How is this different from FLUX.2 Dev?

It runs the same FLUX.2 [dev] model with a LoRA slot: up to three of your own weight files per request, in text to image and in editing. If you do not need custom weights, FLUX.2 Dev has its own page.

Do FLUX.1 LoRAs work here?

No. In our test a FLUX.1 [dev] LoRA loaded without an error and the image came back pixel for pixel identical to the run without it. FLUX.2 [dev] is a different architecture, so use files trained for it.

Can I use uncensored LoRAs for NSFW images?

Yes. Weights are not reviewed and results are not filtered. The base model already draws NSFW content up to the Strong level; in our tests a popular unfiltered LoRA still did not take it past the Very strong level, where results stayed implied.

Why did my LoRA request fail?

Usually the link needs a login (gated Hugging Face files answer with HTTP 401) or points to the model page instead of the file. Use a public URL that downloads the .safetensors file itself. Failed requests are not charged.

How many reference images can I use?

One to three in Image Edit, in the order your prompt mentions them. Write "the man from image 1" and "the woman from image 2" when you combine people, and ask to keep their faces.

What do 1k and 2k mean?

1k is about one megapixel, for example 1024 x 1024 or 1344 x 768. 2k is 2048 pixels on the long side, for example 2048 x 1152 at 16:9. Both cost the same; 2k usually takes around ten seconds longer.

One flat price per image, shown on each endpoint page. The aspect ratio, the resolution tier, the number of LoRAs and the number of reference images do not change it. Image Edit has its own price.

FLUX.2 Dev LoRA API: Uncensored NSFW Images, Your LoRAs

black-forest-labs/flux-2-dev-lora/text-to-image

Text to image on FLUX.2 [dev], Black Forest Labs' 32-billion-parameter open-weights model, with up to three of your own LoRAs per request. Nine aspect ratios at 1k or 2k, one flat price, and no filter on prompts, weights or results.

From $0.022/image

Run model View API

Input

Text to image

Prompt560/5000

Aspect ratio

Resolution

Style LoRAs1/3

Historic Color FLUX2

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-2-dev-lora/text-to-image",
  "input": {
    "aspect_ratio": "3:2",
    "resolution": "2k",
    "loras": [
      {
        "path": "https://huggingface.co/AlekseyCalvin/Historic_Color_FLUX2_v1/resolve/main/Flux2Hst_000002600.safetensors",
        "scale": 0.9
      }
    ],
    "seed": 4101,
    "prompt": "HST photo, a vintage colour photograph inside the dining car of an overnight train at night. An adult Black woman in her early thirties in a backless emerald satin evening gown sits alone at a table set with white linen and silver, glancing back over her bare shoulder at the camera with a slow, knowing smile. Small fringed table lamps glow tungsten amber down the length of the car, a man in a white dinner jacket sits out of focus two tables behind her, cold blue night beyond the glass. 50mm lens, shallow depth of field, rich saturated colour, fine grain."
  }
}

Introduction

FLUX.2 Dev LoRA Text to image draws one image from a written prompt on FLUX.2 [dev], Black Forest Labs' 32-billion-parameter open-weights model, with up to three of your own LoRA files loaded. Pick one of nine aspect ratios and the 1k or 2k tier. The platform adds no content filter to the prompt, the weights or the image.

Choose a mode by starting material

What you have	Mode
a written description	FLUX.2 Dev LoRA Text to image
one to three photos to change or combine	FLUX.2 Dev LoRA Image Edit

Key features & highlights

FLUX.2 LoRAs by direct link

Pass the direct link to a .safetensors file trained for FLUX.2 [dev] and a scale. Hugging Face and Civitai download links both worked in our tests; nothing is uploaded to us in advance and the file is not reviewed.

Three at once

Up to three LoRAs per request, each with its own scale. In our test a painted-interior style at 0.8, a vintage colour-photo style at 0.5 and an impressionist style at 0.4 loaded together without an error.

2k without a surcharge

Nine aspect ratios from 21:9 to 9:21 at 1k (about one megapixel) or 2k (2048 pixels on the long side). Every one of the 18 sizes came back at exactly the requested dimensions, and both tiers cost the same.

No content filter

Every level we test except Very strong rendered without a refusal, Maximum included. Very strong requests came back toned down, also with an unfiltered LoRA loaded.

Request parameters

Request parameters for FLUX.2 Dev LoRA · Text to Image
Parameter	Required	Description
`prompt`	Yes	Put the LoRA's trigger word first, then describe subject, place, light and lens in sentences, up to 5,000 characters. There is no negative prompt. at least 1 characters · at most 5000 characters
`aspect_ratio`	No	Nine options from 21:9 to 9:21; 1:1 by default. The shape does not change the price. `1:1` · `16:9` · `9:16` · `4:3` · `3:4` · `3:2` · `2:3` · `21:9` · `9:21` · Default: `1:1`
`resolution`	No	1k (default) is about one megapixel, for example 1024 x 1024 or 1344 x 768. 2k is 2048 pixels on the long side, for example 2048 x 2048 or 2048 x 1152. `1k` · `2k` · Default: `1k`
`loras`	No	Up to three {path, scale} entries trained for FLUX.2 [dev]; scale 0 to 4, 1 is the default and 0.5 to 1.2 the useful range. Use a public direct link to the .safetensors file. Default: `[]` at most 3 items
`seed`	No	A whole number from 0 to 2,147,483,647. Keep it fixed while you compare LoRAs or scales; it does not guarantee the exact same pixels on every run. `≥ 0` · `≤ 2147483647`

Pricing

$0.022 / image

Billed per image at one flat price. The aspect ratio, the 1k or 2k tier and loading LoRAs do not change it.

Use cases

A house style for a series

Load one style LoRA, keep its trigger word at the front and run a list of prompts: every image shares the look.

Cinematic stills

Write the lens, light and colour grade and the base model alone returns film-like frames, including 21:9 and 9:21 shapes at 2k.

Testing a LoRA you trained

Run your own FLUX.2 [dev] LoRA at full quality, with up to two more files stacked, without hosting a GPU.

Explore key use-case tutorials

Prompt guide & pro tips

Trigger first, then the shot

Start with the LoRA's trigger word, such as "HST photo" or "hmrsh", then describe subject, wardrobe, place, light and lens in sentences.

Check that the LoRA applied

Run the same prompt and seed with and without the file. A style LoRA should be obvious at a glance; identical images mean the file was trained for another model.

Stay near strength 1

0.5 to 1.2 covered everything useful in our tests. At 2 the image turned soft and smeared, and at 4 it dissolved into blur.

Draft at 1k

Settle the prompt and seed at 1k, then switch to 2k for the final. The price is the same; 2k usually takes about ten seconds longer.

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-2-dev-lora/text-to-image",
  "input": {
    "aspect_ratio": "1:1",
    "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?

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.2 Dev LoRA API at a glance

FLUX.2 [dev] with your own LoRAs: photoreal stills and edits in the style you choose, and nothing filtered.

Parameters: 32B; Black Forest Labs' FLUX.2 [dev], released with open weights on 25 November 2025.
LoRAs per call: 3; Up to three weight files per image, each with its own strength from 0 to 4.
Long side at 2k: 2048 px; 1k is about one megapixel; 2k reaches 2048 pixels on the long side, at the same price.
Reference images: 3; Editing takes one to three photos. In our test three travellers from three portraits sat at one table.
Aspect ratios: 9; From 21:9 to 9:21 on both tiers; editing also offers auto.
Content filter: None; Uncensored: no review of prompts, LoRA files or results, and no refusals in our tests.

Results

What you can make with FLUX.2 Dev LoRA

Edit + LoRA
A daytime photo, re-lit for night
Image Edit with a vintage colour-photo LoRA at 0.9: the traveller keeps her face while her coat becomes a burgundy velvet dress and the compartment turns to lamplight.
Image Edit with a vintage colour-photo LoRA at 0.9, auto shape.
Style
Saturated 1950s film colour on request
Text to image with the same LoRA and its trigger first: a dining car at night in rich vintage colour, 3:2 at the 2k tier.
Text to image with a vintage colour-photo LoRA at 0.9, 3:2 at 2k.
Reference images
Two photos, one scene
Two separate portraits in, one sleeper berth out: both faces carried over into a new late-night scene at 16:9, without a LoRA.
Image Edit from two photos without a LoRA, 16:9 at 2k.

More clips made with FLUX.2 Dev LoRA

Image Edit from one photo without a LoRA, 3:4 at 1k.
Text to image without a LoRA, 21:9 at 2k.
Text to image without a LoRA, 2:3 at 2k.
Image Edit with a painted-interior LoRA at 1, auto shape.
Text to image with a painted-interior LoRA at 0.85, 3:4 at 2k.
Text to image with a vintage colour-photo LoRA at 0.8, 4:3 at 2k.

Content test

Is FLUX.2 Dev LoRA uncensored? What we tested

We ran each level on our live setup and checked every clip frame by frame. A pass means the content was actually on screen, not just that the request went through.

Content levels tested on FLUX.2 Dev LoRA, by mode
Level	Image editing	Text to image
MildThe lightest prompts in the test	RendersTested Oct 9, 2026	RendersTested Oct 9, 2026
ModerateWhere filtered models start to refuse	RendersTested Oct 9, 2026	RendersTested Oct 9, 2026
StrongThe middle of the scale	RendersTested Oct 9, 2026	RendersTested Oct 9, 2026
Very strongWhere most models start to hold back	Not tested yet	Not tested yet
MaximumThe top of the scale	RendersTested Oct 9, 2026	RendersTested Oct 9, 2026

Renders: on screen as written. Softened: generated, but milder than what was written. Refused: rejected. Not tested yet: we have not run that level on that mode.

How each mode behaves

All modes: Use LoRAs trained for FLUX.2 [dev]. A FLUX.1 LoRA loads without an error and changes nothing, and a link that needs a login (HTTP 401) or points to a web page instead of the file fails the request. Four LoRAs or a strength above 4 are rejected. There is no negative prompt, no guidance and no step control, and the result is a JPEG.
All modes: No filter is applied and requests are not refused; the limits are the model's own. Every level we test rendered except Very strong, Maximum included, while Very strong requests stayed implied, in both modes and with an unfiltered LoRA loaded.
Image editing: One to three source images. Prompts that only ask to remove or swap something sometimes returned the source almost unchanged; describing the finished picture worked. A request to turn the person in the photo into someone else kept the original subject.
All modes: Every person you describe must be an adult. You are responsible for the lawfulness of what you generate and publish, including the rights of anyone in a source image. Generated media is retained for 14 days and then deleted.

Use cases

What to build with the FLUX.2 Dev LoRA API

A house style for a whole set
Load one style LoRA, such as a 1950s colour-photo look or a muted painted-interior look, and every image in a series shares it at 1k or 2k.
Restyled and re-staged photos
Send a portrait, describe a new outfit, place and light, and add a style LoRA. The face stays recognisable while the whole frame takes the new look.
Several people, one frame
Give it up to three photos and describe a shared scene. In our test three travellers from three separate portraits sat at one dining-car table with their faces intact.
Uncensored stills with community weights
Unfiltered LoRAs load with no review and no filter on the result. On its own the model delivered the Mild, Moderate and Strong levels; the Very strong level came back toned down in our tests, also with an unfiltered LoRA loaded.

Prompts

FLUX.2 Dev LoRA prompt examples you can copy

Paste one into the playground as it is, then change one thing at a time.

Text to image
Emerald dining car, vintage colour
HST photo, a vintage colour photograph inside the dining car of an overnight train at night. An adult Black woman in her early thirties in a backless emerald satin evening gown sits alone at a table set with white linen and silver, glancing back over her bare shoulder at the camera with a slow, knowing smile. Small fringed table lamps glow tungsten amber down the length of the car, a man in a white dinner jacket sits out of focus two tables behind her, cold blue night beyond the glass. 50mm lens, shallow depth of field, rich saturated colour, fine grain.
Why it works. A vintage colour-photo LoRA at 0.9 with its trigger "HST photo" first, 3:2 at 2k. The LoRA brings the saturated 1950s film colour.
Try this prompt
Image editing
From the platform to the dining car
The same man, now at night in the dining car of the same train: the overcoat and scarf are gone, his white dress shirt is unbuttoned at the collar with the sleeves rolled to the forearm, and he leans on a table set with white linen under a small fringed tungsten lamp, giving the camera a slow, smouldering look. Polished walnut and brass behind him, cold blue night in the glass, warm cinematic light, film grain. Keep his face exactly as it is.
Why it works. One photo of a man on a foggy platform in, a dining car at night out: coat off, shirt open at the collar, same face. 3:4 at 1k, no LoRA.
This mode starts from your own image: add it in the playground before you run.
Try this prompt
Text to image
A swaying sleeper corridor
Anamorphic widescreen film still in the narrow wood-panelled corridor of a moving sleeper train at night. An adult Latino man in his mid thirties in a white shirt unbuttoned to the chest and an adult Scandinavian woman in her early thirties with a platinum-blonde bob and pale freckled skin, in a navy satin dress with a plunging neckline, are pressed together as they squeeze past each other, the train swaying, her palm flat on his chest, their faces an inch apart, eyes locked. Brass wall sconces throw warm tungsten pools along the varnished walls, the carpet runner recedes into darkness, motion blur in the black windows. 40mm anamorphic, teal shadows, warm highlights, film grain.
Why it works. No LoRA: the base model alone at 21:9 and 2k, with practical sconces and motion blur in the windows written into the prompt.
Try this prompt

Pricing

FLUX.2 Dev LoRA API pricing

One image starts at $0.022.

The playground shows the exact price for your settings before you run. Failed or timed-out tasks are usually not charged. Some models charge for content refusals as stated on their model page; check those billing terms before submitting.

See every price tier

Get started

How to use the FLUX.2 Dev LoRA API in three steps

Step 1
Use LoRAs trained for FLUX.2 [dev]
Pick files whose base model is FLUX.2 [dev]. In our test a FLUX.1 LoRA loaded without an error and changed nothing: same seed, identical pixels.
Step 2
Put the trigger word first
Most style LoRAs expect a trigger at the start of the prompt, such as "HST photo" for a vintage colour-photo look or "hmrsh" for a quiet painted interior. Then describe subject, place, light and lens in full sentences.
Step 3
Keep strength between 0.5 and 1.2
In our tests 0.5 already shifted the look and 1 gave the full style. At 2 the picture went soft and smeared, and at 4 it dissolved into blur.

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.

Why SpicyAPI

Why run FLUX.2 Dev LoRA on SpicyAPI

No filter from us
We add no content filter of our own. What the model renders is what you get back.
Tested, not guessed
Content labels come from real generations we ran and checked, with the test date on the page.
Pay per result, in dollars
No subscription and no credits. Failed or timed-out tasks are usually not charged. Some models charge for content refusals as stated on their model page; check those billing terms before submitting.
One key for every model
One API key and balance cover image, video, audio and chat models, so you can explore more models from the same account.

FAQ

FLUX.2 Dev LoRA API FAQ

About FLUX.2 Dev LoRA

FLUX.2 Dev LoRA puts Black Forest Labs' open-weights FLUX.2 [dev] behind your own LoRAs: text to image in nine aspect ratios at 1k or 2k, and editing from up to three photos, with up to three weight files per request. LoRAs load without review, and nothing is filtered.