How is this different from Qwen Image 2.1 LoRA?

It runs a different base. Qwen-Image-2512 and Edit-2511 have a much larger library of community LoRAs, and weights made for one base do not transfer reliably to the other. Pick the family whose base your LoRA was trained on.

Can I use uncensored LoRAs?

Yes. The weights are not reviewed and the output is not filtered. With LoRAs trained for unfiltered content loaded, both text-to-image and editing rendered our strongest content probe as written. Rights and consent for everything you generate stay with you.

Which LoRAs work on editing?

LoRAs trained for Qwen-Image-Edit-2511, such as camera-angle, relighting or photo-realism LoRAs. Text-to-image LoRAs for Qwen-Image-2512 are a separate group. Keep each scale at 2 or below on editing.

How big are the images?

Text-to-image goes up to 1536 pixels on the long side: 1328 x 1328 at 1:1 and 1536 x 864 at 16:9. Editing returns about one megapixel in the shape of the first source.

Does the number of edit sources change the price?

No. One, two or three sources cost the same, and so does every text-to-image aspect ratio. Loading LoRAs adds no separate charge.

Why did my LoRA request fail?

Usually the link needs a login, since gated repositories answer with HTTP 401, or it points to a web page instead of the file. Use a public URL that downloads the .safetensors file. Failed requests are not charged. If the request succeeds but the LoRA seems to do nothing, check it was trained for the endpoint you called.

Qwen Image 2512 LoRA API: Uncensored NSFW Images, Your LoRAs

alibaba/qwen-image-2512-lora/edit

Qwen Image 2512 LoRA image editing runs Qwen-Image-Edit-2511 on one to three source images with up to three of your own LoRAs, such as camera-angle or relighting. By default the result follows the first image's shape at about one megapixel.

From $0.03/image

Run model View API

Input

Image editing

Prompt420/5000

Reordering updates prompt references. Removing media also removes its references; review your prompt.

Images2/3

Drop files or media links, paste, or browse · Up to 3 files · Max 10 MB · JPG · PNG · WEBP · GIF

Picture 1
Picture 2

Aspect ratio

Style LoRAs1/3

Anything2Real

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

Animate

Edit

Split into layers

{
  "model": "alibaba/qwen-image-2512-lora/edit",
  "input": {
    "image_urls": [
      "https://cdn.spicyapi.ai/models/examples/inputs/e7184e32f47a78b0.webp",
      "https://cdn.spicyapi.ai/models/examples/inputs/368c1f5546a0fd3c.webp"
    ],
    "loras": [
      {
        "path": "https://huggingface.co/lrzjason/Anything2Real/resolve/main/anything2real_2601.safetensors",
        "scale": 0.85
      }
    ],
    "output_format": "jpeg",
    "seed": 42,
    "prompt": "Dress the woman from Picture 1 in the emerald velvet gown drawn in Picture 2, turned into a real photograph of real fabric: the same deep V neckline, the same thigh-high slit, the same long train pooling on the floor. Keep her face, her freckles, her auburn waves, her pose by the window and the sunlit Paris apartment from Picture 1. Natural skin texture, soft morning light on the velvet, editorial fashion photograph."
  }
}

Introduction

Qwen Image 2512 LoRA Edit changes one to three source images with Alibaba's Qwen-Image-Edit-2511, guided by your instruction and up to three of your own LoRA files. The result follows the first image's shape at about one megapixel and comes back as JPEG, PNG or WebP at one flat price, with no content filter.

Choose a mode by starting material

What you have	Mode
one to three images to change	Qwen Image 2512 LoRA Edit
a written description only	Qwen Image 2512 LoRA Text to image

Key features & highlights

Edit LoRAs, not only style LoRAs

Qwen-Image-Edit-2511 has a large set of community edit LoRAs: new camera angles, relighting, photo-realism and more. Each one turns a plain instruction into a specific, repeatable edit.

Identity carries over

Alibaba built 2511 to hold people consistent between input and output. In our tests the same man, outfit and room survived a camera move, and an illustrated singer kept her curls and dress as a photograph.

Uncensored with custom weights

The sources, the instruction and the result are never filtered. With an uncensored edit LoRA, our strongest content probe came back as written.

Request parameters

Request parameters for Qwen Image 2512 LoRA · Image Editing
Parameter	Required	Description
`prompt`	Yes	Say what each picture contributes and what should change. Some edit LoRAs expect a fixed phrase; the model card gives it. at least 1 characters · at most 5000 characters
`image_urls`	Yes	One to three source images, addressed in the prompt as Picture 1, Picture 2 and Picture 3. Without `aspect_ratio` the result follows the first image's shape. at least 1 items · at most 3 items
`aspect_ratio`	No	Optional. Leave it empty and the result follows the first source image at about one megapixel. `1:1` · `16:9` · `9:16` · `4:3` · `3:4` · `3:2` · `2:3` · `21:9` · `9:21`
`loras`	No	Up to three {path, scale} entries trained for Qwen-Image-Edit-2511; scale 0 to 2 on this endpoint. Default: `[]` at most 3 items
`output_format`	No	JPEG by default, PNG or WebP if you need them. `jpeg` · `png` · `webp` · Default: `jpeg`
`seed`	No	Keeps edits comparable while you change the instruction. `≥ 0` · `≤ 2147483647`

Pricing

$0.03 / image

Billed per image at one flat price, whether you send one, two or three sources. LoRAs add no separate charge.

Use cases

Product and character angles

One photo becomes a set of views for a listing, a lookbook or a character sheet. Run the multiple-angles LoRA several times with different view phrases and the same seed to build the set.

Illustration to photograph

Turn concept art or a drawn character into a believable photo that keeps the design: hair, clothing and pose carry over while the rendering becomes photographic.

Uncensored edits

Uncensored edit LoRAs apply to your own references with no review and no filter on the result, at the same flat price as any other edit.

Explore key use-case tutorials

Prompt guide & pro tips

Use the LoRA's phrase exactly

Edit LoRAs are often trained on a fixed wording. The multiple-angles LoRA expects "" and then the view, for example " front-left quarter view eye-level shot medium shot"; copy the format from its model card.

Name every source

With two or three images, say what each one contributes: Picture 1 is the person, Picture 2 is the outfit, Picture 3 is the room. The result takes its shape from the first image, so put the main subject there.

Stay at or below 1 first

The scale limit here is 2, but most edit LoRAs are made for 1. Raise it only when the effect is too weak, in steps of 0.25, and keep the seed fixed while you compare.

Say what must not change

When the edit should be narrow, add what stays: keep her face, keep the dress, keep the window light. Saying it outright gives the model something to hold on to.

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": "alibaba/qwen-image-2512-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"
  }
}'

# 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.

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

Bigger frames for your LoRAs, and edits that can move the camera

Qwen Image 2512 LoRA runs two of Alibaba's open models with up to three of your own LoRA files per request: Qwen-Image-2512 for text-to-image, at up to 1536 pixels on the long side, and Qwen-Image-Edit-2511 for editing, from one to three source pictures. The two take different LoRAs, and getting that right is the first step. On a 20-billion-parameter base, style LoRAs keep faces, hands and small objects together; edit LoRAs can re-shoot a photo from a new angle or turn a drawing into a photograph. Every image costs the same, and nothing is filtered.

Open the playground

Bigger frames for your LoRAs, and edits that can move the camera

Text to image: Qwen-Image-2512, 20B parameters; nine ratios, up to 1536 px on the long side
Editing: Qwen-Image-Edit-2511; 1 to 3 sources, output follows the first at about one megapixel
LoRAs per request: Up to 3, each {path, scale}; scale 0 to 4 on text to image, 0 to 2 on editing
Which LoRAs: Trained for that endpoint's own base; other Qwen-Image versions do little
Price: One flat price per image, whatever the size or the number of sources
Content filter: None; both endpoints rendered our strongest content probe with a LoRA trained for it

Two bases: Match each LoRA to its endpointText-to-image and editing run different models, so every LoRA is trained for one or the other. An edit LoRA on text-to-image, or the reverse, loads and then does little. Keep the two lists apart in your configuration and the most common mistake disappears.
Frame size: More pixels than a one-megapixel frameText-to-image renders 1328 x 1328 at 1:1, 1536 x 864 at 16:9 and 1472 x 1104 at 4:3, at the same price as every other ratio. Editing returns about one megapixel in the shape of the first source.
Limits: Three sources and a scale of 2 when editingThe editing endpoint takes at most three pictures, caps each LoRA scale at 2 and has no aspect ratio choice: the result follows the first image. Weights made for other Qwen-Image versions, including the base behind Qwen Image 2.1 LoRA, do not transfer reliably.

Working with LoRAs on Qwen Image 2512 LoRA

Stylised, with the detail intact

A storybook-anime LoRA and an outdoor hot spring at dusk: two men resting together, steam, lanterns and a soft pastel palette. The style belongs to the LoRA, and faces and hands hold because the base is large. Name the style words the LoRA was trained on and write the scene in full sentences.

Move the camera with an edit LoRA

The multiple-angles LoRA on the editing endpoint re-shot one photo from a front-left quarter view: the same man, the same clothes, the same hotel window. It expects its own fixed phrase, "<sks>" followed by the view, which its model card spells out. Most edit LoRAs are made for a scale of 1, and this one ran at 1.

Name every source when you send more than one
With two or three pictures, say what each one contributes: Picture 1 is the person, Picture 2 is the room. The result follows the shape of the first picture, so put the one whose framing you want first. One, two or three sources cost the same.
Drawings into photographs
A photo-realism edit LoRA at 0.85 turned an illustrated singer into a photograph with the same curls, the same green velvet and the same vintage microphone. A plain instruction was enough; the LoRA supplied the realism.
Links that download the file
Each file is fetched when the request runs, so a link that needs a login fails the request, since gated repositories answer with HTTP 401, and so does a link to a web page instead of the file. Use a public URL to the .safetensors file. Failed requests are not charged.

Who uses Qwen Image 2512 LoRA

Illustrators and story apps
One hand-drawn look at a large size
Watercolour, storybook and pixel-art LoRAs give a series one hand-made look, and the frame goes up to 1536 pixels on the long side for covers and banners. Fix the seed per layout and change only the scene.
Game and retro art
Pixel art with the detail kept
A pixel-art LoRA triggered by "Pixel Art" at 16:9, where 2512 renders 1536 x 864: a neon arcade, rows of machines, two people mid-conversation. The large base keeps faces and small objects together inside the pixel grid.
Studios and concept artists
From drawing to photograph
Send an illustration to editing with a photo-realism LoRA at about 0.85 and a plain instruction. Our illustrated singer came back as a photograph with the same curls, dress and microphone, ready to serve as a reference or a finished frame.
Platforms offering community LoRAs
Stills and edits, uncensored
Community LoRAs of every kind load without review and neither endpoint filters its output; both rendered our strongest content probe as written. One flat price per image, whatever the size or the number of sources, keeps the cost of each user request predictable.

From LoRA to finished image with Qwen Image 2512 LoRA

Decide which endpoint the LoRA belongs to
Its model card names the base: Qwen-Image-2512 LoRAs go to text-to-image, Qwen-Image-Edit-2511 LoRAs go to editing.
Copy the direct file link
Use a public URL that downloads the .safetensors file. Gated repositories answer with HTTP 401 and the request fails.
Write in the LoRA's own format
Trigger phrase first for style LoRAs. Some edit LoRAs expect a fixed phrase, such as "<sks>" followed by the view you want.
Keep each scale in range
Start at 1. Text-to-image accepts up to 4 per LoRA; editing caps each one at 2.
Order your sources
For edits, put the picture whose framing you want first, name each one in the prompt as Picture 1, 2 or 3, and fix the seed while you adjust the instruction.

Qwen Image 2512 LoRA prompts worth copying

A storybook series line

For a storybook-anime LoRA on text-to-image. The prompt opens with the style words the LoRA was trained on, then the scene.

storybook anime illustration of two old friends sharing a picnic under a cherry tree at dusk, paper lanterns in the branches, a thermos and two cups on the blanket, soft pastel palette

A new camera angle
On the editing endpoint with the multiple-angles LoRA at scale 1 and one source photo. The phrase is the LoRA's own format; only the view changes.
```
<sks> front-left quarter view eye-level shot medium shot
```

Two sources, one instruction

On the editing endpoint with two pictures. Say what each one contributes; the result takes the shape of Picture 1.

Picture 1 is the woman. Picture 2 is the room. Place her in the room, seated at the window table in the same grey knit sweater, soft morning light from the left

Workbench

Input

Output

Introduction

Choose a mode by starting material

Key features & highlights

Edit LoRAs, not only style LoRAs

Identity carries over

Uncensored with custom weights

Request parameters

Pricing

Use cases

Product and character angles

Illustration to photograph

Uncensored edits

Prompt guide & pro tips

Use the LoRA's phrase exactly

Name every source

Stay at or below 1 first

Say what must not change

API quick start

Related models and next tasks

Bigger frames for your LoRAs, and edits that can move the camera

Working with LoRAs on Qwen Image 2512 LoRA

Stylised, with the detail intact

Move the camera with an edit LoRA

Name every source when you send more than one

Drawings into photographs

Links that download the file

Who uses Qwen Image 2512 LoRA

One hand-drawn look at a large size

Pixel art with the detail kept

From drawing to photograph

Stills and edits, uncensored

From LoRA to finished image with Qwen Image 2512 LoRA

Decide which endpoint the LoRA belongs to

Copy the direct file link

Write in the LoRA's own format

Keep each scale in range

Order your sources

Qwen Image 2512 LoRA prompts worth copying

A storybook series line

A new camera angle

Two sources, one instruction

More Qwen Image 2512 LoRA questions

How is this different from Qwen Image 2.1 LoRA?

Can I use uncensored LoRAs?

Which LoRAs work on editing?

How big are the images?

Does the number of edit sources change the price?

Why did my LoRA request fail?

All Qwen Image models11

Related comparisons

LoRA guide