What this audit solves
Searchers want a chatbot that will do adult roleplay. That is a policy question first: does the operator sell NSFW, or forbid it? Prompt workarounds on Character.AI and Replika are the wrong job — those operators say pornographic content is outside the product ([Replika / CAI status](/en/snippets/replika-nsfw-unlock-status/)).
Practical pick: for hosted NSFW text plus a written adult product, start with Candy AI. Large custom-bot catalog on a phone → CrushOn (official website). Voice or photos inside the thread → Muah AI. Dedicated stills → DreamGF, not a chatbot ranking.
Affiliate disclosure: Dating Auditors may earn a commission if you join Candy AI, CrushOn, Muah AI, or DreamGF through links on this page. Policy limits below come from each operator’s public pages and apply regardless of that relationship.
Evidence used
Checked 31 August 2026. This hub is a buyer map of written adult policy, not a lab of refusal rates, response times, or “24 tested models.” Older Dating Auditors copy invented those scores. They are not reused here.
- Candy Blocked Content Policy (30 July 2026), Terms (17 August 2026), and help: Classic vs Roleplay Mode for Create My AI only; Premium Plus names Generate 18+ images. Deep Mode was not found as a current setting. Detail: Is Candy uncensored?.
- Character.AI help and ToS (ToS last updated 27 August 2025): pornographic content will not be supported; requesting NSFW-filter removal can ban; guidelines forbid bypassing safety systems.
- Replika ToS (no last-updated date on the page): 18+; vulgar/obscene/pornographic user content prohibited. Pro help documents Romantic Partner and unblurred romantic photos — not an NSFW unlock.
- Janitor help: free default JLLM plus an optional proxy that needs your key. Adult rules on the proxy follow the upstream host.
- CrushOn FAQ / install wiki: hidden content unlocked on the website; store apps claimed, listing unknown. Not PWA-only.
- DreamGF: mix-and-match stills, 10 tokens/image, 100/month. Not a chatbot replacement.
Girlfriend-specific stack (memory, visuals, “apps” shopping): AI girlfriend apps. Privacy/logs: safe NSFW AI chat.
How to judge “uncensored” without fake indexes
Ignore invented friction scores. Check:
- Policy — Is adult RP a sold feature, or a ToS violation? Marketing “no filters / talk about anything” is weaker than the legal page.
- Hosting — One account (Candy) vs a UI plus optional proxy key (Janitor) vs a community catalog (CrushOn).
- Hard bans that still apply — Candy’s Blocked Content Policy still forbids minors, CSAM, non-consensual intimate imagery, and deepfakes of real people. “Uncensored” is not “no rules.”
- Access — Browser or Add to Home Screen vs a filtered store build (NSFW AI on phones).
- Job fit — Text chat, character catalog, voice/media, or stills. One product should not win every column.
Live refusal rates, filter internals, and free-vs-paid model routing are unknown here. We did not re-run chats.
Platform comparison (adult chat)
Qualitative snapshot from operator documents — re-check the live legal page before you renew. This is not a scored ranking.
| Platform | What public documents support | Best for | Limit |
|---|---|---|---|
| Candy AI | Sells NSFW chat and 18+ images. Free generally 5 messages (Terms). Deep Mode is not a current documented setting | Primary: hosted adult girlfriend chat | Blocked Content Policy still applies. Cloud processing, not zero-log. Chat languages on Candy’s uncensored marketing page: EN/FR/DE/ES/JP |
| CrushOn | Character catalog; hidden content unlocked on the website | Phone + many custom bots | Use the official website. Store listing unknown. Not a privacy pick. FAQ forbids throwaway email |
| Muah AI | In-chat voice and media is the product | Alt: only when that is the job | Not the default chatbot. Privacy: voice/media |
| DreamGF | Mix-and-match adult stills; token-metered | Stills of a recurring character | Not a chatbot hub winner. No video. NSFW image generators |
| Janitor AI | JLLM by default; optional proxy + your key | Stay if you want model picking | Proxy adult policy = upstream host. Not “UI-only.” Janitor alternatives |
| Replika / Character.AI | Adult / pornographic content forbidden or unsupported | SFW chat or Replika’s sold romance companion | Do not jailbreak. Migrate if you need adult RP |
SpicyChat, Chai, Talkie, and similar app-store “spicy” brands are not reviewed on this hub. Do not treat an omitted row as a hidden recommendation.
1. Why mainstream “uncensored” hopes fail
Character.AI’s help center says pornographic content is against its terms and will not be supported. Asking to remove NSFW filters can result in a ban. Its community guidelines forbid bypassing safety systems. Minors can be on the platform (ToS 13+, 16+ in EEA/UK).
Replika is 18+ and still lists vulgar/obscene/pornographic user content as prohibited. Pro’s Romantic Partner status and unblurred romantic photos are not a documented NSFW unlock.
That is product design, not a missing prompt. This hub does not publish bypass recipes. Filter internals (classifiers, interceptors, keyword lists) are unknown; older copy invented that architecture.
2. Primary pick + focused alternates
Primary: Candy AI
Default uncensored chatbot on this hub: hosted inference (no user-pasted OpenRouter key), adult RP sold as a product, and a written Blocked Content Policy. Replaces both CAI/Replika dead-ends and Janitor proxy babysitting.
It is not Deep Mode, “no filters,” or a zero-moderation guarantee. Free is generally five messages, then pay — not a sourced “filtered free stack.” Policy vs marketing: Candy uncensored status. Retention and vendors: Is Candy AI safe?.
Do not choose Candy if you need a documented zero-log host, deepfakes of real people, or a verified filter-pass score from Dating Auditors.
CTA: Open Candy AI — adult product with a written policy, not Deep Mode
Alternate (catalog / phone): CrushOn
When the job is browsing many custom characters on a phone. Prefer the official website and Add to Home Screen. CrushOn is not PWA-only; FAQ also claims store apps. Hidden content is unlocked on the site. Install steps: CrushOn on iOS & Android.
Alternate (voice / in-chat media): Muah AI
Only when photos or voice have to appear in the conversation. Not a gallery tool and not the default text chatbot. Product vs privacy: Muah review, voice/media privacy.
Wrong job on this hub
- DreamGF — stills builder, token-metered. Use the NSFW image page.
- Stay on Janitor if JLLM or a working proxy is the product.
3. Related maps
| Need | Guide |
|---|---|
| Is Candy actually uncensored? | Candy uncensored status |
| Girlfriend memory / apps shopping | Best AI girlfriend apps |
| Logs, deletion, trackers | Safe NSFW AI chat |
| CAI / Replika status | Why workarounds fail |
| Hosted vs Janitor proxy | Janitor AI alternatives |
| Phone access | CrushOn on mobile |
FAQ
Why do Character.AI and Replika refuse NSFW prompts?
Their public help and terms put pornographic / obscene content outside the product. Character.AI says NSFW will not be supported and that asking to remove filters can ban you. Sustained adult RP needs a host that sells it — Candy AI if you accept cloud processing — not a jailbreak paste.
How do I use uncensored AI on iPhone or Android without App Store filters?
Use the browser or Add to Home Screen, not a filtered store “AI girlfriend.” CrushOn is the catalog path via the official website. Candy’s help documents no native iOS/Android app. Do not treat CrushOn as PWA-only; its FAQ also claims store apps, and listing status is unknown.
Are free “uncensored” tiers actually open?
Candy’s Terms cap free users at generally five messages, then a paywall. That is a message cap, not a sourced “safety-wrapped free model.” Whether a given explicit prompt is refused on any plan is unknown without an in-account test. This page is not that test.