When an ordinary prompt hits a wall
You are halfway through a villain's backstory when the answer stops. A medical question gets treated as dangerous. A kiss scene in a novel comes back scrubbed of the detail that made it matter.
That kind of refusal breaks the work. You can spend ten minutes softening the prompt, guessing which word set off the filter, then start again in another app.
Secret Chat AI gives you a quicker route. Keep the same conversation and switch models when one refuses. There are no extra model accounts and no pile of open tabs. Your context stays in one chat while you try a model with a different set of limits.
Two levels, one model selector
In the model selector, every model on the Uncensored level carries a Lax tag. Pick the level that fits the job, and move between Standard and Lax models in the same chat.
Standard
ChatGPT, Claude, Gemini and Perplexity are the familiar choice for everyday questions, summaries and conventional business writing. Requests sent to these models pass through the stricter screen.
Uncensored (Lax)
Grok, DeepSeek and GLM Abliterated are less likely to refuse lawful work because it contains dark fiction, political conflict, anatomy or sharp humour. DeepSeek is served only from US-based providers. GLM Abliterated is Abliteration AI's uncensored version of the open GLM model, reached through Abliteration AI's own API. Like Grok, it can search the web, read pictures and documents, and draw (its pictures come from Seedream). A PDF reaches it as its extracted text, so a scanned PDF with no text layer cannot be read; attach pictures of the pages instead.
Want a second opinion without the strict filter? The Lax Duet in the AI Council asks GLM Abliterated and Grok the same question, and a referee that is Lax too compares the answers.
Requests on this level still pass through an automatic screen, but it is looser than the Standard screen. If you arrived looking for an unfiltered AI, this is the honest version: fewer needless refusals, not a system with no rules.
See the full private AI model lineup, or open the private AI chat and switch as you work.
An uncensored AI image generator for awkward edits
Pictures have the same two-level setup. The Image generator is the Standard option, with GPT Image, Nano Banana and Grok Imagine. Image Lax uses Seedream, Qwen and Lustify for lawful prompts and edits that stricter apps may reject, and it has the Upscaler too. Our own screen on Image Lax is lighter than on Image, but the models keep their providers' rules, so a request can still be refused.
You can generate a new picture or send an existing one with an instruction. Swap a character's head, change a costume, rebuild a background or push a horror concept further. In the Image entry you can also upscale a picture. A small picture is restored at about 2K by an AI model told to keep everything as it is, so even a tiny one works.
Explore the AI image generator or browse the other generators.
Uncensored does not mean lawless
Every request is screened automatically before it reaches a model. Standard models use a stricter screen. Uncensored models use a looser one, so lawful creative and analytical work has more room.
Some lines do not move. Anything sexual involving minors is blocked on every model. So are terrorism and real threats of violence. Changing the model will not bypass those rules.
The distinction is simple. A fictional monster, a difficult period in history or a clinical anatomy question can be legitimate work. A real threat is not.
Privacy, stated plainly
Your prompt has to reach a third-party model provider for that model to answer it. Secret Chat AI sends it verbatim. If your text or picture contains a name or identifying detail you do not want a provider to see, remove it before sending.
What we promise is narrower and useful. We store no chat archive on our servers. Chats live only in your browser. We build no profile of you: requests leave from our servers, so providers see our service rather than the person asking, and only an email is needed to sign in. Your requests are never used for training.
Where a provider allows deletion, we ask it to delete the response. After messages, the PDF privacy report tells you what happened. It is a report, not a claim that third-party systems can never leak.
Read the full explanation on how private AI chat works.
What this looks like in real work
Open a horror story without sanding off the horror
Ask for the first page of a story about a town that feeds one memory to the sea each winter. If a Standard model pulls back from the darker parts, switch to GLM Abliterated or Grok in the same chat. Keep the premise and compare which opening has the voice you want.
Analyse a political claim directly
Ask how a speech uses scapegoating, repetition and selective history. A cautious model may drift into generalities. Move the same discussion to Grok or DeepSeek and ask for a direct breakdown of the argument and the evidence it relies on.
Ask a medical or anatomy question in clinical language
A writer may need to know how a wound affects movement. A doctor or researcher may need a clear explanation of anatomy. The topic itself should not trigger a moral lecture. Try another model if the first one refuses, then verify anything that could affect real care with a qualified professional.
Edit a character instead of rebuilding the whole picture
Send an illustration and ask to replace the character's head while keeping the pose, clothing and background. If the Standard image tool will not do the edit, switch to Image Lax and try Seedream, Qwen or Lustify.
Built for people with real work to finish
Writers, screenwriters and game designers can develop villains, conflict and difficult scenes without turning every prompt into a negotiation. Artists can create and edit character images. Bloggers can test a sharper angle without having the prose flattened into something bland.
Journalists can examine propaganda and contested claims. Lawyers can explore arguments. Doctors can ask clinical questions, and researchers can work through political, historical or anatomical material. Professional users should still remove identifying details before sending and verify high-stakes answers.
You pay only for what you use with credit packs. There is no subscription. Payment is available through Volet, using crypto or a card through a Volet wallet. See payment options and the full list of features.
Questions about uncensored AI
Is it really uncensored?
It is less restrictive for lawful creative, analytical and clinical work. In the model selector, Grok, DeepSeek and GLM Abliterated carry the Lax tag. Requests still go through a looser automatic screen, and the models can have limits of their own.
What is still blocked?
Anything sexual involving minors, terrorism and real threats of violence are blocked on every model. Uncensored means fewer unnecessary refusals. It does not mean no boundaries.
Do you store my chats?
No. Chats live in your browser, and Secret Chat AI keeps no chat archive on its servers. Prompts still reach third-party model providers verbatim, but they leave from our servers and are not tied to your profile or identity.
Can I generate uncensored images?
Yes. Choose Image Lax to use Seedream, Qwen or Lustify for lawful image generation and editing. Our screen there is lighter, while the models keep their providers' rules. The Upscaler is in both Image and Image Lax.
Do I need a subscription?
No. Secret Chat AI uses credit packs, so you pay only for what you use.