiniMax H3 X 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

    Night, a rain-soaked rooftop above a glittering city. A couple dances a slow tango in the downpour: she wears a clinging crimson satin slip dress, he wears a soaked white shirt open at the collar. He dips her low, she pulls him back up by the collar, their faces an inch apart, and she says, "Not yet." Rain hissing on the tiles, distant traffic, a bandoneon playing a slow tango. Cinematic, shallow depth of field, wet reflections, warm city glow against blue night.

    Try this model

  • Text to video

    Three shots in a dim, art deco hotel at night. From 0 to 4 seconds: wide shot, a woman in a black backless satin dress walks away from the camera down a long carpeted corridor, heels clicking. At 4 seconds the shot cuts to a close-up of her hand sliding a brass room key into a door lock, the latch clicks. At 8 seconds the shot cuts to a medium shot inside the room: she slips off her heels, glances back over her shoulder at the camera and smiles. Same woman in all three shots. Soft jazz from another room, cinematic, warm amber light. Full-frame 16:9 widescreen picture filling the whole frame, no black bars.

    Try this model

  • Image to video

    She sets the glass down slowly, leans across the bar toward the camera, rests her chin on her hand and whispers, "Last call." A record crackles on the turntable, low soul music, ice clinking. The camera pushes in slowly. Warm amber light, shallow depth of field.

    Try this model

  • Reference to video

    The man from Image 1 performs the same moves as the dancer in Video 1, in a sunlit concrete loft: a smooth spin, a shoulder roll, then he points at the camera and laughs. Keep his face and white tank top from Image 1, copy only the motion from Video 1. Upbeat funk music with a clear beat.

    Try this model

  • Reference to video

    The woman from Image 1, in her black slip dress, tries on the trench coat from Image 2 in the fitting room from Image 3. She belts it, turns to the gold mirror, then lets the coat slide off one shoulder and glances back at the camera over that shoulder. Soft spotlight, rustle of wool, quiet lounge music. Cinematic, shallow depth of field.

    Try this model

Make your own with MiniMax H3 X

Open the Playground

The verdict

Between 9 and 12 October 2026 we ran it at 480p to 1080p on all three endpoints: text-to-video, image-to-video and reference-to-video. Not one request was refused. Mild and Moderate came back as written everywhere. Image-to-video and reference-to-video also delivered Strong, Very strong and Maximum; at Very strong the clip stays implied, with the most direct moments kept out of frame. Text-to-video is the soft spot: single-person scenes at Strong and requests with male subjects were toned down in every run.

Pick it for a cheap NSFW video that still talks, cuts between shots on cue and holds two faces across a scene. Start from a still or from references whenever the scene is one person or involves male subjects. Pick base MiniMax H3 when you need the full grid from a written prompt: on the same prompts it delivered 21 of 21.

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

Key facts

  • Content: no refusals across more than 250 clips from this build and its Turbo sibling on 9 to 12 Oct 2026. Image and reference endpoints delivered every level; text-to-video toned down single-person Strong scenes and male-subject requests. Details under How far it goes.
  • Price: from $0.022 per second at 480p; 720p and 1080p cost more. The model page shows live prices.
  • Limits: 2 to 15 seconds per clip, 480p, 720p (default) or 1080p, 24 fps, stereo sound that cannot be switched off. Prompts run up to 7,000 characters.
  • Inputs: text-to-video takes six aspect ratios; image-to-video takes an opening frame and an optional closing frame and keeps the frame's shape; reference-to-video takes up to 9 images, 2 video clips (15 seconds in total) and 3 voice clips.
  • Speed: a 5-second clip took about 50 to 65 seconds at 480p and 75 to 130 seconds at 720p; 1080p took about three minutes.

What we liked and what held it back

What we liked

  • Spoken lines come out word for word. We put a quoted line in the prompt and ran speech-to-text on the soundtrack; it matched in all three checks, with lip movement to match, including a run that used a reference voice clip.
  • Timed shots land where you put them. A 15-second prompt with three timed shots cut at about 5 and 10 seconds, with the same two people in every shot.
  • Two faces survive a scene. Reference-to-video held both faces and both outfits when we sent two portraits, and the model-page example keeps a face, a coat and a room from three separate photos.
  • Stronger input is not blocked. Opening frames, closing frames, reference images and reference clips that already showed the higher levels were all accepted, and the clips came back as written.
  • Prompt expansion is optional. It is off by default and made no visible difference in our reruns, so what you write is what you get.

What held it back

  • Text-to-video tones down more than the other two. Single-person scenes at Strong were toned down in 2 of 2 runs, male-subject requests in every run, and close-ups and scenes with more than two people were softened too. A scene with two men stayed implied.
  • Very strong stays implied. On every endpoint the most direct moments are kept out of frame. Reference-to-video delivered it once and stopped short once.
  • 1080p is a bigger frame, not a sharper one. The file is 1920 x 1088 but encoded at about 3.3 Mbps against about 10 Mbps at 720p, and in our crops it looked no crisper. Use 720p unless you need the pixels.
  • No aspect-ratio control on image-to-video. The clip keeps the shape of your opening frame. Opening and closing frames cannot be mixed with references in one request.
  • Slower than the Turbo build. The same 5-second clip takes roughly twice as long here as on H3 X Turbo.
  • A tuned build is not a new model. Its content limits are slightly stricter than base H3, and we have not measured the quality gain the trainers claim.

What the clips look like

The gallery holds published, safe-for-work examples from the MiniMax H3 X model page. Our content-test clips are not shown; we report those results in words.

A line in the rain. A couple dances a slow tango on a rooftop, he dips her low, and in a close-up she says "Not yet." The prompt names rain, traffic and a bandoneon; the line is heard over slow, melancholic music.

Three shots in 12 seconds. One prompt lists a wide corridor walk, a hand turning a key at 4 seconds and a medium shot in the room at 8. The cuts came at about 3.9 and 8.3 seconds with the same woman throughout.

A frame that whispers. A woman behind a walnut bar goes in as the opening frame, and the prompt only says what follows: she sets her glass down, leans in and says "Last call." Quiet soul music sits under the line.

Copying a move. One portrait and one short dance clip go in as Image 1 and Video 1. "Copy only the motion from Video 1" and "keep his face and white tank top from Image 1" split the two jobs, and the clip repeats the spin and the shoulder roll.

Three references, one scene. A portrait, a coat photo and an empty fitting room go in as Images 1, 2 and 3, and face, outfit, coat and room all match.

Prompts that worked

Write dialogue in double quotes, say who speaks, and give each shot a start time. For image-to-video, describe only what happens next. For reference-to-video, name references by type and position: "the woman from Image 1", "the motion of Video 1".

1. Timed shots, one woman (text-to-video, 16:9, 12 s). The prompt behind the corridor clip.

text
Three shots in a dim, art deco hotel at night. From 0 to 4 seconds: wide shot, a woman in a black backless satin dress walks away from the camera down a long carpeted corridor, heels clicking. At 4 seconds the shot cuts to a close-up of her hand sliding a brass room key into a door lock, the latch clicks. At 8 seconds the shot cuts to a medium shot inside the room: she slips off her heels, glances back over her shoulder at the camera and smiles. Same woman in all three shots. Soft jazz from another room, cinematic, warm amber light. Full-frame 16:9 widescreen picture filling the whole frame, no black bars.

Run this prompt on MiniMax H3 X text-to-video →

2. A frame, then a spoken line (image-to-video, 6 s). Send your own opening frame; the prompt starts after it.

text
She sets the glass down slowly, leans across the bar toward the camera, rests her chin on her hand and whispers, "Last call." A record crackles on the turntable, low soul music, ice clinking. The camera pushes in slowly. Warm amber light, shallow depth of field.

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

3. Move transfer (reference-to-video, 16:9, 5 s). Needs one portrait and one short motion clip.

text
The man from Image 1 performs the same moves as the dancer in Video 1, in a sunlit concrete loft: a smooth spin, a shoulder roll, then he points at the camera and laughs. Keep his face and white tank top from Image 1, copy only the motion from Video 1. Upbeat funk music with a clear beat.

Run this prompt on MiniMax H3 X reference-to-video →

Weak: "A woman walks down a hallway and talks." No shot times, no line in quotes, no light, and the model fills every gap itself. Strong: prompt 1, where shots, times, sound and the repeat-the-person sentence are all written out.

How far it goes

SpicyAPI adds no content filter. Between 9 and 12 October 2026 we ran adult fictional subjects through all three endpoints, one prompt per level, then reran the weak spots with prompt expansion off; the test prompts are not published. Zero requests were refused. Results for the first full grid on this build:

LevelText-to-videoImage-to-videoReference-to-video
MildDeliveredDeliveredDelivered
ModerateDeliveredDeliveredDelivered
Strong, one personToned down (2 of 2)Delivered (2 of 2)Delivered (2 of 2)
Very strongDelivered (2 of 2)Delivered (2 of 2)1 delivered, 1 toned down
Maximum, restraint and impactDelivered (2 of 2)Delivered (2 of 2)Delivered (2 of 2)
Maximum, cinematic goreDeliveredDeliveredDelivered
Strong, male subjectsToned down (2 of 2)Toned down (2 of 2)Delivered (2 of 2)

Reruns on 11 and 12 October, with prompt expansion off and requests sent one at a time, gave the same picture: text-to-video still toned down single-person Strong scenes and male-subject requests, and the other two endpoints did not change. Reference-to-video is the endpoint that delivered male-subject requests as written.

Same-prompt check: on 11 October base MiniMax H3 delivered all 21 of 21 clips on the same seven levels across the three tasks, including the two cases where H3 X toned down text-to-video. That is the reason to choose base H3 for text-only work at the upper levels.

Not tested yet: a blind output-quality comparison against base H3, 1:1, 4:3 and 3:4 text-to-video ratios, and long 15-second clips at the upper levels.

Adults only, in every prompt, frame and reference. Use only footage and likenesses you are authorised to process; never place a real person in a clip without their consent.

Alternatives

How far MiniMax H3 X 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
1/1 delivered12.0 / 20 pts
1/1 delivered12.0 / 20 pts
1/1 delivered12.0 / 20 pts
ModerateWhere filtered models start to refuse
1/1 delivered12.0 / 20 pts
1/1 delivered12.0 / 20 pts
1/1 delivered12.0 / 20 pts
StrongMid-scale; must be tested before a model is ranked
1/1 delivered12.0 / 20 pts
1/1 delivered12.0 / 20 pts
Not tested yet
Very strongWhere most models start to hold back
1/1 delivered12.0 / 20 pts
Not tested yet
1/1 delivered12.0 / 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
Class A: 1/1 deliveredClass B: Not tested yet6.0 / 20 pts
Class A: 1/1 deliveredClass B: Not tested yet6.0 / 20 pts
Task score54.0 / 100.0Tested Oct 9, 202642.0 / 100.0Lowest: pulls the average downTested Oct 9, 202642.0 / 100.0Lowest: pulls the average downTested Oct 9, 2026

Tested Oct 9, 2026 to Oct 9, 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

Open the Playground

How MiniMax H3 X 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
Not yet tested

No exact match on a tracked public leaderboard yet. How it works

Blind output test 35% of quality
66.9
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
81.5
Freedom Score 30% weight
46.0

Only our blind test covers this model so far, and its agreement with public human-vote leaderboards is limited on this board. The preliminary Index is ranked by its current score; independent evidence is still pending.

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 resolution1080p
25.0 / 25
Longest single clip15 s
9.0 / 20
Native audioYes
15.0 / 15
Reference inputs9 reference images · 5 video or audio
15.0 / 15
Frame controlFirst and last frame
10.0 / 10
Task coverageImage to video, Reference to video, Text to video
7.5 / 10
Customization (LoRA)No LoRA support
0.0 / 5

What MiniMax H3 X 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 video
Resolutions480p · 720p · 1080p480p · 720p · 1080p480p · 720p · 1080p
Clip length2–15 s2–15 s2–15 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 audioYesYesYes
Reference inputs—14—
Seed controlNoNoNo
Negative promptNoNoNo
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
1.8 min
90th percentile
6.0 min
Wait per second of video
24.5 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.

Models to consider instead of MiniMax H3 X

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 X FAQ

Try MiniMax H3 X yourself

Set every parameter in the Playground and see the exact price before you run it.

$0.16 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 laughing in a swimming pool at night, hair swept back, pink and teal neon light rippling across the water around her.

    MiniMax H3 HighSpeed

    Pick MiniMax H3 HighSpeed if you want 768p takes with sound that we saw come back as written at Very strong; for cheaper drafts or 2K, look at Lite or H3.

    Spicy Index
    52.0
    Freedom Score
    20.0

    Tested Oct 27, 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
    82.2
    Freedom Score
    83.6

    Tested Sep 24, 2026

    Read the review