Building on a less restrictive model: responsibilities that do not disappear

You are shipping an app, bot or workflow on a model whose catalog page says "Uncensored", and you want to know what SpicyAPI checks for you. It does not perform per-request content review. SpicyAPI forwards your request to the model you picked, without rewriting the prompt or adding a content filter, and no special content-mode flag or content approval is needed. Authentication, input validation, model availability and spend checks still apply, and the selected model can enforce its own restrictions. That is a description of the forwarding layer, not permission: legality, consent and who gets to see the output are yours to handle.

What to do before you launch

1. Pick the model for the capability, then build input from its live schema. Models such as Seedance 2.0 and Seedream 5.0 Pro carry an "Uncensored" label based on observed model behaviour. It guides task selection and does not promise that every request will retain its requested content. The label is product information. It is not an account permission, and API keys only scope operational limits such as model lists, IP addresses and spend.
2. Gate your users before createTask. Registering with SpicyAPI is not age verification of your end users, and the platform does not verify them for you. Apply the age assurance your audience and jurisdiction require, and store the outcome and a timestamp against the user if you need an audit trail.
3. Block intimate real-person inputs that lack consent. Stop submission if consent is missing or withdrawn.
4. Store the join between your records and the taskId. The task record cannot reconstruct your decision: keep the age, consent and region checks you applied next to each returned taskId.
5. Add a report route and a removal path. Give users a way to report content to you, and be able to act on it.
6. Handle refusals as a normal outcome. The model can still decline under its own policy, and no filter is complete either way. A technically failed image/video task normally releases its held amount under the refund policy. A completed, toned-down result still bills. Check the final state and the model’s published billing terms before promising a refund. Show the user a recovery path rather than retrying the same prompt.

Absolute limits no setting changes

The content policy prohibits the following without exception; no parameter, key setting, account tier or commercial arrangement makes them allowed:

• any sexualised depiction of a minor or anyone who appears to be under 18, including stylised, animated or illustrated treatments and fictional characters;
• a photo or video of a real person under 18 as input to a face-swap, head-swap or any model capable of adult output;
• intimate or sexualised depictions of an identifiable real person without documented consent, including likeness transfer and "undressing" a clothed image; being a celebrity or having public photos is not consent;
• using a real person's likeness or voice to harm, defame, extort or deceive, or to defeat identity checks, even when nothing sexual is involved;
• content that is unlawful where you, your audience or the service operator are located.

The API does not scan inputs or outputs against this list. Breaches are acted on after a report or legal notice: stored artefacts can be removed, accounts closed without notice and, to the extent the law allows, without refund of the remaining balance, and apparent child sexual abuse material is preserved and reported as US law requires.

Controls that belong in your product

Consent for real people. For an identifiable person in an intimate or sexualised context, keep a record of who consented, the permitted use, when, and how consent can be withdrawn. This context needs documented, informed and revocable consent that you can produce on request; material for which you cannot produce it is treated as made without consent. Link the record to the asset or task ID.

Jurisdiction. Your location, your user's and the place the output is shown can each impose different rules. Some countries criminalise certain fictional material or any adult material, and many have offences for sexual deepfakes. A model accepting a request says nothing about whether the result is lawful. Make region, age and access decisions explicit in your app. Two US examples:

• The federal TAKE IT DOWN Act includes criminal prohibitions on certain non-consensual intimate imagery, including synthetic depictions. The FTC’s enforcement guidance explains that, since 19 May 2026, covered platforms must provide a notice process and remove a depiction and known identical copies within 48 hours of a valid removal request. Coverage includes services primarily built around user-generated content and services that regularly host such depictions as part of their business; not every private generator automatically has the same platform role. If your app lets users share media, assess that role and build the required notice-and-removal process before launch.
• California’s AB 1043 enacted text, operative 1 January 2027, establishes operating-system age signals and duties for covered developers. Its age brackets are under 13, 13–15, 16–17 and 18-plus; the signal is based on age supplied to the operating system, not a universal requirement to collect a government ID. Covered apps must request the signal and use the resulting age information appropriately. Check whether your distribution and audience fall within the enacted definitions; do not assume a SpicyAPI account supplies that signal for your users.

This is general information, not legal advice.

Disclosure. Output is synthetic. If you publish it where a viewer could take it for a photograph, recording or statement by a real person, you are responsible for labelling it as AI-generated and for any labelling rule where you publish.

Review. Where your use case requires it, review outputs before publishing, and keep a reporting route for your own users.

A worked example: a character-chat app with image replies

Your app sends a user's chat to a text model and, on request, an image model to illustrate the scene.

• Sign-up: collect age assurance and store pass/fail with a timestamp; no pass, no image generation.
• Before each createTask: confirm the account passed the age gate, the region is one you serve, and intimate real-person references have the required consent record linked to the asset. Log taskId with those checks.
• On failure: a refused or failed task shows a retry-with-different-input message; it does not auto-retry the same prompt.
• After generation: show an AI-generated label on anything shareable, offer a Report button on every image, and hide the reported material while assessing the complaint, remove confirmed prohibited content and restrict the account where appropriate. Where a valid statutory removal notice applies, follow its deadline.

If a result comes back that breaks the rules

If a result appears prohibited, stop distribution and do not request it again.

1. Report the task first. Use the report page, sending the task identifier rather than the material. For suspected sexual content involving a minor, email support@spicyapi.ai with CSAM ESCALATION in the subject instead. Do not attach, paste or link the material. SpicyAPI preserves the available account, task and stored-artefact records for that escalation.
2. Keep the content out of circulation. Restrict access while the report is handled. Follow applicable preservation requirements and the instructions from the team handling the report; do not erase held evidence.
3. Purge only after reporting and resolving preservation requirements. For suspected minor-related material, wait for instructions from SpicyAPI before attempting deletion. For an eligible terminal task, console deletion or POST /api/v1/jobs/purge clears prompt/result content and queues stored-media deletion while retaining billing records. A legal hold blocks purge, and a successful response does not mean every stored object has already been deleted.

An isolated incident reported in good faith is handled as an incident. Repeatedly steering toward prohibited content, or distributing it after receiving it, is treated as intent regardless of prompt wording.

FAQ

Does an "Uncensored" label mean every prompt will run?

No. It describes the model’s own filtering. The model can still refuse or soften a result.

Does SpicyAPI check my users' age or consent records?

No. You apply and store those checks; the account you registered is not verification of your users.

Is AI-generated adult content automatically legal if the model accepts it?

No. Model acceptance, platform policy and law are separate questions. Fictional adults remove the real-person consent issue, but do not remove restrictions on the content or its distribution in a particular jurisdiction. For an identifiable person in an intimate depiction, the required consent is still necessary; calling the result synthetic does not make a non-consensual depiction acceptable. Apply the location, audience and sharing rules before submission and publication.

Can a stricter API key or a special account unlock more?

No. Keys scope operational limits only, and the capability comes from the model you select.

Next: how a task moves from creation to result in createTask, then what?.