Is MiniMax H3 Uncensored? Our Test Results

If you’re making an adult video, you might start with a text prompt or a still you already like. MiniMax H3, H3 Max and H3 Spicy support different workflows and differ in how fully they deliver the requested scene.

Short answer: yes. Based on our September 2026 tests, SpicyAPI-hosted MiniMax H3 delivered all 12 adult test clips, from text, from an opening image and from reference images, plus 2 clips of non-adult violence. The grid below shows where each landed. Five seconds costs $0.185 at 768p. H3 Max refused or softened every L4 request, so base H3 is the starting point for higher-intensity adult work.

The 30-second version

• H3 Max is for acting, not range. It suits talking heads and short emotional beats at L1 and L2, where its acting and sound are the draw. It is not the choice for L3 and L4: it refused or toned down most of those prompts that H3 delivers (H3 Max review).
• H3 Spicy animates a still with no prompt at all, for clips from 3 seconds. Its 1080p is an upscaled tier.
• Pick H3 for almost everything. Look elsewhere only for a take longer than 15 seconds, 4K, or the lowest price per image-to-video clip (comparison below).

Try MiniMax H3 text-to-video →

How far each MiniMax model goes

L1–L4 and L5a identify the adult test categories below. L5b is a separate non-adult probe, not another step up the adult ladder.

• L1 Suggestive: suggestive imagery.
• L2 Mature themes: moderate NSFW imagery.
• L3 Strong: advanced NSFW imagery.
• L4 Intense: higher-intensity adult scenes between two adults.
• L5a: a consensual adult scene with restraint themes. L5b: cinematic gore, a separate non-adult probe.

✅ came back as written · ⚠️ toned down · ❌ refused · — no clips.

Base H3's L1 to L5a counts add up to the 12 adult clips; L5b adds 2 more. The H3 Max and Spicy rows show three runs per category (Spicy also has three earlier clips, one at L4: pooling that L4 clip with the later three gives three matched and one softened). The sample sizes differ, so read this as a map of what each model delivers, not a quality contest.

Uncensored runs with no content filter added on our side; Softened still succeeds and still bills, but comes back toned down; Filtered refuses, and a refusal normally releases the hold on your balance.

Test reports: H3, H3 Spicy, H3 Max. Try a model: H3 text-to-video · H3 image-to-video · H3 reference-to-video · H3 Spicy

What the clips actually look like

L3, Strong. The published test frame was pixelated by SpicyAPI for presentation; that mosaic is not model filtering.

L4 and L5. L4 and L5a clips on H3 held the action for the full clip instead of cutting away, including the L4 text request at 2k.

Sound. H3 and H3 Max generate sound. H3 Spicy has no sound option and its sample clip is silent; add audio in the edit.

What a softened clip looks like. H3 Max, L4 prompt: a woman in a white silk nightgown. The task succeeds and bills like any other clip; no error says the scene was toned down.

The everyday look. The moody L1 register most people start from, from H3 text-to-video.

How to write MiniMax H3 prompts

The recorded H3 tests accepted direct wording. For a useful shot, specify what should appear instead of relying on a generic mood label. A prompt that lands names five things, in this order: who (and what they're wearing), the one action, the camera, the light, the sound.

Solo, suggestive (text-to-video)

An adult woman in an ivory silk slip sits on rumpled hotel sheets at dusk. She holds the lens with a slow, steady look and leans back on her hands. Warm window backlight, city lights behind her, slow push-in, 35mm grain. Quiet room tone and distant traffic.

Run this prompt on H3 →

Two people, one bed (text-to-video)

Two adults, a man and a woman, lie in white sheets in a dim loft at night. She rests her forehead against his and they share a slow kiss. The camera orbits the bed once. One warm bedside lamp, rain streaking the skylight. Soft breathing and rain, no music.

Run this prompt on H3 →

Say "two adults" and who is who; keep to one shared action and one camera move.

Motion and camera (text-to-video)

An adult woman in a black evening dress walks slowly toward the camera down a dark hotel corridor, stops, turns her back to the lens and lets her hair down, then looks back over her shoulder. The camera tracks backward at her pace, then holds on the look. Hard rim light, deep shadows. Heels on carpet, the low hum of the corridor.

Run this prompt on H3 →

Write actions in the order they happen, joined by "then", and give the camera its own sentence.

Animating your own still (image-to-video)

The opening frame supplies the initial subject, outfit and composition, and so sets the content level: the prompt only moves it. Describe the motion, not the picture; the generated motion can still change details or reframe the shot.

She exhales and turns her head toward the camera, stretches both arms above her head and sinks back into the pillows. The bedside lamp flickers once. Slow push-in, then hold. Quiet room tone, rain on the window.

Run this prompt on H3 image-to-video →

No prompt at all (H3 Spicy)

H3 Spicy is the only model in this line that takes a still with no prompt: send the image alone to let the model choose the motion. Use a prompt when the direction matters. For a specific beat, keep it short:

Slow breath, a small turn toward the lens, hair shifting. Hold on her eyes.

Run this prompt on H3 Spicy →

Same character across shots (reference-to-video)

Give each reference image one job and call them Picture 1, Picture 2. Put the outfit on the person image itself; a separate garment reference was ignored in our three-reference example. More in the MiniMax H3 prompt guide.

Picture 1 is the woman, already wearing the red silk dress. Picture 2 is the bedroom. She sits on the edge of the bed in the room from Picture 2, pulls her hair over one shoulder and looks up at the camera. Slow handheld drift, warm lamp, quiet room tone.

Run this prompt on H3 reference-to-video →

Where H3 trips up, and the fix

• Two runs, two different people. A new text-only run can change the person and wardrobe; base H3 has no seed control. Fix: make the still you like with an uncensored image model, then animate it with image-to-video.
• A soundtrack you didn't write. H3 generates picture and sound together. Fix: describe it (room tone, rain, breathing, "no music") or mute it in the edit. H3 Spicy has no sound option.
• H3 Max text-to-video makes a square 8-second clip unless told otherwise. Set the length and aspect ratio every time: left unset, it bills $0.64 instead of $0.40 for 5 seconds.
• H3 Spicy's 1080p isn't sharper. Its native canvas is 768p and 1080p is the same picture enlarged. Fix: render at 768p unless you need the bigger frame.
• 15 seconds is the ceiling. Fix: take the last frame of a clip you like and use it as the next opening frame in image-to-video. The longer uncensored video guide walks through the chain. For a longer single generation, see the next section.

H3 or another model: which to pick

Prices are for 5 seconds at the cheapest tier, from the public catalogue.

Which one? A text prompt in the adult categories tested here: start with H3. A folder of stills and no time to write prompts: H3 Spicy. A single generation over 15 seconds from text: Wan 3.0 ($1.35 for 30 seconds at 480p), or LTX 2.5 for 20 seconds with a pinnable seed. Cheapest from a still: Wan 2.2 Spicy ($0.095 per 5 s), if you can live with a softened L5; Seedance 1.5 Pro Spicy is cheaper still at about $0.06 per 5 s at 480p. Deeper tests: Seedance, Wan, LTX 2.5, Vidu Q3. Choosing between H3, Spicy, the LoRA builds and Singularity: four-version comparison.

What a clip costs

Example: four 768p drafts plus a 2k keeper on H3 come to 4 × $0.185 + $0.29 = $1.03. H3 takes 5 to 15 seconds; a 15-second take at 768p is $0.555; a 3-second H3 Spicy loop at 480p is $0.114.

Unset resolution on H3 means 768p ($0.185 per 5 s); reference-to-video adds $0.0068 per reference image. Softened clips are billed. A content refusal normally releases the hold in full, but a model page can state an exception, so check cost and settled on the task; see the content and usage policy.

Try it in the Playground first

1. Open the H3 text-to-video page and stay on the Playground tab in Form view. Sign in to run it.
2. Paste the solo prompt from above into Prompt.
3. Leave Resolution at 768p (the default; the other option is 2k) and set Duration to 5. Aspect ratio opens at 16:9. The price estimate updates before you submit.
4. Run it, check the clip, then step up to 2k once the motion is right.

The MiniMax H3 text-to-video Playground in Form view, showing the Prompt field and the Run button.

For your own still, use the image-to-video page: add the still under Input image (an optional Closing frame tightens the ending) and write only what moves. The Playground body is the same JSON the API takes, so a setup that works there copies straight into code. H3 Max defaults to a square 8-second clip, so set length and shape there too.

Then run it from code

Install the Python SDK (pip install spicyapi), set SPICY_API_KEY in your environment and run:

import uuid
from spicyapi import SpicyClient, output_assets

client = SpicyClient()  # reads SPICY_API_KEY from the environment

task = client.create_task(
    model="minimax/h3/text-to-video",
    input_data={
        "prompt": "An adult woman in an ivory silk slip sits on rumpled hotel sheets at dusk. "
                  "She holds the lens with a slow, steady look and leans back on her hands. "
                  "Warm window backlight, city lights behind her, slow push-in, 35mm grain. "
                  "Quiet room tone and distant traffic.",
        "resolution": "768p",
        "duration_seconds": 5,
    },
    idempotency_key=str(uuid.uuid4()),
)

final = client.wait_for_terminal(task["taskId"])
if final["state"] != "succeeded":
    raise RuntimeError(f"{task['taskId']} ended as {final['state']}: {final.get('errorCode')}")
print(output_assets(final)[0]["url"])  # copy the clip to your own storage now

To switch models, change model and send that model's own input fields: the image models need an image_url, and H3 Spicy accepts a seed that H3 does not. Creating, waiting, downloading and error codes are in the tasks guide and the Python SDK reference; every H3 option is on the MiniMax H3 model page. Results expire, so read retention and delivery before you build on the returned link.

Or let an Agent run it

When an assistant should run the clips, all three official tools show the price and ask you before a paid call.

• CLI. Put the same input object in input.json, then:

``bash spicyapi tasks create --model minimax/h3/text-to-video --input-file input.json --wait ``

See the CLI docs; npx @spicyapi/cli works without installing.

• MCP server. Add the MCP server to your assistant, then ask: "Run MiniMax H3 text-to-video with this prompt at 768p for 5 seconds." It calls spicyapi_task_create, which stops for your price confirmation, then spicyapi_task_wait.
• Skill. npx skills add spicyapi-ai/spicy-skill installs the Agent Skill, which has a coding assistant read the live model schema instead of guessing fields.

FAQ

Is MiniMax H3 uncensored?

On our API, yes: all 12 adult requests delivered the requested content across text, image and reference tasks, and SpicyAPI adds no content filter. The grid shows each level.

Is MiniMax H3 uncensored if I run it myself or use MiniMax's own service?

This page covers SpicyAPI-hosted H3. MiniMax's official repository describes moderation in its own hosted preprocessing workflow. The downloadable weights use the H3 Community License, which grants rights worldwide except the EU, UK, South Korea and US. The acceptable-use policy and the downstream-safeguard clause still apply, so open weights do not mean unrestricted permission. Local deployment also needs task-specific checkpoints and suitable hardware.

Why did my H3 Max video come back toned down?

A completed Max task is not proof the scene matched: from L3 up it softens or refuses, and image-to-video refused every L4 request. Use base H3 for that work.

What's the difference between MiniMax H3 and H3 Spicy?

H3 takes text, an opening frame or reference images, goes up to 2k, makes sound and costs less per second. H3 Spicy is an uncensored build of H3 that only animates images and has no sound option, but it doesn't need a prompt, starts at 3 seconds and adds 540p and 1080p.

Does MiniMax H3 make sound?

Yes on H3 and H3 Max: describe the track you want. H3 Spicy has no sound option.

Can MiniMax H3 go all the way to L5?

Yes: both L5a clips rendered as asked. L5b also rendered, but it is a non-adult violence probe.

Do I pay for refused or softened clips?

Softened: yes, at the normal price. Refused: normally no, because the hold is released, unless the model page states an exception. That is why you check the clip and the task's cost, not the status alone.

How long does a MiniMax H3 video take?

The published H3 report lists a median of about 35 seconds. Check the task state rather than waiting on a fixed time.

Can I use a photo of a real person?

Only with that person's documented consent, and never for an intimate depiction of someone who hasn't given it.

Your part

We don't put a content filter in front of these models, so that does not remove your responsibilities for the inputs and outputs. Everyone on screen must be an adult. Any real, identifiable person must have consented. You need the rights to everything you upload, and you're responsible for the law where you publish. SpicyAPI is not available in every country; the policy page lists the regions we can't serve.

Run your first clip

Create an account, open MiniMax H3 in the Playground and paste the solo prompt from above. Draft at 768p for $0.185, then step up to 2k once the motion is right.