How is Base different from the distilled klein 9B?

Black Forest Labs step- and guidance-distilled klein 9B to run in four steps. Base is the same 9B model without that distillation: its makers describe it as having higher output diversity and recommend it for fine-tuning and LoRA training. Here it took two to five times longer per image, gave softer, more film-like pictures. Their reference settings run Base at 50 steps against four for the distilled build.

Do LoRAs trained on the distilled klein 9B work here?

In our tests, yes: a pixel-art LoRA trained on the distilled 9B redrew the scene on Base, and unfiltered LoRAs trained on Base also loaded on the distilled build. FLUX.1 and klein 4B LoRAs ran without an error and changed nothing.

My LoRA loaded but nothing changed. Why?

Three causes in our tests: the file was trained for another model (FLUX.1 or klein 4B), it was saved in diffusers PEFT format (keys ending in .default.weight), or it is a style LoRA whose look the prompt already asks for. None of these raise an error. Compare the same seed with and without the file.

Should I use this or FLUX.2 Klein 9B LoRA?

Use Base to check a LoRA you trained on Base, or for a softer, more varied look. Use the distilled build when speed matters: it renders in about ten seconds and gives a more polished photographic finish.

Can I use uncensored LoRAs for NSFW images?

Yes. LoRA files are not reviewed and the output is not filtered. On its own Base tends to crop or soften NSFW content and tone down the Very strong level; unfiltered LoRAs trained on Base removed that in our tests.

Why did my LoRA request fail?

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

Each endpoint has one flat price per image. The aspect ratio, 1k or 2k, the number of edit sources and loading LoRAs do not change it. Failed requests are not charged.

FLUX.2 Klein Base 9B LoRA API: Uncensored NSFW Images

black-forest-labs/flux-2-klein-base-9b-lora/text-to-image

FLUX.2 Klein Base 9B LoRA text to image runs Black Forest Labs' undistilled FLUX.2 [klein] Base 9B with up to three of your own LoRA files. Nine aspect ratios at 1k or 2k, softer and more varied than the distilled build, and no content filter.

From $0.024/image

Run model View API

Input

Text to image

Prompt469/5000

Aspect ratio

Resolution

Style LoRAs1/3

ANALOG REDMOND FLUXKLEIN9B

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-klein-base-9b-lora/text-to-image",
  "input": {
    "aspect_ratio": "2:3",
    "resolution": "2k",
    "loras": [
      {
        "path": "https://huggingface.co/artificialguybr/ANALOG-REDMOND-FLUXKLEIN9B/resolve/main/%5BFLUX.2.Klein%5DAnalog_Redmond.safetensors",
        "scale": 0.8
      }
    ],
    "seed": 101,
    "prompt": "AnalogRedmAF, Analog. Dawn inside a Victorian palm house: giant fronds, curved glass streaked with condensation, white wrought-iron ribs overhead, mist hanging in the first light. A woman in her mid-thirties with dark brown skin and close-cropped natural hair stands among the palms in an emerald silk robe loosely tied at the waist, her hair damp from the misting system, an alluring look toward the camera. 35mm film photograph, soft green and gold tones, fine grain."
  }
}

Introduction

FLUX.2 Klein Base 9B LoRA Text to image draws one image from a written prompt on Black Forest Labs' FLUX.2 [klein] Base 9B, the undistilled 9B model its makers recommend for fine-tuning and LoRA training, with up to three of your own LoRA files loaded. Choose one of nine aspect ratios at 1k or 2k. Base is slower than the distilled klein 9B and gives softer, more varied pictures. The platform adds no content filter.

Choose a mode by starting material

What you have	Mode
a written description and a LoRA	FLUX.2 Klein Base 9B LoRA Text to image
one to three images to change	FLUX.2 Klein Base 9B LoRA Image editing

Key features & highlights

The build LoRAs are trained on

A LoRA trained on klein Base 9B runs here on the model it was trained against, which is the cleanest place to judge it. LoRAs trained on the distilled klein 9B loaded as well in our tests; FLUX.1 and klein 4B LoRAs ran without an error and changed nothing.

Softer, more varied pictures

Without any LoRA, Base drew natural, film-like light and skin and more variety from seed to seed than the distilled build. Its makers run it at 50 sampling steps against four for the distilled build, which is why it takes longer.

Clean A/B checks

In our text-to-image tests a repeated seed returned identical pixels, so a pair of images with and without a LoRA shows exactly what the LoRA adds.

Request parameters

Request parameters for FLUX.2 Klein Base 9B LoRA · Text to Image
Parameter	Required	Description
`prompt`	Yes	Required, up to 5,000 characters. Put the LoRA's trigger phrase first, then describe subject, clothing, place, light and lens in sentences. at least 1 characters · at most 5000 characters
`aspect_ratio`	No	Nine shapes 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 1344 x 768 or 1024 x 1024; 2k puts 2048 pixels on the long side. 2k takes longer but costs the same. `1k` · `2k` · Default: `1k`
`loras`	No	Up to three {path, scale} entries trained for FLUX.2 [klein] 9B, Base or distilled; scale 0 to 4, 1 by default. Use a public direct link to the .safetensors file. Default: `[]` at most 3 items
`seed`	No	Optional, 0 to 2,147,483,647. In our text-to-image tests the same seed and inputs repeated the same pixels, so it isolates the effect of a LoRA or one word. `≥ 0` · `≤ 2147483647`

Pricing

$0.024 / image

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

Use cases

Checking a LoRA you trained

Run your file on Base at 1k or 2k and compare it against the same seed without it before you build on it or move it to the faster distilled build.

Natural stills

Portraits and interiors with soft light and believable skin, in nine shapes up to 2048 pixels, when you can wait half a minute per image.

Explore key use-case tutorials

Prompt guide & pro tips

Trigger first

Put the LoRA's trigger phrase at the start, then describe the picture in full sentences: subject, clothing, place, light and lens. There is no negative prompt, so state what you want to see.

One change per pair

Keep the seed fixed and change one thing at a time: the LoRA, its scale or one sentence. On Base the repeat is exact in text to image, so every difference you see comes from that one change.

Budget the time

About 20 seconds at 1k and around 35 at 2k in our tests, two to five times the distilled build. Send requests in parallel rather than waiting on each one, and use 1k while you are still choosing a scale.

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-klein-base-9b-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 Klein Base 9B LoRA API at a glance

FLUX.2 [klein] Base 9B, the undistilled build LoRAs are trained on, with up to three of your own LoRAs and no content filter.

LoRAs per call: 3; Up to three weight files per image, each with its own strength from 0 to 4.
Undistilled build: Full 9B; The full klein 9B model, without step or guidance distillation.
Klein 9B builds: Both; LoRAs trained on Base 9B or on the distilled 9B both loaded in our tests.
Typical render: 20–55 s; Median model time: about 20 s for 1k text to image, up to about 55 s for a 2k edit.
Largest size tier: 2k; 2048 pixels on the long side; 1k is about one megapixel. Both cost the same.
Content filter: None; No review of your LoRAs and no filter on the result.

Results

What you can make with FLUX.2 Klein Base 9B LoRA

Base, no LoRA
Soft light and natural skin
Without any LoRA, Base draws a round lily house at first light with mist on the water and a man kneeling on the rim, in the soft, film-like tones it favours, at 2048 by 1152.
Text to image without a LoRA, 16:9 at 2k.
Film-still LoRA
A grade from a LoRA, a scene from the prompt
A film-still LoRA trained on the distilled klein 9B, loaded on Base at 0.8: two people on a wrought-iron spiral stair among tree ferns, with the LoRA setting the haze and grade.
Text to image with a film-still LoRA at 0.8, 3:4.
Edit
From a crosswalk to a palm house
One street photo became the same woman in a rust silk dress among palm fronds at dawn, with a film-still LoRA setting the grade.
Edit of one street photo with a film-still LoRA, auto shape.

More clips made with FLUX.2 Klein Base 9B LoRA

Text to image with an analog-photo LoRA at 0.8, 2:3 at 2k.
Edit of one office photo with an analog-photo LoRA, 16:9.
Edit combining two portraits without a LoRA, 3:4 at 2k.
Text to image without a LoRA, 1:1.

Content test

Is FLUX.2 Klein Base 9B 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 Klein Base 9B 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	RendersTested Oct 9, 2026	RendersTested Oct 9, 2026
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 [klein] 9B, Base or distilled. FLUX.1 [dev] and klein 4B LoRAs loaded without an error and changed nothing in our tests, and so did a file saved in diffusers PEFT format. A link that needs a login or points to a web page fails the request. Up to three LoRAs, scale 0 to 4. There is no negative prompt and no step or guidance control; output is JPEG, one image per request.
All modes: There is no content filter. On its own the model holds back: in our tests it cropped results short of the Strong level, softened a request with a male subject, kept a Very strong scene implied, and in editing sometimes returned the source unchanged. With unfiltered LoRAs trained on Base loaded, every level we test rendered, Very strong included. A strong LoRA in an edit can replace the framing and the person, so lower its scale to keep the source.
All modes: It is slow next to the distilled klein 9B: about 20 seconds for a 1k image and 30 to 75 seconds for edits and 2k in our tests.
Image editing: Editing takes one to three sources. With auto the result keeps the first image's shape at about one megapixel; choose a ratio for the 1k or 2k tier.
All modes: Every person you describe or upload must be an adult. You are responsible for the lawfulness of what you generate and publish, including the rights of anyone shown in a source image. Generated media is retained for 14 days and then deleted.

Use cases

What to build with the FLUX.2 Klein Base 9B LoRA API

Checking your own LoRA
Train on klein Base 9B, then run the file here at full quality against the same seed without it, without hosting a GPU.
Varied, natural stills
Base gives softer light, natural skin and more variety from seed to seed than the distilled build.
Restyling photos
Editing keeps a person and changes clothes, place or medium, from a film grade to an inked illustration.
Your own LoRAs, unfiltered
Nothing is reviewed or filtered. Unfiltered LoRAs trained on Base are how you get the scenes the model holds back on its own.

Prompts

FLUX.2 Klein Base 9B 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 silk in the palm house
AnalogRedmAF, Analog. Dawn inside a Victorian palm house: giant fronds, curved glass streaked with condensation, white wrought-iron ribs overhead, mist hanging in the first light. A woman in her mid-thirties with dark brown skin and close-cropped natural hair stands among the palms in an emerald silk robe loosely tied at the waist, her hair damp from the misting system, an alluring look toward the camera. 35mm film photograph, soft green and gold tones, fine grain.
Why it works. An analog-photo LoRA at 0.8 with its trigger first, 2:3 at 2k. Compared with the same seed without it, the LoRA made the picture lighter and hazier with softer greens.
Try this prompt
Image editing
Denim to silk among the palms
Cinematic, Film Still. Keep the face, curly hair and smile of the woman in image 1. Move her into a glasshouse at dawn among huge palm fronds, condensation running down the glass, and change her outfit to a rust-coloured silk slip dress with a low back, her hair a little damp from the humid air. Soft golden haze.
Why it works. One street photo, auto shape, the film-still LoRA at 0.8. Her face, curls and smile stay; the denim becomes a rust silk dress, among palms at dawn.
This mode starts from your own image: add it in the playground before you run.
Try this prompt
Text to image
First light in the lily house
First light in a round glasshouse built around a pond of giant water lilies, mist rising from the water, wrought-iron columns and beads of condensation on the panes. A man in his early forties with South Asian features, a salt-and-pepper beard and a white shirt open at the collar with the sleeves rolled up kneels on the stone rim, one hand trailing in the water, looking up at the camera with a quiet smile. Cinematic wide shot, 35mm lens, pale gold and jade palette.
Why it works. No LoRA, 16:9 at 2k: Base on its own, with soft gold light, mist on the water and a natural pose.
Try this prompt

Pricing

FLUX.2 Klein Base 9B LoRA API pricing

One image starts at $0.024.

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 Klein Base 9B LoRA API in three steps

Step 1
Test the LoRA you trained
Base is the build klein 9B LoRAs are usually trained on, so a file you trained runs here the way it was trained. Run the same prompt and seed with and without it; in our text-to-image tests a repeated seed gave identical pixels.
Step 2
Distilled LoRAs load too
A pixel-art LoRA trained on the distilled klein 9B redrew our test scene here. FLUX.1 and klein 4B LoRAs, and a file saved in diffusers PEFT format, ran without an error and changed nothing.
Step 3
Start the scale at 0.8 to 1
0.5 gave a lighter version of a style in our tests, 1 the full look, and 4 broke the picture into blocks. Unfiltered LoRAs can change framing and faces; lower them if that happens.

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 Klein Base 9B 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 Klein Base 9B LoRA API FAQ

About FLUX.2 Klein Base 9B LoRA

FLUX.2 Klein Base 9B LoRA runs the undistilled klein 9B, the build LoRAs are trained on, behind your own LoRA files: text to image and editing at 1k or 2k and no content filter. It is the slower, more varied sibling of the distilled klein 9B.