Skip to main content

Uncensored Content & Policy-Tier Routing

AgentOS ships uncensored generation as a first-class feature. Any high-level API call that accepts a policyTier of 'mature' or 'private-adult' automatically reroutes to a curated uncensored model and disables the provider's default safety filter, so callers get working output without wiring bespoke model IDs for every surface.

This document covers:

  • Text generation (generateText, generateObject, streamText)
  • Image generation (generateImage)
  • Image editing (editImage)
  • The underlying model catalog and how to extend it

Policy tiers​

TierBehaviour
safeFamily-friendly. Default censored providers. Refuses sexual/violent content.
standardDefault for most sessions. Mild violence, romantic tension, no explicit content.
matureRoutes to uncensored catalog. Sexual content, graphic violence, dark themes.
private-adultRoutes to uncensored catalog. All legal fictional content.

Tiers are inclusive: every tier above safe allows the tiers below it. Two content categories remain prohibited at every tier (including private-adult): sexual content involving minors, and real-world instructions for weapons, explosives, or poisons.

Text — uncensored chat and structured output​

Set policyTier on any text call. The PolicyAwareRouter picks from the UncensoredModelCatalog (Hermes 3, Dolphin, MythoMax on OpenRouter) and bypasses the default OpenAI/Anthropic chain.

import { generateText, PolicyAwareRouter, createUncensoredModelCatalog } from '@framers/agentos';

// Option 1 — explicit router (most control)
const router = new PolicyAwareRouter(
  createUncensoredModelCatalog(),
  undefined,
  undefined,
  'private-adult',
);
const result = await generateText({
  router,
  system: 'You are a confident adult character named Cleopatra.',
  prompt: 'Describe the scene the way you want me to see it.',
});

// Option 2 — route through a tier-aware agent
import { agent } from '@framers/agentos';
const a = agent({
  instructions: 'You are Cleopatra, confident and direct.',
  router: new PolicyAwareRouter(createUncensoredModelCatalog(), undefined, undefined, 'mature'),
});
const reply = await a.generate('Tell me about the court intrigue.');

Catalog entries (text)​

Tier mature / private-adult routes to OpenRouter text models in this priority order (high quality first):

  1. nousresearch/hermes-3-llama-3.1-405b — flagship, follows system prompts reliably
  2. nousresearch/hermes-3-llama-3.1-70b — faster, same instruction-following
  3. cognitivecomputations/dolphin-mixtral-8x22b — fully uncensored MoE
  4. cognitivecomputations/dolphin3.0-llama3.1-8b — cheap tail-end fallback
  5. gryphe/mythomax-l2-13b — last-resort creative-writing model

Safe/standard tiers ignore the router and use the default provider chain.

Refusal-retry​

When a model still produces wellness-bot speak despite the tier (happens with some OpenRouter routing hops), PolicyAwareRouter does not retry on its own — refusal detection is the caller's responsibility. The wilds-ai CompanionOrchestrator wraps this into a runRefusalRetry helper that iterates the full uncensored catalog on detection; cooperating apps can use the same pattern via createUncensoredModelCatalog().getTextModels().

Image — uncensored generation and editing​

Set policyTier: 'mature' or 'private-adult' on either entry point:

import { generateImage, editImage } from '@framers/agentos';

// Generate an uncensored portrait
const portrait = await generateImage({
  prompt: 'Cleopatra in the palace, oil painting style',
  policyTier: 'mature',
  referenceImageUrl: 'https://cdn.example.com/cleopatra-anchor.png',
});

// Edit an existing avatar (e.g. outfit change) while preserving the face
const outfit = await editImage({
  image: existingAvatarBuffer,
  prompt: 'wearing formal court attire, seated on the throne',
  policyTier: 'mature',
  capabilities: ['face-consistency', 'img2img'],
});

Automatic behaviour when policyTier is set​

When policyTier is 'mature' or 'private-adult', the API:

  1. Routes through PolicyAwareImageRouter to pick a model from the uncensored catalog. generateImage infers 'face-consistency' when referenceImageUrl is supplied; editImage defaults to 'img2img' and honours explicit capabilities hints.
  2. Sets providerOptions.replicate.disableSafetyChecker = true so the community model's own NSFW filter does not silently veto the prompt.
  3. Initialises Replicate as the primary provider (Fal and stable-diffusion-local remain in the fallback chain).

Safe and standard tiers bypass all of this; they use whatever provider / model the caller passes in (or env-detected defaults).

Catalog entries (image)​

Tier mature / private-adult routes to Replicate community models:

ModelQualityContent permissionsCapabilities
lucataco/realvisxl-v4.0highgeneral, romantic, erotictxt2img, img2img, photorealistic
stability-ai/sdxlhighalltxt2img, img2img
lucataco/ip-adapter-faceid-sdxlmediumgeneral, romantic, erotictxt2img, img2img, face-consistency
zsxkib/instant-idmediumgeneral, romantictxt2img, face-consistency
lucataco/animate-diffmediumgeneral, romantic, violenttxt2img, video
stability-ai/stable-video-diffusionhighgeneral, romanticimg2video, video

The catalog is sorted by quality and filtered by capabilities. Call createUncensoredModelCatalog().getImageModels({ capabilities: ['face-consistency'] }) to inspect the filtered list at runtime.

Outfit / costume editing (wardrobe_preview)​

For the canonical "change a character's clothes while preserving their face" use case, pass a reference image and the face-consistency capability. The router picks lucataco/ip-adapter-faceid-sdxl, which is trained precisely for this pattern and allows erotic content.

const lingerie = await editImage({
  image: cleopatraAvatarBuffer,
  prompt: 'wearing elegant lingerie, seductive pose, Egyptian palace setting',
  policyTier: 'mature',
  capabilities: ['face-consistency', 'img2img'],
  strength: 0.75,
});

Without policyTier, this call routes to Flux Kontext Dev (or whatever the caller configured), which refuses the prompt at the model-provider level — no amount of disable_safety_checker changes that because Flux Kontext's moderation is baked into the model weights.

Extending the catalog​

createUncensoredModelCatalog() returns an immutable catalog. To add models, compose your own implementation of UncensoredModelCatalog:

import type { UncensoredModelCatalog, CatalogEntry } from '@framers/agentos';

const EXTRA_ENTRIES: CatalogEntry[] = [
  {
    modelId: 'your-org/your-nsfw-model',
    displayName: 'Custom Erotic SDXL',
    providerId: 'replicate',
    modality: 'image',
    quality: 'high',
    contentPermissions: ['general', 'romantic', 'erotic'],
    capabilities: ['txt2img', 'img2img', 'face-consistency'],
  },
];

function makeExtendedCatalog(): UncensoredModelCatalog {
  const base = createUncensoredModelCatalog();
  return {
    getTextModels: (f) => base.getTextModels(f),
    getImageModels: (f) => [...EXTRA_ENTRIES, ...base.getImageModels(f)],
    getPreferredTextModel: (t, i) => base.getPreferredTextModel(t, i),
    getPreferredImageModel: (t, caps) => {
      const preferred = EXTRA_ENTRIES.find(
        (e) => !caps || caps.every((c) => e.capabilities.includes(c)),
      );
      return preferred ?? base.getPreferredImageModel(t, caps);
    },
  };
}

const router = new PolicyAwareImageRouter(makeExtendedCatalog());

Wire the extended router into generateImage / editImage by passing provider and model resolved from router.getPreferredProvider(...) explicitly — the high-level APIs currently embed createUncensoredModelCatalog() as the default, but the routers themselves are caller-owned objects.

Environment variables​

Uncensored routing requires credentials for the uncensored providers. At least one of each is recommended:

# Text (at least one for mature/private-adult)
OPENROUTER_API_KEY=sk-or-...

# Image (Replicate is the primary uncensored host)
REPLICATE_API_TOKEN=r8_...

# Optional fallbacks
FAL_API_KEY=...
STABLE_DIFFUSION_LOCAL_BASE_URL=http://localhost:7860

If no uncensored-capable credentials are set, the APIs fall back to whatever censored provider is configured and the tier has no effect.

Logging and telemetry​

Both generateImage and editImage emit OpenTelemetry spans with llm.provider and llm.model attributes. When policy-aware routing kicks in, the resolved provider/model reflect the uncensored choice, so dashboards can filter by tier and spot hits on the uncensored path.