Introduction
MiniMax H3 Image LoRA Image editing re-renders the people and things in your reference images into a new scene, outfit, pose or style, using the open-weights MiniMax H3 with up to three of your own LoRA files. Send one to nine images, refer to them as Picture 1, Picture 2 and so on, and get one image at 1k or 2k in the first reference's shape or any of fifteen aspect ratios.
Choose a mode by starting material
| What you have | Mode |
|---|---|
| a written description | MiniMax H3 Image LoRA Text to image |
| one to nine images to change | MiniMax H3 Image LoRA Edit |
| a still you want to animate | MiniMax H3 LoRA Image to video |
Key features & highlights
Identity carried over from the references
Faces, hairstyles and distinctive clothing details carry over while everything else changes. In our test a single portrait went into a completely different setting and pose, and the face and hair stayed recognisably the same person.
Up to nine numbered sources
Combine a person, an outfit, an object and a place from separate images in one request. Numbering them in the prompt tells the model which picture does which job.
Uncensored editing with custom weights
References are accepted as they are, whatever they show, and with a matching LoRA and our strongest content prompt the edit rendered as written. Nothing is reviewed or filtered.
Request parameters
| Parameter | Required | Description |
|---|---|---|
prompt | Yes | Name what each picture contributes, then describe the new scene, outfit or style. Identity comes from the references, so you do not need to describe the face. at least 1 characters · at most 5000 characters |
image_urls | Yes | One to nine references, addressed in the prompt as <Picture 1>, <Picture 2> and so on. The first image also sets the output shape unless you pick an aspect ratio. at least 1 items · at most 9 items |
aspect_ratio | No | Leave it unset to follow the first reference, or pick one of fifteen ratios. Does not change the price. |
resolution | No | 1k is about one megapixel, 2k about four. |
loras | No | Up to three {path, scale} entries trained for MiniMax H3; scale 0 to 4. A character LoRA plus the reference holds identity tightest. Default: |
output_format | No | JPEG by default, PNG or WebP if you need them. |
seed | No | Keeps edits comparable while you change the instruction. |
Pricing
Billed per image by resolution; each reference image after the first adds a small charge. LoRAs add no separate charge.
| Specification | Price |
|---|---|
| $0.054 / image |
| $0.126 / image |
Use cases
New scenes for the same person
One clean portrait is enough to place someone on a balcony, in a hotel room or on a stage. Change the light and the clothes in the prompt and keep the face from the picture.
Wardrobe and product swaps
Send the person as Picture 1 and a garment or product as Picture 2, and ask for the person wearing or holding it. Useful for catalogues and for keeping a character's outfit consistent.
Restyle with a LoRA
Load an anime, film-stock or VHS LoRA and ask for the same image in that style. The composition and the person stay; the medium changes.
Prompt guide & pro tips
Map the pictures, then describe the change
Start with one short sentence per reference, then describe only what should be different. Describing the face again tends to fight the reference rather than help it.
Fewer, clearer references
Two or three references with distinct jobs usually beat nine overlapping ones, and each image after the first adds to the price. Crop references so the part you care about fills the frame.
Keep scales moderate when identity matters
A strong style LoRA at a high scale can pull faces toward its training data. Start style LoRAs at 0.7 to 1 and raise them only if the look is not coming through.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${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: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "minimax/h3-image-lora/edit",
"input": {
"image_urls": [
"https://example.com/reference.png"
],
"output_format": "jpeg",
"prompt": "A ceramic teapot on a linen cloth, soft window light, muted palette",
"resolution": "1k"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.
