Unlimited Open Source Models

Get Plan
Skip to main content
Developer guide

Uncensored AI API: Content Filter Controls by Endpoint

On ModelsLab's self-hosted image models the content filter is a request parameter, safety_checker, and it is off unless you turn it on. Third-party models always keep their provider's moderation. This page lists what each endpoint does, and the rules every request must follow. Adults 18+ and legal use only.

Get an API KeyTerms of Service

Last updated · By ModelsLab Engineering

What each endpoint does with content filtering

Endpoint familyFilter controlDefault
Self-hosted image: /api/v6/images/text2img (Stable Diffusion, SDXL, Flux, imported Civitai checkpoints)safety_checker true / false; safety_checker_type black, blur or sensitive_content_textOff
Self-hosted image: /api/v6/images/img2img, /api/v6/image_editing/qwen_editsafety_checker true / false; safety_checker_typeSend false explicitly
Third-party image: OpenAI, Google, xAI, ByteDance, Alibaba and others on /api/v7/imagesProvider moderation, always onOn
Self-hosted video: Wan 2.2, LTX 2.3Prompt keyword filter, always onOn
Third-party video: Kling, Seedance, MiniMax, Veo and othersProvider moderation, always onOn
Self-hosted chat: model uncensored-chat on /api/v7/llm/chat/completionsNo provider moderation layerOff
Third-party chat: every other model on /api/v7/llmProvider policy; a content refusal pauses /api/v7/llm for 48 hoursOn

Read from the API code on 2026-09-23. Third-party providers change their own moderation without notice; a refused request returns an error.

Content rules, before anything else

Turning a filter off does not change what is allowed. Every request must be legal where you and your users are, and the service is for adults 18 or older. You are responsible for what you generate, store and publish, and for age checks and moderation in your own product.

  • Never: anything involving minors, or anyone who appears to be a minor.
  • Never: images or text of real, identifiable people without their consent.
  • Never: non-consensual content, or content that promotes violence against real people.
  • Never: content that is illegal in your country or your users’ countries.
  • Breaking these rules can get an account suspended.

Uncensored image generation API

On the self-hosted image endpoints the filter is a parameter. `safety_checker: true` screens each output, and `safety_checker_type` decides what replaces a flagged image: `black`, `blur` or `sensitive_content_text`. On /api/v6/images/text2img the default is off. On img2img and the Qwen image edit endpoint, send `safety_checker: false` explicitly if you do not want screening.

This covers the open-source checkpoints ModelsLab runs on its own GPUs, including Civitai checkpoints and LoRAs you import. The $149/month Open Source Unlimited plan covers them without per-call charges; plans start at $21/month.

Set the filter per request

The same request with the filter off and on.

Filter off (Python)

Python
1import requests
2

3response = requests.post(
4    "https://modelslab.com/api/v6/images/text2img",
5    json={
6        "key": "YOUR_MODELSLAB_API_KEY",
7        "model_id": "sd-xl-v1-0-vae-fix",
8        "prompt": "a moody film-noir detective office, rain on the window",
9        "width": 1024,
10        "height": 1024,
11        "safety_checker": False,
12    },
13)
14

15print(response.json())

Filter on, blur flagged output (curl)

bash
1curl https://modelslab.com/api/v6/images/text2img \
2  -H "Content-Type: application/json" \
3  -d '{
4    "key": "YOUR_MODELSLAB_API_KEY",
5    "model_id": "sd-xl-v1-0-vae-fix",
6    "prompt": "a moody film-noir detective office, rain on the window",
7    "safety_checker": true,
8    "safety_checker_type": "blur"
9  }'

Uncensored LLM API

For chat, the self-hosted model `uncensored-chat`, listed as Custom LLM, runs on the same OpenAI-compatible endpoint as every other model, /api/v7/llm/chat/completions, at $0.30 per million input tokens and $1.20 per million output tokens. Because ModelsLab runs it, there is no third-party moderation layer in front of it. Your own system prompt and your own moderation decide what your product allows.

Third-party LLMs on /api/v7/llm keep their provider’s policies. If a provider refuses a request on content grounds, /api/v7/llm is paused for that account for 48 hours. The OpenAI-compatible endpoint at /api/uncensored-chat/v1/chat/completions stays available during that time.

Self-hosted chat model

OpenAI request format on /api/v7/llm.

OpenAI SDK (Python)

Python
1from openai import OpenAI
2

3client = OpenAI(
4    api_key="YOUR_MODELSLAB_API_KEY",
5    base_url="https://modelslab.com/api/v7/llm",
6)
7

8response = client.chat.completions.create(
9    model="uncensored-chat",
10    messages=[
11        {"role": "system", "content": "You are a candid writing assistant for an 18+ fiction app."},
12        {"role": "user", "content": "Outline a dark thriller in five chapters."},
13    ],
14)
15

16print(response.choices[0].message.content)

Models that always filter

Every third-party model applies its provider’s moderation, and ModelsLab cannot switch it off: OpenAI, Google, xAI, ByteDance, Kling, MiniMax, Alibaba and the rest. The self-hosted video models, Wan 2.2 and LTX 2.3, filter prompts by keyword before rendering. If your product needs unrestricted output, build it on the self-hosted image models and the self-hosted chat model above.

Before you build, read the Terms of Service. For plan details see pricing.

Get Expert Support in Seconds

We're Here to Help.

Want to know more? You can email us anytime at support@modelslab.com

View Docs

Yes, within limits. On ModelsLab's self-hosted image endpoints the `safety_checker` filter is a request parameter, off unless you turn it on, and a self-hosted chat model runs without a provider moderation layer. Third-party models always keep their provider's moderation. All use must be legal and by adults 18 or older.

On /api/v6/images/text2img, img2img and the self-hosted image editing endpoints, `safety_checker: true` screens the output and `safety_checker_type` decides what replaces a flagged image: `black`, `blur` or `sensitive_content_text`. On text2img it is off by default; send `safety_checker: false` explicitly on the other endpoints to be sure.

Every third-party model: the OpenAI, Google, xAI, ByteDance, Kling, MiniMax and Alibaba endpoints apply their own moderation, and a refused request returns an error. The self-hosted video models Wan 2.2 and LTX 2.3 also filter the prompt, so they are not suited to unrestricted video.

Yes. The self-hosted model `uncensored-chat` (listed as Custom LLM) runs on /api/v7/llm/chat/completions with the OpenAI request format, at $0.30 per million input tokens and $1.20 per million output tokens. Third-party LLMs on the same endpoint keep their provider's policies. If a provider refuses a request on content grounds, /api/v7/llm is paused for that account for 48 hours; the OpenAI-compatible endpoint at /api/uncensored-chat/v1/chat/completions stays available.

Self-hosted Stable Diffusion, SDXL and Flux checkpoints, including imported Civitai models, return output without a filter when `safety_checker` is off. You are responsible for what you generate and publish, and the prohibited-content rules on this page apply to every request.

Anything involving minors or anyone who appears to be a minor, real people without their consent, non-consensual content, content that is illegal where you or your users are, and anything that breaks the Terms of Service. Breaking these rules can get an account suspended.

No. The `safety_checker` parameter works on every plan, from $21/month. The $149/month Open Source Unlimited plan covers the self-hosted image models without per-call charges.