Which LoRAs work here?

LoRAs trained for FLUX.1 [dev], which is the largest group of community image LoRAs. Pass a direct download link to the .safetensors file. SDXL, Z-Image or Qwen-Image LoRAs will not transfer.

Can I use uncensored LoRAs for NSFW images?

Yes. The weights are not reviewed and the output is not filtered. In our tests unfiltered LoRAs rendered, but at our strongest probe level the model held back part of the scene three times out of three.

Can I use the images commercially?

Black Forest Labs' FLUX.1 [dev] license lets you use generated images for personal, scientific and commercial purposes. The license of each LoRA you load is set by its author, so check its model card.

What does guidance do?

It sets how strictly the image follows the prompt. The default 3.5 suits most LoRAs; realism LoRAs often look more natural a little lower, around 2 to 3.

How big are the images?

About one megapixel. 1:1 is 1024 x 1024, 16:9 is 1344 x 768 and 21:9 is 1536 x 640. Every ratio costs the same.

Is image editing available?

Yes. Image editing takes one source image and a strength from 0 to 1, and an optional black-and-white mask redraws only the white area while the rest stays as it was. Up to three FLUX LoRAs load there too, at the same flat price.

FLUX.1 Dev LoRA API: Uncensored NSFW Images, Any FLUX LoRA

black-forest-labs/flux-1-dev-lora/edit

FLUX.1 Dev LoRA image editing from Black Forest Labs' FLUX.1 [dev], with up to three of your own LoRA files per request. A strength control sets how far the result moves from one source image, and a mask can limit the redraw to one area.

From $0.018/image

Run model View API

Input

Image editing

Prompt186/5000

Input image1/1

Drop files or media links, paste, or browse · 1 file · Max 10 MB · JPG · PNG · WEBP · GIF

Transformation strength

Style LoRAs0/3

Mask1/1

Drop files or media links, paste, or browse · 1 file · Max 10 MB · JPG · PNG · WEBP · GIF

Prompt adherence

Sampling steps

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": "black-forest-labs/flux-1-dev-lora/edit",
  "input": {
    "strength": 0.95,
    "guidance_scale": 3.5,
    "steps": 28,
    "output_format": "jpeg",
    "seed": 7,
    "prompt": "a black satin slip dress with thin spaghetti straps, bare shoulders and collarbones, a low neckline, her right hand gripping the chrome microphone stand in front of her, warm stage light",
    "mask_url": "https://cdn.spicyapi.ai/models/examples/inputs/e9c801386cf8fcd8.png",
    "image_url": "https://cdn.spicyapi.ai/models/examples/inputs/616a1d1536981fed.jpeg"
  }
}

Introduction

FLUX.1 Dev LoRA Image editing changes one source image on Black Forest Labs' FLUX.1 [dev], guided by your prompt and up to three of your own LoRA files. Without a mask it restyles the whole frame, and a strength value from 0 to 1 decides how far it moves from the source. With a black-and-white mask it redraws only the white area and keeps the rest. The result comes back as JPEG, PNG or WebP at one flat price, and nothing is filtered.

Choose a mode by starting material

What you have	Mode
a picture to change, whole or in part	FLUX.1 Dev LoRA Image editing
a written description only	FLUX.1 Dev LoRA Text to image

Key features & highlights

Masked inpainting

Paint the area to change white and everything else black. In our tests a mask over the upper body of a studio portrait changed only that area while the face, the hair and the trousers stayed as they were.

Strength you can reason about

The default 0.8 keeps pose and layout; 0.95 lets the model redraw nearly everything, the face included. With a mask, 0.9 to 1 makes sure the white area really changes.

Uncensored with custom weights

The source, the prompt, the weights and the result are never filtered. At strength 0.95 a studio portrait was edited as our strongest test asked, with and without an unfiltered LoRA.

Request parameters

Request parameters for FLUX.1 Dev LoRA · Image Editing
Parameter	Required	Description
`prompt`	Yes	Describe the whole image you want back. With a mask, describe what should appear in the white area. at least 1 characters · at most 5000 characters
`image_url`	Yes	The picture to edit. The result keeps its shape at about 0.7 megapixels: in our tests a 1024 x 1536 source came back as 672 x 1008.
`mask_url`	No	Optional. A black-and-white image the same size as the source: white areas are redrawn from the prompt, black areas are kept as they are.
`strength`	No	How far to move from the source, 0 to 1 (default 0.8). With a mask, use 0.9 to 1 so the white area really changes. `≥ 0` · `≤ 1` · Default: `0.8`
`loras`	No	Up to three {path, scale} entries trained for FLUX.1 [dev]; scale 0 to 4, 1 is neutral. Use a public direct link to the .safetensors file. Default: `[]` at most 3 items
`guidance_scale`	No	How closely the result follows the prompt, 0 to 20 (default 3.5). Realism LoRAs often look more natural a little lower. `≥ 0` · `≤ 20` · Default: `3.5`
`steps`	No	Denoising steps, 1 to 50 (default 28). More steps take longer to generate. `≥ 1` · `≤ 50` · Default: `28`
`output_format`	No	JPEG by default, PNG or WebP if you need them. Does not change the price. `jpeg` · `png` · `webp` · Default: `jpeg`
`seed`	No	Keeps results comparable while you adjust strength or the mask; the same seed does not promise an identical image. `≥ 0` · `≤ 2147483647`

Pricing

$0.018 / image

Billed per image at one flat price. The mask, strength, steps and LoRAs add no separate charge.

Use cases

Change one part of a photo

Swap an outfit, a background or an object and keep the rest of the picture untouched, for product variants, wardrobe tests or portrait retouching.

Restyle with a community LoRA

Run a photo through one of the thousand-plus FLUX LoRAs to turn it into film stock, editorial gloss or illustration while the composition stays.

Prompt guide & pro tips

Describe the result, not the instruction

Write what the final image should contain. With a mask, describe only what belongs in the white area, such as "a black velvet top, soft studio light".

Feather hard mask edges

A hard rectangle leaves a visible seam where it meets the kept pixels. Soften the edge of the mask or follow the outline of the object you want to replace.

Start at 0.8, then move in steps of 0.05

If the source bleeds through, raise strength; if the face drifts, lower it. Keep the seed fixed while you compare so that strength is the only thing changing.

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": "black-forest-labs/flux-1-dev-lora/edit",
  "input": {
    "image_url": "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.

LORA IMAGE API

FLUX.1 Dev with your own LoRAs, from the biggest community library of any open image model.

Key numbers

LoRAs per call: 3; Up to three weight files per image, each with its own strength.
Community LoRAs: 1,000+; Hugging Face lists more than a thousand adapters built on FLUX.1 [dev].
Parameters: 12B; Black Forest Labs' rectified flow transformer, released with open weights.
Aspect ratios: 9; From 21:9 to 9:21 at about one megapixel, every ratio at the same price.
Guidance range: 0 – 20; 3.5 by default. Lower it for looser, more photographic results.
Content filter: None; Uncensored: no review of your LoRAs and no filter on the result.

Before you spend

all

Text-to-image and image editing, with an optional mask for inpainting. Weights must be trained for FLUX.1 [dev]; LoRAs for other bases do little. About one megapixel, one image per request. At our strongest content probe level the model held back part of the scene in all three tries even with an unfiltered LoRA, so it is not the pick for the strongest requests.

What it does

Two men in a red vinyl booth by a rain-streaked window at night, one in a leather jacket, the other in an open white shirt

Realism

Photographs, not renders

A Super Realism LoRA and a late-night booth by a rainy window: two men leaning into each other under a warm pendant lamp. Skin, fabric and window light read like a photograph.

Editorial

A magazine look in one trigger

A woman in a backless red satin gown on marble stairs, glancing over her shoulder in side light

"MJ v6" and a Midjourney-style LoRA give FLUX a glossy editorial finish: a backless red satin gown on marble stairs, lit from the side.

Cartoon

Swap the whole medium

A long-limb cartoon LoRA at 1.2 turns the same model into a pin-up illustrator: a red umbrella, a wet cobblestone street and exaggerated proportions.

Made with FLUX.1 Dev LoRA

Image editing with a mask over the dress, strength 0.95.
Image editing without a mask, strength 0.85.
Image editing without a LoRA, strength 0.75.
Text to image without a LoRA, guidance 3.5, 16:9.

Where it fits

The LoRA you already found
If the look you want lives on Hugging Face or a community hub as a FLUX LoRA, this is where it runs as a plain API call: paste the file link, set the scale, write the prompt.
Fashion and editorial stills
Realism and magazine-style LoRAs give you backless gowns on marble stairs and late-night booths by rainy windows with believable light and skin.
Stylised campaigns
Cartoon, pin-up and illustration LoRAs turn one prompt template into a consistent series across nine aspect ratios.
Uncensored stills
Unfiltered LoRAs load without review and the output is not filtered, for platforms whose users expect that freedom.

Model reference

FLUX.1 Dev LoRA API reference

FLUX.1 Dev LoRA API specifications

Endpoints: Text to Image and Image Editing (strength, optional inpainting mask), both with LoRA support
Base model: FLUX.1 [dev], Black Forest Labs' 12-billion-parameter rectified flow transformer
LoRA input: Up to 3 entries, each {path, scale}; path is a direct HTTPS URL to a .safetensors file
LoRA strength: 0 to 4 per entry (1 is the default)
Guidance: 0 to 20 (3.5 is the default)
Output size: About one megapixel: 1024 x 1024 at 1:1, 1344 x 768 at 16:9, 1536 x 640 at 21:9
Aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9, 9:21
Output formats: JPEG (default), PNG, WebP

How to call the FLUX.1 Dev LoRA API

Find the LoRA's trigger
Most FLUX LoRAs name a trigger phrase on their model card, such as "Super Realism" or "MJ v6". Put it at the start of the prompt, then describe the picture.
Tune guidance with realism LoRAs
Guidance 3.5 is the default. Realism LoRAs often look more natural between 2 and 3.5; very high values push towards a glossy, over-sharpened look.
Raise the scale for strong styles
Some style LoRAs need a scale above 1 to take over. Our cartoon example ran at 1.2; start at 1 and move in steps of 0.2.
Check the file is public
Use a direct link that downloads the .safetensors file. Gated repositories answer with HTTP 401 and the request fails before any image is made.

The short version

FLUX.1 Dev LoRA runs Black Forest Labs' FLUX.1 [dev] with up to three of your own LoRAs per request: the base with more than a thousand community LoRAs, nine aspect ratios at about one megapixel, a guidance control and one flat price per image. Realism, fashion, cartoon and unfiltered LoRAs load without review, and nothing is filtered.