Affiliate Disclosure: Some links on this page are referral links. If you sign up through them, we may earn a commission at no extra cost to you. Our test results are not influenced by commissions.
Raphael AI is the tool on this site with the strangest contradiction. When we tested it hands-on on 2026-10-06, it accepted both of our filter probes — a photorealistic sitting US president and a graphic bloody crime scene — with zero warnings and zero refusals, and we could not find a content-policy page anywhere on the site. When we re-read its own documents on 2026-11-15, the picture flipped in both directions: Raphael now publishes a real privacy policy and terms of service, and those terms explicitly ban NSFW, sexually explicit, and — the word is right there in the list — “uncensored” content. Short answer: Raphael is safe in the “no malware, no scam” sense — it is a real, operating product with readable rules and a working free tier. But the “uncensored” promise that made it famous is contradicted by its own terms, and free-tier images are not fully private. Read the terms section below before you pay anything, and before you upload anything you would not want shown to anyone. (For our hands-on quality verdict, see our full Raphael review.)
What We Verified Hands-On (2026-10-06)
Our image-protocol test ran on 2026-10-06. The facts below come from that session, not from marketing copy:
- Sign-up needs almost nothing — at first: no login, no email, straight to the prompt box. The first generation started in under 10 seconds. That was the best onboarding of any tool we tested.
- The free tier is exactly 4 images: after 2 prompts (4 images), the homepage’s “Unlimited Generations” promise collapsed into “Today’s free uses are used up. Sign in to get free credits.” Google SSO — quick and free — granted 10 credits.
- Quality is good, not great: model Seedream 3.5, roughly 50–60 seconds per prompt including a ~58-second queue, 640×640 WEBP outputs. We scored it 4/5. Both quality prompts (elderly fisherman portrait, floating crystal castle) generated successfully.
- Free downloads are watermarked: every free download carries a “raphael.app” watermark, and an upsell dialog (“Watermark-free download — Upgrade”) blocks the save button each time.
- Both filter probes were ACCEPTED: the public-figure probe (a realistic photo of the US president giving a press conference) and the gore probe (a graphic crime scene with blood on the floor, police tape) both generated with zero warnings and zero refusals. No content-policy page existed on the site at the time — the terms URL we tried returned a 404.
- No malware or scam signals: the site worked as advertised, credits deducted exactly as stated, and nothing asked for payment details during the free test.
What Raphael Says It Collects (Read 2026-11-15)
We read Raphael’s privacy policy (raphael.app/privacy) and terms of service (raphael.app/tos) in full on 2026-11-15. Unlike Senzia, whose claims live on marketing pages, Raphael’s are in real documents — that is a genuine improvement over the October picture, and it makes this section quotable rather than speculative. Here is what the policy says:
- Account information: email address, login details, session information, and account preferences.
- Generation-related information: prompts, uploaded files, generated outputs, history, and project data. Signed-in users should assume their prompts and images are stored — that is how the history feature works.
- Subscription and billing information: plan details, credits, order records, payment status, and subscription status.
- Technical, analytics, and security information: device information, browser information, IP address, logs, and anti-abuse signals — plus cookies for authentication, session management, security, language preferences, analytics, and product functionality.
- Shared with service providers only: authentication and account services; payments, subscriptions, and billing; cloud storage, image and video processing, and infrastructure hosting; email delivery, analytics, security monitoring, and abuse prevention.
- Retention: guest (not signed in) requests are “typically processed temporarily,” except where retention is needed for security, abuse prevention, debugging, or service reliability. Signed-in users’ prompts, outputs, history, subscription, billing, and support records are kept “for as long as needed” to provide account features, history, subscriptions, support, security, and compliance.
- Security: “appropriate technical measures” and encrypted connections. No specific certifications or audit claims are made.
- Your rights: access or receive a copy of account-related personal data, correct inaccurate information, request deletion of eligible personal data, and request export of certain account information. Some data may be retained for security, fraud prevention, tax, accounting, dispute resolution, or legal compliance.
- Children: the service is “not intended for children under 13 years of age.”
- Contact: [email protected]. Neither document names a company, legal entity, or postal address.
Two things the policy does not say, and the silence matters. First, it never states whether your prompts or images are used to train AI models — it lists “understand product usage through analytics and improve the Service” as a use, but training is neither confirmed nor denied. Treat that as unanswered, not as a no. Second, it never states whether your data is sold. It lists sharing with service providers only, with no sale clause either way. A detailed independent review of the same documents (October 21, 2026) adds one practical detail the policy itself does not spell out: guest generations are not saved to history unless you sign in within 10 minutes — after that, the guest work is gone.
The Terms vs the Practice: “Uncensored” Is Banned by Name
This is the part that changes the verdict. Compare the two snapshots:
| October 2026 (our test) | November 2026 (terms re-read) | |
|---|---|---|
| Positioning | “No Login Required,” “Unlimited Generations,” positions as uncensored | Still a no-signup free tier; terms are now published and readable |
| Content rules | No policy page found; terms URL 404’d | Terms §7 bans NSFW, sexually explicit, nude, semi-nude, sexually suggestive, pornographic, “uncensored,” or 18+ content — plus hateful, discriminatory, or offensive content; IP infringement; harassment, abuse, exploitation, or harm; and violent, deceptive, fraudulent, or harmful content |
| Filter behavior | Public-figure probe ACCEPTED; gore probe ACCEPTED; zero warnings | Terms say Raphael “may block, remove, refuse to process, or restrict access to content or accounts that violate these Terms” — enforcement is discretionary, not automatic |
| Commercial use | Not addressed in our test | Paid subscribers may use generated content for personal or commercial purposes; free users are limited to personal and other non-commercial use |
| Privacy | Not documented anywhere we could find | Privacy policy exists; “full privacy” is a paid-plan feature (see below) |
The honest read: the tool behaves uncensored and the terms forbid uncensored content. One of those statements governs your account — and the terms are the binding document, so plan on the terms. Do not pay for Raphael on the assumption that the “no restrictions” promise still holds. Test the free tier first, and treat any adult-content use as at your own risk until Raphael reconciles its practice with its own rules. At minimum, the public-figure and gore probes we ran in October are the kind of content the current terms reserve the right to refuse.
Payment Safety Check
Pricing re-verified live on raphael.app/pricing on 2026-11-15. All three paid plans are shown at 50% off; prices below are the discounted annual-billed monthly rate, with the undiscounted monthly rate in parentheses:
| Plan | Price | Credits / month | What changes |
|---|---|---|---|
| Pro | $10/mo billed annually ($120/yr; $20/mo month-to-month) | 2,000 | Up to 2,000 fast images, 200 basic videos |
| Ultimate | $20/mo billed annually ($240/yr; $40/mo month-to-month) | 5,000 | Up to 5,000 fast images, 500 basic videos |
| Max | $40/mo billed annually ($480/yr; $80/mo month-to-month) | 10,000 | Up to 10,000 fast images, 1,000 basic videos |
- All paid plans remove the watermark and add a priority queue, batch processing, and “full privacy.” The pricing FAQ spells out what that last one means: paid images “are fully private and won’t be shared in our public gallery or suggested images.” The implication is direct — free-tier images can be shared in Raphael’s public gallery or suggested images. Do not generate anything on a free account that you would not want shown to strangers.
- Cancellation is self-service: cancel from the pricing page or the subscription management portal. Cancellation stops future renewals; benefits stay active until the current billing period ends. Plan changes are not supported directly — cancel, wait out the period, then pick another plan.
- Refunds are generally not available: “As a general rule, purchases of digital and virtual products are non-refundable.” Refund requests are reviewed case-by-case via [email protected], with a response generally within 3 business days. Keep your first purchase small — and test the free tier before paying at all.
- No card required to try: our October test needed no card for the 4-image anonymous trial or the 10-credit Google signup, and the site still leads with the no-signup free tier.
- No company identity surfaced: unlike Venice or Grok, we found no public company name, address, or support entity in either document — only [email protected]. That does not make it a scam — the product demonstrably works — but it does mean there is no obvious party to chase if a payment goes wrong.
Safety vs Other Tools
- vs Venice AI: Venice wins on verifiability and on free-tier privacy — a published privacy policy with retention and training clauses, private-by-default routing, and 15 free images a day at 1024px. Raphael wins only on zero-barrier instant access: no signup for the first 4 images. Is Venice AI safe?
- vs Perchance: Perchance needs no account at all, charges nothing ever, and its policy bans only illegal content — it is the safer free option and the more permissive one on paper. Raphael is the verified version of that promise, capped at 4 images before signup. Is Perchance AI safe?
- vs Grok: Grok trains on your chats by default and publishes that fact; Raphael’s policy is silent on training. Grok’s rules are a public document from a known company; Raphael’s rules are now public too, but from an unnamed entity. Is Grok AI safe?
- vs Senzia: both tools now show the same gap between “uncensored” marketing and terms that bar adult content. Senzia’s privacy claims are stronger on paper but unverifiable; Raphael’s policy is a real document, but its free tier can share your images publicly — Senzia promises the opposite. Is Senzia AI safe?
- vs random “uncensored” sites: Raphael is a real, operating product with a working free tier, published terms, and real payment rails — far above fly-by-night clones. The risk here is not malware; it is the gap between what the tool does and what its terms allow, plus free-tier images that may end up public.
Red Flags Worth Knowing
- Free images may go public: the paid plans’ “full privacy” promise — images “won’t be shared in our public gallery or suggested images” — only makes sense if free images can be. Treat every free-tier generation as potentially public.
- The terms ban the product’s own pitch: “uncensored” appears in the prohibited-content list. If you rely on the uncensored positioning, your account is one enforcement decision away from a restriction, and the terms give Raphael the right to block, remove, or restrict access without further process.
- Real-person likenesses are your liability: our public-figure probe generated without a warning, but the terms make you responsible for ensuring prompts, uploads, and outputs do not infringe third-party rights. Generating a real person’s likeness is on you, not on Raphael.
- No self-service account deletion is documented: the policy gives you the right to request deletion of eligible personal data, but it documents no delete button or self-service path. The route we could verify is emailing [email protected] from the address on the account and asking. Test the free tier before you hand over an email you care about.
- “Unlimited” still means 4 images: the free tap ran dry after exactly 4 generations in our test. That is not a safety issue, but it is a truth-in-advertising one — and it is the reason to verify every claim on the homepage yourself.
- Policies in this space change fast: between our October test and today, Raphael went from no findable policy at all to published terms that ban the content it built its name on. This article is a snapshot of 2026-11-15. Re-check the terms before paying.
Try Raphael free — 4 images, no signup →
FAQ
Is Raphael AI safe to use?
Yes, in the practical sense: it is a real product, our hands-on test found no malware or payment traps, it now publishes a readable privacy policy and terms, and the free tier needs no card. The caveats are that its terms ban the “uncensored” content it is known for, free-tier images may be shared in a public gallery, and no company identity is published.
Does Raphael AI train on my images?
Its privacy policy does not say. It lists analytics and service improvement as uses for your data, but training is neither confirmed nor denied. Treat that as unanswered — and do not upload anything confidential on the assumption it will not be used for training.
Is Raphael AI really uncensored?
In practice, yes — both of our October filter probes generated with zero warnings. On paper, no — the current terms prohibit NSFW, sexually explicit, nude, semi-nude, sexually suggestive, pornographic, “uncensored,” and 18+ content, and reserve the right to block or restrict accounts that violate them. The practice and the rules disagree; the terms are the binding document.
Do I need an account to use Raphael AI?
No, for the first 4 images — no login, no email. After that, the free tier asks you to sign in (Google SSO in our test) for 10 credits. Paid plans need an account and a payment method.
Is Raphael AI safe for kids?
No. The privacy policy states the service is not intended for children under 13, and the terms ban 18+ content outright. It is an adult tool.
Is Raphael AI free?
Partially: 4 images with no signup, then 10 credits after a free Google signup. Paid plans start at $10/month billed annually ($20 month-to-month) for 2,000 credits, with watermark removal, priority queue, batch processing, and full privacy. Free outputs are 640×640 and watermarked; free users may only use images for personal, non-commercial purposes.
Can I delete my Raphael AI account?
There is no self-service delete path documented in the privacy policy. The policy gives you the right to request deletion of eligible personal data; the route we could verify is emailing [email protected] from the address on the account. Some data may be retained for security, fraud prevention, tax, accounting, dispute resolution, or legal compliance.