Can my MiniMax H3 LoRAs generate stills here?

Yes: in our tests MiniMax H3 VHS and Studio1939 cartoon LoRAs affected P-H3-Image output. The original [MiniMax H3 Image LoRA](/models/minimax-h3-image-lora) page provides the original H3 still-image route at 1k and 2k; this adaptation adds 1.5k, 3k and 4k settings.

Why did a wrong LoRA link still produce a billed image?

An invalid link can be silently skipped while image generation completes at the LoRA rate. Our repository-page and missing-file tests returned the same pixels as a scale-0 control. Supply the direct .safetensors download address, then check the result against a same-seed run without the LoRA.

How can I tell whether a LoRA changed the result?

Keep the prompt, seed, ratio and tier fixed while switching the LoRA on and off. Our VHS file showed a clear difference at scale 1; the realism file at 0.8 had little visible effect. Similar results warrant checking the link, but do not establish a loading failure by themselves.

Does P-H3-Image LoRA support uncensored NSFW images?

The model has no content filter and returned images without refusals in our tests, including LoRA requests. Extreme instructions may be softened in the result; for these requests, every person you describe must be an adult.

What if I leave the LoRA list empty?

In our tests an empty list produced nearly the same composition as the base P-H3-Image endpoint with a matching prompt and seed, though the files differed. The fixed per-image LoRA surcharge still applies with zero attachments.

Do all three LoRA slots cost extra?

A single fixed surcharge applies per image at the LoRA endpoint whether zero, one or three files are attached. Resolution determines the tier charge, and editing also adds a small charge per source after the first. A failed request is uncharged; a skipped invalid LoRA link is a completed, charged generation.

Can a LoRA transform an existing photograph?

Yes, both modes accept loras. The stage-photo cartoon test retained the pose, green dress and microphone after applying Studio1939 style at scale 1.

P-H3-Image LoRA API: NSFW Images, H3 LoRAs at 4K

pruna-ai/p-h3-image-lora/text-to-image

Generate a P-H3-Image still with up to three MiniMax H3 LoRA attachments, each controlled by its own scale.

From $0.042/image

Run model View API

Input

Text to image

Prompt118/5000

Aspect ratio

Resolutionaffects price

Style LoRAs1/3

Vh5tape Vhs Lora Minimax H3

Output format

JPEGPNGWebP

Seed

Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run

Example result

Edit

Split into layers

Animate

{
  "model": "pruna-ai/p-h3-image-lora/text-to-image",
  "input": {
    "aspect_ratio": "3:2",
    "resolution": "1k",
    "loras": [
      {
        "path": "https://huggingface.co/KennethFal/vh5tape-vhs-lora-minimax-h3/resolve/main/vh5tape.safetensors",
        "scale": 1
      }
    ],
    "output_format": "png",
    "seed": 42,
    "prompt": "vh5tape, a woman in a red raincoat standing in a lit motel doorway at night, rain, neon sign, adult, 1990s home video."
  }
}

Introduction

P-H3-Image LoRA Text to Image accepts MiniMax H3 LoRA weights for a still generated from words. It uses the same underlying P-H3-Image model as the endpoint without attachments.

Key features & highlights

Attach no more than three files and assign a scale between 0 and 4 to each. In our tests vh5tape changed a same-seed image at strength 1, while a realism file at 0.8 produced little visible difference. Five size tiers remain available through 4k.

Request parameters

Request parameters for P-H3-Image LoRA · Text to Image
Parameter	Required	Description
`prompt`	Yes	Required, with no preset wording. Include the LoRA's trained trigger followed by the scene you want; keep the description concise. at least 1 characters · at most 5000 characters
`aspect_ratio`	No	For LoRA generation choose one of 15 supported ratios, defaulting to 1:1. The set includes 1:3, 3:1, 9:21 and 21:9; each has the same charge within a tier. `1:1` · `16:9` · `9:16` · `4:3` · `3:4` · `3:2` · `2:3` · `21:9` · `9:21` · `2:1` · `1:2` · `3:1` · `1:3` · `5:4` · `4:5` · Default: `1:1`
`resolution`	No	The size setting begins at 1k (~1 MP); you may choose 1.5k (~2.4 MP), 2k (~4 MP), 3k (~9.4 MP) or 4k (~16.8 MP). Each resolution tier has its own image price. `1k` · `1.5k` · `2k` · `3k` · `4k` · Default: `1k`
`loras`	No	Optional list of up to 3 {path, scale} entries trained for MiniMax H3; omit attachments for a control. Scale ranges from 0 to 4 and defaults to 1. Default: `[]` at most 3 items
`output_format`	No	Choose jpeg, png or webp for the generated file; leaving the field out selects jpeg. File-format choice does not increase the charge. `jpeg` · `png` · `webp` · Default: `jpeg`
`seed`	No	Reuse a seed as the same random source when checking a single LoRA change; this optional field has no required default value. `≥ 0` · `≤ 2147483647`

Pricing

Billing combines the selected per-image size tier and a fixed small LoRA surcharge, with zero through three files priced alike. Ratios and formats add no fee; editing adds a small fee per input after the first. Failed requests are uncharged, but silently skipped invalid LoRA links still produce charged images.

Pricing
Specification	Price
`1k`	$0.042 / image
`1.5k`	$0.054 / image
`2k`	$0.066 / image
`3k`	$0.114 / image
`4k`	$0.162 / image

Use cases

Try the motel scene with a VHS file or the club trio with Studio1939 style. For a source-photo conversion, open LoRA Image Editing.

Explore key use-case tutorials

Prompt guide & pro tips

Begin with a 1k control without a LoRA, then repeat the prompt and seed with the file enabled. Use the actual .safetensors download URL and the file's trigger. Repository-page and missing-file URLs silently skipped the LoRA in our tests; generation still completed and was charged. A fixed surcharge is charged per image regardless of the number of attachments.

API quick start

# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
  -H "Authorization: Bearer $SPICY_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
  -d '{
  "model": "pruna-ai/p-h3-image-lora/text-to-image",
  "input": {
    "aspect_ratio": "1:1",
    "output_format": "jpeg",
    "prompt": "A ceramic teapot on a linen cloth, soft window light, muted palette",
    "resolution": "1k"
  }
}'

# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
  -H "Authorization: Bearer $SPICY_API_KEY"

See the API reference for authentication, the complete schema, polling, and error handling.

Missing the model you need?

H3 LORA STILL IMAGE API

P-H3-Image LoRA API at a glance

Bring MiniMax H3 LoRAs to stills, then verify their effect before scaling up.

LoRA file slots: 3; MiniMax H3 training is required for the files used here.
Per-file scale: 0 – 4; An omitted scale uses 1; compare against 0 to inspect the effect.
Output tier ceiling: 4k; LoRA requests retain the five P-H3-Image size settings.
Source slots for edits: 9; The first slot holds the image to transform.
Output shapes: 15; Editing inherits the first source ratio unless you select one.

Results

What you can make with P-H3-Image LoRA

VHS comparison
Verify the tape effect with a control
In our tests vh5tape at scale 1 changed the composition and image finish relative to scale 0 at the same seed. A completed request alone cannot confirm that the weights were applied.
1k, 3:2; vh5tape LoRA at 1.
Cartoon edit
Redraw the singer with Studio1939
The Studio1939 file at strength 1 changes the photograph into a cartoon. Our test retained the stage pose, outfit and microphone through the conversion.
1k; inherited 2:3 source ratio, Studio1939 LoRA at 1.
Photo instruction
Request a living-room tape scene
The home-video example combines the vh5tape trigger, one source portrait and a denim-jacket scene instruction. It uses a 4:3 frame at 1k.
1k, 4:3; vh5tape LoRA at 1.

Content test

Is P-H3-Image LoRA uncensored? What we tested

Content levels tested on P-H3-Image LoRA, by mode
Level	Image editing	Text to image
Mild	Not tested yet	Not tested yet
Moderate	Not tested yet	Not tested yet
Strong	RendersTested Nov 5, 2026	RendersTested Nov 5, 2026
Very strong	Not tested yet	Not tested yet
Maximum	Not tested yet	Not tested yet

How we test

How each mode behaves

All modes: A wrong LoRA link silently skips the LoRA: the image is still generated and charged at the LoRA endpoint rate. Use a direct .safetensors file URL; repository pages and missing-file links produced this behavior in our tests.
All modes: Effects vary by file; our realism LoRA comparison showed little change at scale 0.8. An empty loras list still incurs the fixed LoRA endpoint surcharge.
All modes: For each LoRA request, every person you describe must be an adult. You remain responsible for generated and published material.
All modes: Images generated here remain available for 14 days before deletion.
All modes: The API returns a single still and exposes no negative prompt, guidance or sampling-step settings.

Use cases

What to build with the P-H3-Image LoRA API

Home-video stills
The vh5tape example pairs a motel doorway with a VHS LoRA trained for MiniMax H3.
Cartoon conversion
In our tests the Studio1939 LoRA redrew a stage photograph as a 1930s cartoon while retaining the outfit, pose and microphone.
Evaluate a public weight file
Make a comparison at the 1k tier before relying on a LoRA for larger stills. A subtle result can reflect either the file's effect or a link problem.

Prompts

P-H3-Image LoRA prompt examples you can copy

Text to image
Raincoat at the motel
vh5tape, a woman in a red raincoat standing in a lit motel doorway at night, rain, neon sign, adult, 1990s home video.
Why it works. The vh5tape trigger connects the requested home-video treatment to the loaded VHS file.
Try this prompt
Text to image
Cartoon club trio
Studio1939 style. A jazz trio playing in a small smoky club, a double bass player, a drummer and a trumpeter in bow ties, a round lamp on every table, 1930s rubber-hose cartoon.
Why it works. The Studio1939 trigger and period cartoon instruction align the scene with the attached style file.
Try this prompt
Image editing
Cartoon stage singer
Studio1939 style. Turn <Picture 1> into a 1930s rubber-hose cartoon frame, same pose and outfit.
Why it works. The edit requests a cartoon treatment while preserving the source pose and clothing.
Try this prompt
Image editing
Home-video sofa scene
vh5tape. <Picture 1> is a woman. Keep her face and hair. Show her in a 1990s living room on a home video tape, wearing a denim jacket, sitting on a floral sofa and laughing at the camera.
Why it works. The source token identifies the person while the trigger and living-room instruction request the new treatment and setting.
Try this prompt

Pricing

P-H3-Image LoRA API pricing

One image starts at $0.042 at 1K and $0.162 at 4K.

See every price tier How pricing works

Get started

How to use the P-H3-Image LoRA API in three steps

Step 1
Supply the weight-file address
Use path for a direct .safetensors download, such as a Hugging Face /resolve/main/ URL. A repository web page is not the weight file.
Step 2
Add the trained trigger
Use vh5tape for the VHS example or Studio1939 style for the cartoon example, followed by a short description or edit instruction.
Step 3
Run an on/off check at 1k
Use one prompt and seed for two runs, with the chosen LoRA enabled in one and absent or at scale 0 in the other. Compare the pictures before committing to a larger tier.

Open the playground Get an API key

Parameter reference and code samples are in the API and Readme tabs at the top of this page.

FAQ

P-H3-Image LoRA API FAQ

About P-H3-Image LoRA

Test your MiniMax H3 LoRA on a still at 1k, inspect it beside a same-seed control, then choose a larger tier if needed. Treat image completion and LoRA application as separate checks.