Is FLUX Uncensored? What FLUX.1 Dev Will and Won't Generate

Many answers to "is FLUX uncensored" say yes, because the open weights ship without a content filter. The useful question is what the model draws when you ask for something adult, and that depends on how much you ask for.

Short answer: partly. On our API there is no content filter in front of FLUX.1 Dev LoRA, but it does not draw everything. It draws the first two levels of our ladder and the restraint probe as written. Without an adapter it softens the middle: a whole-figure request (L3) and a two-person scene (L4) come back milder than asked. With an adult LoRA loaded it produces advanced NSFW imagery (L3), but not the L4 two-person scene. If you need L4 every time, pick a different model; if you already own a FLUX LoRA, it runs here with no filter at $0.018 an image, text to image or editing.

Based on our September 2026 tests of FLUX.1 [dev] as we sell it, with every image judged by looking at it. FLUX.2 and FLUX 3 joined our catalogue in October 2026 and have their own section below.

Try FLUX.1 Dev LoRA text to image →

FLUX uncensored: how far each route goes

Five levels, each stronger than the one before. ✅ came back as written · ⚠️ toned down · ❌ refused. Three runs per cell unless stated.

• L1 suggestive imagery of an adult.
• L2 moderate NSFW imagery, the upper body only.
• L3 advanced NSFW imagery of one adult, the whole figure rather than the upper body.
• L4 a higher-intensity scene between two adults.
• L5 two separate extreme probes: a consensual restraint and impact-play scene between adults, and fictional gore.

The editing row is a separate batch (one upper-body request on a clothed portrait, four runs with different settings), so don't compare it with the text-to-image row. The FLUX.1 Dev LoRA test report lists one of these runs (strength 0.95, no LoRA, no mask) under L3.

Nothing was refused: every toned-down run was drawn, just weaker than the prompt. Uncensored on our model pages means we put no content filter of our own in front of the model or the weights you load.

What the images actually looked like

Where the toning down shows up. At L3 the no-LoRA frames put the woman in underwear. At L4 they showed two people embracing and nothing more. In the gore probe FLUX drew blood on the floor and left out the injury. This is not blur or mosaic: the model draws a milder picture than you asked for, so the status code won't tell you. The restraint probe was not held back.

What the adapter changes. In our FLUX.1 Dev tests, the adult LoRA enabled L3 content, but the L4 two-person scene remained softened. That is a practical improvement for L3 work, not evidence that an adapter removes every limit or identifies the cause of the unadapted results.

Editing pictures with FLUX

Image editing takes one source image, a strength from 0 to 1 (default 0.8) and an optional black-and-white mask. Edit only your own fictional characters or people who have given documented consent (see "Your part" below). The test request was an upper-body NSFW imagery edit on a clothed studio portrait:

Strength is the setting to change first. The default of 0.8 is the weaker setting: 0.85 fell short and 0.95 delivered, with or without a LoRA. The cost is identity: at 0.95 the model redraws nearly everything, the face included. Only the masked run kept the face and trousers unchanged. For a recurring character, mask the part you want to change and use 0.9 to 1. Without a mask, start at 0.8 and raise it by 0.05 until the change appears. With an adult LoRA, editing reaches the same ceiling as text to image: advanced NSFW imagery (L3), not the L4 scene.

The model page's masked-edit example, source first:

Loading an adult LoRA

FLUX.1 [dev] has a very large community library of LoRAs, adult styles included. On our API you paste a download link, set a scale and write the prompt. We don't review the weights.

With a community adult LoRA (Lustly Uncensored v1) at scale 1, the L4 probe, a higher-intensity scene between two adults, gave advanced NSFW imagery in all three seeds, clearly drawn, but never the two-person scene. In ladder terms, the adapter delivers L3 and not L4. Our FLUX NSFW LoRA guide lists seven popular files with the strength each author recommends and licence flags.

FLUX plus a LoRA fits advanced NSFW portraits and styled adult stills. It does not fit a brief that needs the L4 scene; for that, go elsewhere:

• Z-Image Turbo LoRA ($0.012 an image) needs no adapter: all 18 text-to-image runs of the ladder, including the L4 scene and both L5 probes, came back as asked (see its test report). It also drew the L4 probe with an adult LoRA.
• Qwen Image 2512 LoRA ($0.03) rendered the L4 probe as written with an adult LoRA, but without one its L4 runs were toned down, so bring a LoRA trained for its own base.

See the Qwen Image 2512 LoRA write-up for its editing route and the Z-Image test for three more Z-Image tiers that pass L3 with no adapter.

How to load a LoRA

1. Use a direct link. The path must be an HTTPS link that downloads the .safetensors file itself.
2. Match the base. The file must be trained for FLUX.1 [dev]. SDXL, Z-Image and Qwen LoRAs will not transfer.
3. Trigger phrase first. Most LoRA model cards name one, such as "Super Realism" or "MJ v6". Put it at the start of the prompt.
4. Scale. 0 to 4 per LoRA; 1 applies the file at the strength its author trained. Authors often recommend another value on the model card (0.8 for some adult files), so start there, and if a style is faint step up by 0.2.
5. Up to three LoRAs per image, each with its own scale, at no extra charge.
6. Gated repositories fail. A link that needs a login returns a "Failed to load LoRA" error (401) and no image is made or charged.

For the wider picture on hosted adapters, see our comparison of hosted image LoRA models and the guide to running your own LoRA.

Prompts that work on FLUX

FLUX reads long, specific prompts well. Write full sentences: who, what they wear, the light, the camera look.

Super Realism, two men in their thirties share a red vinyl booth by a rain-streaked window at 2am, one in a worn leather jacket with his arm along the seat behind the other, the other in an unbuttoned white shirt leaning into him, warm pendant lamp, two coffee cups, quiet intimate smiles, shallow depth of field.

MJ v6, editorial fashion photograph of a beautiful woman in a red satin gown with a plunging neckline and an open back, descending marble stairs and glancing over her bare shoulder, dramatic side light, glossy magazine look.

FLUX keeps clothing and light first and drops pose clauses: in the booth picture the arm along the seat is missing, and in the gown picture the woman faces the camera instead of descending the stairs. Put the one pose you care about early and drop the rest.

Lower guidance from the default 3.5 towards 2.5 if a realism LoRA looks waxy or over-sharpened. There is no negative-prompt field, so describe what is in the frame rather than what to leave out. Leave signs and screens out: FLUX writes misspelled words on them.

Where FLUX trips up, and the fix

• L3 and L4 come back milder than asked without an adapter, and L4 never appears even with one. Fix: load an adult LoRA for advanced NSFW portraits; for L4 use Z-Image Turbo LoRA.
• Gore is implied, not shown. Fix: use a model that draws injury if the wound is the point.
• A LoRA for another base does almost nothing. The job still succeeds and the image looks fine, so you may blame the LoRA. Fix: check the model card says FLUX.1 [dev].
• A gated LoRA link fails. Fix: host a direct public link to the file.
• Edits that won't take. Fix: add a mask and use 0.9 to 1, or raise strength in steps of 0.05.

Why we don't offer FLUX Kontext dev LoRA

An adult editing request on FLUX.1 Kontext dev LoRA came back as the original input image, unchanged, with a completed job and no error. A silent copy of your input looks like a success, so Kontext is not in our catalogue. For FLUX editing, use FLUX.1 [dev] with strength and mask, above, or the FLUX.2 models in the next section.

FLUX.2 and FLUX 3

Since 8 October 2026 our catalogue also has six FLUX.2 models, each with text to image and editing, and FLUX 3 for video. Everything above was tested on FLUX.1 [dev]; the newer models fall into two groups.

• Moderated by Black Forest Labs: FLUX.2 Max, FLUX.2 Pro, FLUX.2 Flex and FLUX 3. L1 comes back as asked. L2 and L3 usually come back covered up, with a sheet, a robe or a crop at the waist, or are refused; L4 is refused, and edits of photos of people are refused more often than text to image. FLUX 3 is stricter still: only L1 came through, plus gore in text to video. Refused requests are not charged.
• Open weights, no moderation: FLUX.2 Dev, FLUX.2 Klein 9B and FLUX.2 Klein 4B. Nothing is refused: all 72 of our Klein 9B and 4B test runs completed. Dev drew L3, the restraint probe and gore as written. The Klein models often add clothing on their own, 4B most of all. None of the three draws L4: it comes back as two people in bed and no more.

At 1k, text to image runs from $0.0093 an image on Klein 4B to $0.07 on Max; FLUX 3 video starts at $0.06 a second in draft quality. Each model's full readings are in its test report: Max, Pro, Flex, Dev, Klein 9B, Klein 4B and FLUX 3.

What it costs

Price is per image; mask, strength, steps and LoRAs add nothing. Each model page shows the live estimate before you submit.

Black Forest Labs' FLUX dev Non-Commercial License says you may use outputs for any purpose, commercial included, with conditions: you either run content filters or review outputs for unlawful or infringing content before you distribute them, and you disclose AI generation where required. The non-commercial limit applies to the model weights, not to the outputs. The licence of each LoRA you load is set by its author, and many FLUX LoRAs are non-commercial too, so check each model card.

Try it in the Playground, then run it from code

1. Open the FLUX.1 Dev LoRA text-to-image page and stay on the Playground tab in Form view. Sign in with a funded account to run.
2. Write the Prompt: the LoRA's trigger phrase first, then who, what they wear, the light and the camera look.
3. Under Style LoRAs, paste the direct file link into LoRA file and set Strength (1 is a normal start). Add item gives you up to three.
4. Pick an Aspect ratio. If a realism LoRA looks waxy, open Advanced parameters and lower Prompt adherence from 3.5. A Seed keeps two runs comparable, so run once with the LoRA and once after Remove item to see what it changed.
5. Press Run model.

To edit a picture, use the image-editing page: add the Input image, set Transformation strength (0.9 to 1 with a mask), and put the black-and-white file in Mask under Advanced parameters. The Playground request body is the same JSON the API takes, so a setup that works here can be copied into code.

This is the booth example from the model page, with its LoRA. Install the Python SDK (pip install spicyapi) and set SPICY_API_KEY:

import uuid
from spicyapi import SpicyClient, output_assets

client = SpicyClient()  # reads SPICY_API_KEY from the environment

task = client.create_task(
    model="black-forest-labs/flux-1-dev-lora/text-to-image",
    input_data={
        "prompt": "Super Realism, two men in their thirties share a red vinyl booth by a rain-streaked window at 2am, "
                  "one in a worn leather jacket with his arm along the seat behind the other, the other in an "
                  "unbuttoned white shirt leaning into him, warm pendant lamp, two coffee cups, quiet intimate "
                  "smiles, shallow depth of field.",
        "aspect_ratio": "3:2",
        "guidance_scale": 3.5,
        "seed": 7,
        "loras": [{
            "path": "https://huggingface.co/strangerzonehf/Flux-Super-Realism-LoRA/resolve/main/super-realism.safetensors",
            "scale": 1,
        }],
    },
    idempotency_key=str(uuid.uuid4()),
)

final = client.wait_for_terminal(task["taskId"])
if final["state"] != "succeeded":
    raise RuntimeError(f"{task['taskId']} ended as {final['state']}: {final.get('errorCode')}")
print(output_assets(final)[0]["url"])  # copy the image to your own storage now

A generation took about 21 seconds; the same seed does not promise an identical picture. Creating, waiting, downloading, retries and error codes are in the tasks guide and the Python SDK reference; results expire, so see retention and delivery. Every option is on the FLUX.1 Dev LoRA API page.

Or let an Agent run it

When an assistant should run the generations, the CLI, the MCP server and the Agent Skill all read the live model schema, and the CLI and MCP show the price and ask you to confirm before a paid call.

# CLI: put the same input object in input.json
npx @spicyapi/cli tasks create --model black-forest-labs/flux-1-dev-lora/text-to-image --input-file input.json --wait

# MCP server: the command your assistant's MCP config runs (tools: spicyapi_task_create, spicyapi_task_wait)
npx --yes --package=@spicyapi/mcp spicyapi-mcp

# Agent Skill
npx skills add spicyapi-ai/spicy-skill

FAQ

Is FLUX uncensored?

Partly. FLUX.1 [dev] has no content filter, but L3, L4 and gore come back toned down without an adapter (see the table), and L4 does not appear even with an adult LoRA.

Does FLUX.1 Dev do advanced NSFW imagery?

Not by default: load an adult LoRA. In editing, an upper-body request worked at strength 0.95 or with a mask.

Can I use uncensored LoRAs with FLUX?

Yes: weights are not reviewed and output is not filtered. The file must be trained for FLUX.1 [dev] and be a direct public link (the steps are above).

Does FLUX have restrictions?

Not from us, but FLUX.2 Max, Pro, Flex and FLUX 3 run Black Forest Labs' own moderation (see the FLUX.2 section). Black Forest Labs' usage policy still bars sexual depictions of real people without their verified, documented and informed consent, and any sexual or harmful content involving minors, and says its restrictions apply to derivatives of the models.

Is FLUX.2 uncensored?

It depends on the model. FLUX.2 Max, Pro and Flex run Black Forest Labs' moderation: L1 comes through, L2 and L3 are usually covered up or refused, and L4 is refused. FLUX.2 Dev, Klein 9B and Klein 4B are open weights with no moderation and refuse nothing, though the Klein models often add clothing and none of the three draws L4. Black Forest Labs describes pre-release mitigations for the FLUX.2 family on the FLUX.2 dev model card.

Can FLUX edit an existing picture?

Yes, on FLUX.1 [dev], with a strength control and an optional mask (details above), and on all six FLUX.2 models, which take up to three source images. FLUX Kontext dev LoRA is not offered because it silently returned the input unchanged.

Your part

We add no extra content filter of our own. Everyone in an image must be an adult. Don't depict a real, identifiable person without their consent: Black Forest Labs' policy and licence prohibit it for sexual content. You're responsible for the law where you publish.

Make your first image

Create an account, follow the Playground steps above with a LoRA link and the first prompt. A hundred drafts cost $1.80.