Solution
The uncensored AI API
"Uncensored" has never meant "no rules". It means the rules are written down, enforced at the account layer, and placed in a policy you can read — not inside a filter that refuses your users probabilistically and bills you for the privilege.
50 free credits · no card · private mode on every account
Refusals are a product killer
A filtered model refusing mid-scene is not safety, it is breakage: the character dies mid-story, the user churns, and the refusal still billed. Worse, it is random — the same prompt passes on Tuesday and fails on Thursday, so you cannot design around it.
eroq's engines are selected so fiction renders. The limits live in a four-paragraph acceptable-use policy — no minors, no real people, nothing illegal — enforced at the account level. Requests outside it return an explicit content_blocked error code and are never charged.
Uncensored, across every modality
The same policy-not-filter posture holds for all 35 models:
- →Chat: RP+ stays in character through dark and intense scenes.
- →Images: Eroq Uncensored and Anima render the scene as written at 10 credits, with negative prompts and CFG control.
- →Video on Motion One, eroq's own engine, and voice: same range, same rules, same error codes. The third-party video engines are moderated by their providers.
Private, too
Uncensored is half the promise; privacy is the other half. Private mode — a Studio toggle, or private: true on the image and video endpoints — returns the render once and keeps nothing: no file, no library entry, and the prompt starred out in the usage ledger. eroq's own engines (RP+, RP mini, the voices, the image models and Motion One) retain nothing upstream; the third-party video engines may log prompts at their provider, and GET /v1/engines says which is which.
Accountable, not anonymous
Half the "uncensored" market is anonymous endpoints with no entity and no tomorrow. eroq is operated by Qore Games (Paris) with real terms of service and sale, an invoiced billing trail, and a ledger you can audit call by call. Permissive and accountable are not opposites — they are the product.
How eroq compares
The uncensored space has real options with different jobs. The honest comparison is by what you are building:
| eroq | Venice AI | OpenRouter | Self-hosted | |
|---|---|---|---|---|
| Built for | Shipping uncensored creative products | Private, uncensored inference | Routing across many models | Full control, your GPUs |
| Content posture | Written AUP, uncensored own engines, content_blocked on refused renders, uncharged | Uncensored models, privacy-first framing | Inherited from each upstream provider | Whatever you enforce |
| Privacy | Private mode keeps nothing; own engines retain nothing upstream | Privacy-first — their headline feature | Varies by upstream provider | Yours |
| Behind one key | RP chat · image · video · TTS · STT · CDN storage | Chat and image generation | Chat-first, varies by model | What you deploy |
| Pricing model | Flat credits per call, failed calls auto-refund | Token-based | Per-token, per provider | GPU-hours plus your ops time |
| Roleplay tuning | Sampling profiles measured on production character platforms | General-purpose models | Depends on the model you pick | Yours to build |
| Media hosting | Included — the eroq Store CDN | — | — | Bring your own infra |
Comparison reflects public positioning as of August 2026 — verify current details on each provider's site. If you want raw private inference, Venice is a fine choice; if you are shipping an uncensored creative product, the full surface above is the argument.
One call to see it
curl https://eroq.ai/v1/chat/completions \
-H "Authorization: Bearer $EROQ_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "eroq-rp-plus",
"context": "Mira: sardonic starship mechanic, dry humor, hates small talk. Scene: engine bay, mid-shift.",
"messages": [
{
"role": "user",
"content": "The reactor is making that noise again."
}
]
}'Base URL https://eroq.ai/v1 · full reference in the docs
| Model | Kind | Price |
|---|---|---|
| eroq-rp-plus | chat | 3 credits / completion |
| eroq-rp-mini | chat | 1 credit / completion |
| eroq-uncensored | image | 10 credits / image |
| eroq-krea2 | image | 10 credits / image |
| eroq-one | image | 20 credits / image |
| eroq-anime | image | 10 credits / image |
| eroq-motion-one | video | 12 credits / second |
| eroq-motion-pro | video | 30 credits / second |
| seedance-2-5 | video | 62 credits / second |
| veo-3-1 | video | 106 credits / second |
| sora-2-pro | video | 80 credits / second |
| kling-3-pro | video | 45 credits / second |
| seedance-2-0 | video | 40 credits / second |
| hailuo-3 | video | 35 credits / second |
| runway-gen-4-5 | video | 32 credits / second |
| veo-3-1-fast | video | 27 credits / second |
| kling-3 | video | 34 credits / second |
| seedance-2-0-fast | video | 24 credits / second |
| hailuo-3-max | video | 22 credits / second |
| grok-imagine | video | 37 credits / second |
| seedance-2-0-mini | video | 9 credits / second |
| veo-3-1-lite | video | 14 credits / second |
| wan-3 | video | 27 credits / second |
| eroq-flash-image | image | 10 credits / image |
| eroq-flux-image | image | 12 credits / image |
| gpt-image-2-5-sunburst | image | 31 credits / image |
| gpt-image-2-5-flare | image | 31 credits / image |
| nano-banana-2-1 | image | 31 credits / image |
| seedream-5-0-pro | image | 36 credits / image |
| nano-banana-pro | image | 124 credits / image |
| flux-3-image | image | 32 credits / image |
| grok-imagine-2-0 | image | 32 credits / image |
Full table on the pricing page.
Questions builders ask
What does "uncensored" mean here, concretely? +
Engines selected so fiction renders instead of refusing — dark, violent or provocative themes included — plus a written acceptable-use policy as the only wall: no minors, no real-person impersonation, nothing illegal — enforced at the account level with explicit error codes.
What is the best uncensored AI API? +
For shipping uncensored creative products, we make the case that it is eroq: roleplay-tuned uncensored chat, image, video and voice behind one key, a Private mode that keeps nothing, flat credit pricing with automatic refunds, CDN storage, and a written policy rather than a filter. For raw private inference or model shopping, Venice AI and OpenRouter are the serious alternatives — the ranked breakdown is on the blog.
How is eroq different from Venice AI? +
Venice is a privacy-first inference platform with uncensored models. eroq makes the same two promises — private and uncensored — and adds the product layers: persona-tuned chat, image, video, speech and storage in one API, flat per-call credits that refund on failure, and a written acceptable-use policy. Different jobs — if you are building a character product, the product surface is the difference.
Will I get banned for dark or intense content? +
No — fiction that goes to dark places is the supported use case. Enforcement targets the policy's hard limits, not a theme's intensity.
What happens when a prompt is outside policy? +
The call returns HTTP 400 with code content_blocked, is not charged, and your product can react programmatically — no silent failures, no lectures.
Why not just use a jailbroken general API? +
Jailbreaks are probabilistic and break on every model update; a policy-level guarantee is contractual. Your product deserves infrastructure, not a trick.