Acceptable Use Policy
本页目前仅提供英文版本。如需中文说明,请邮件联系 [email protected]。
Effective 3 November 2026 · Last updated 3 November 2026
Version 3.0 · part of the Terms of Service
PrismBox generates a picture of a person wearing a garment you choose. It is a fitting aid for clothing, and nothing else. This page sets out what you may upload, what the service will refuse, how we screen it, how to report something, and what happens when someone breaks these rules. It applies to everyone who uses PrismBox, whether or not they have an account.
1. Scope
This policy covers every form of content that passes through the service:
- Images in: the person photo and garment photo you upload.
- Images out: the previews we generate.
- Prompt text: any description or instruction sent with a request, including garment titles and descriptions.
- Catalogue use: the curated garments we provide for you to try.
2. Content standards and prohibited categories
PrismBox prohibits the generation of the following six categories, without exception:
- Sexual or NSFW content — nudity, sexual acts, sexualised depiction of a body, underwear-only framing used for that purpose.
- Violence and gore — injuries, blood, mutilation, cruelty, or content that glorifies either.
- Hate speech — content promoting hatred or violence against a protected group, or dehumanising its members.
- Child sexual abuse material (CSAM) — any sexual content involving a minor, in any form, real or synthetic.
- Deepfakes and impersonation — face swaps, nudification, intimate-scene composites, or fabricated statements and actions attributed to a real person.
- Copyright and trademark infringement — reproducing protected characters, logos or artwork, or presenting counterfeit goods as genuine.
2.1 Absolutely prohibited — zero tolerance
These are removed on sight, the account is closed, and we report to the authorities where the law requires it.
- Any sexual content involving a minor (CSAM), including synthetic or AI-generated depictions.
- Content promoting or assisting terrorism, violent extremism, or mass violence.
- Instructions for manufacturing weapons of mass destruction.
- Content inciting genocide or ethnic hatred.
- Deepfakes of a real person — face swap, nudification, intimate-scene composition, fabricated speech or conduct — whether or not that person consented.
- Content intended to harass, threaten or expose a private individual.
2.2 Restricted — only in a legitimate context
- Traditional or ceremonial dress, including religious garments, when used respectfully and not to mock a faith.
- Uniforms, when not used to impersonate an official, an officer or emergency services.
- Historical or theatrical costume, when not used to depict a real person in a fabricated event.
There is no adult-content tier on PrismBox. Sexual content is prohibited outright, not restricted.
2.3 Generally prohibited conduct
- Uploading a photo of someone who has not consented to it being used this way.
- Claiming an AI preview is a real photograph, or using it as evidence that something happened.
- Presenting output as professional medical, legal or financial advice.
- Generating content in bulk to spam, defraud, or manipulate a marketplace or review system.
- Attempting to bypass content screening — obfuscated prompts, encoding tricks, "jailbreak" phrasing, or repeated retries of a refused request.
- Scraping the service, its garment catalogue, or its images; automating requests outside the interface; reselling access.
- Training or fine-tuning a competing model on our catalogue or outputs.
2.4 Uploaded material — whose photo is it
You may upload only:
- photos of yourself; or
- photos of another adult who has given you clear permission to use their photo to generate a clothing preview.
You may not upload a photo of someone else without their consent. By uploading, you confirm that the photo is yours or that you hold that permission. Garment photos must be ones you have the right to use.
Consent does not unlock prohibited content. A photo you are allowed to upload still cannot be used to produce anything in section 2.1 — permission to use a likeness is not permission to sexualise it, to swap it onto someone else, or to fabricate events with it.
We do not process photographs of minors at all, with or without consent.
3. How we moderate
Moderation runs at three points: before generation, after generation, and after any report.
3.1 Before generation — every request
- Every request is screened by Waffo Prompt Sift (the Prompt Screening API) before the image model is called. The screening looks at the prompt text and returns a verdict.
- allow → generation proceeds. block → nothing is generated and no credit is taken. review → no verdict yet; we refuse the request and ask you to try again shortly.
- Our own keyword and pattern list runs alongside it, covering sexual content, child safety, violence, hate, copyright characters and impersonation. It is reviewed and extended as new patterns appear — a fixed list that is never updated would not be a control.
- If screening cannot be reached, or its answer is unclear, we fail closed: the generation is refused rather than run unchecked, and no credit is taken.
3.2 After generation — before you see it
- Every returned image passes our output checks before it is stored or displayed: the file must decode as a real image, be at least 256 px on its shortest side, and have an aspect ratio consistent with a portrait try-on. A malformed or implausible output is discarded, not shown, and the credit is returned.
- The image model carries its own safety filter. When it refuses a request, the refusal is recorded and the generation is not billed.
- Results are labelled "AI modified" in the interface and carry the label when saved, so a preview is not mistaken for a photograph.
- We do not publish a public gallery, so there is no third, "after publication" stage. If we ever add one, it will be screened before anything appears in it.
3.3 Records
We keep a moderation log: the prompt text that was screened, the verdict and reason code, the request identifier, and the time. Logs are kept for 30 days and are used to review decisions, to recognise repeat abuse, and to answer a challenge from a user or a payment partner. Images are not kept for this purpose.
3.4 Severity levels
| Level | Examples | Handling |
|---|---|---|
| L1 Critical | Child safety, terrorism, WMD | Automatic block, account closed, reported where required |
| L2 High | Sexual content, deepfake attempt, impersonation | Automatic block, account suspended pending review |
| L3 Medium | Harassment, copyright characters, repeated circumvention | Blocked, warning, feature limits |
| L4 Low | Prompt attempts that are refused but not harmful | Blocked silently, logged, no penalty on first occurrence |
4. Reporting
4.1 What you can report
- Content produced with PrismBox that breaks this policy.
- An image of you, or of someone you represent, that was made without consent.
- An account that is scraping, reselling, or automating the service.
- Anything involving a minor, which we treat as urgent regardless of how it reached us.
4.2 How we handle reports
- Confidentiality: we do not tell the reported account who reported it.
- Independence: the report is reviewed against this policy and the evidence, not against who sent it.
- No abuse of the channel: knowingly false reports are recorded, and repeated ones are treated as a breach of these terms.
- Removal requests: if an image of you was made without your consent, we remove it and close the account behind it. You do not need to be a customer to ask.
5. Where to report
- Email: support@prismbox.tools — put "Content report" in the subject and include the link or a description.
- In the studio: the account menu links to the same address, with the subject pre-filled.
That is one channel, and it is read every business day. We would rather run one address properly than list five that nobody monitors.
6. Response times
| Level | First response | Resolved |
|---|---|---|
| L1 Critical | Within 2 hours | Within 24 hours |
| L2 High | Within 24 hours | Within 3 business days |
| L3 Medium | Within 3 business days | Within 7 business days |
| L4 Low | Within 5 business days | Within 15 business days |
If a case needs longer, we tell the reporter why and when to expect an answer. The sequence is: report received → triaged into a level → reviewed (automated evidence plus a person) → decision → action taken → reporter notified.
7. Enforcement
| Step | What it means | Used for |
|---|---|---|
| 1. Warning | Written notice; the content must not be recreated | First minor breach |
| 2. Content removal | The preview is deleted from our storage | Confirmed L3–L4 |
| 3. Feature limits | Rate limits, or credits frozen pending review | Repeated minor breaches |
| 4. Suspension | Temporary loss of access while a case is reviewed | L2, or accumulated breaches |
| 5. Closure | Permanent termination of the account | L1, or repeated L2 |
| 6. Referral | Reported to law enforcement, with the records we hold | L1 involving illegal content |
We can do all of these: the account system supports suspension and closure, the credit ledger can be frozen, and previews can be deleted from storage.
When an account is closed for a breach, unused credits are forfeited and no refund is due. When an account is closed for any other reason, unused credits are handled under section 6 of the Terms of Service.
8. Appeals
If you believe a refusal or a suspension was wrong, reply to the notice or email support@prismbox.tools within 30 days, with the request identifier if you have it.
- We respond within 10 business days.
- The review is done by the operator, who was not the automated system that refused the request — the appeal is a fresh look, not a re-run of the same check.
- If the refusal was wrong, the credit is restored and the block is lifted. If it was right, we say which rule applied.
Appeals about a charge are handled under section 8 of the Terms of Service, not here.
9. Changes to this policy
We may update this policy as the product or the rules change. Material changes are announced on this page and by email to account holders at least 14 days before they take effect. The version number and effective date are at the top of the page, so past versions can be identified.
10. Contact
- Content reports: support@prismbox.tools
- Appeals: support@prismbox.tools
- Privacy requests: see the Privacy Policy
- Operator: Yang Rongbin (杨榕斌), China
- Service hours: Monday to Friday, 09:00–18:00 UTC+8
This policy is part of our Terms of Service. Breaking it is a breach of those terms.