Pre-Screening and Post-Screening Policy

No content on Ember is generated or published without passing automated checks first. This policy explains the concrete screening steps we run before a generation happens ("pre-screening") and after ("post-screening"), and what you can expect from them. It works alongside the Content Moderation Policy, which covers how we govern moderation overall, our enforcement ladder, and how decisions are made; this policy focuses on the screening mechanics themselves.

1. Why we screen

The Service only produces content through generations that we run. Because we control that step, we screen every prompt, reference image, and character before it becomes an Output, and we keep checking after. This is what lets us keep the Service free of content involving minors, non-consensual depictions of real people, and other content prohibited under the Prohibited Content Policy.

2. Pre-screening: prompts

Before any generation runs, your prompt is classified by an automated content-policy system. One of three outcomes applies:

• Allow. The prompt proceeds as written.
• Rewrite. If wording is youth-coded or leaves someone's age ambiguous, the prompt is automatically rewritten so that every person described is explicitly an adult of 18 or older. We tell you when a prompt has been rewritten and why.
• Block. Any prompt that shows intent to depict a minor in a sexual, nude, suggestive, or fetishized context is blocked before generation, including attempts using misspellings, euphemisms, non-English wording, or anime or cartoon styles. We tell you when a prompt has been blocked and why.

3. Pre-screening: reference images

Every reference image you upload is checked with automated age estimation on each face it detects. If anyone in the image appears possibly under 18, the image is rejected and cannot be used, regardless of any age you claim for them. This check is separate from, and does not affect, the rule that real, identifiable people may not be depicted on the Service at all, whatever their age.

4. Pre-screening: characters

Character names, descriptions, and persona text are screened under the same rules as prompts. Characters you keep private go through this automated screening like anything else. If you make a character public, it may also receive additional human review, either before it is listed or afterward, on top of the automated checks every character receives.

5. Records of blocked material

When a prompt or reference image is blocked, we keep a record of it: the prompt or image itself, the reason it was blocked, the account involved, and the time. We keep these records to enforce this policy, detect patterns of abuse, and meet legal reporting obligations, including the reporting described in Section 9. Access is limited to our safety and legal teams and, where the law requires it, to law enforcement. We retain these records for as long as your account exists and for as long as the law requires afterward.

6. Post-screening: watermarking

Every Output carries a "MADE WITH" badge and a tiled brand mark, so its synthetic origin is disclosed wherever it appears. Watermarking does not depend on whether the Output stayed private or was made public.

7. Post-screening: review and audits

Screening does not stop once a generation runs. Flagged Outputs and public characters go through automated and human review after the fact. We also run periodic audits of how well our screening is performing and use the results to retrain or tune our classifiers, so the systems keep improving as new patterns emerge.

8. Reports and removal requests

Post-screening also includes what other people tell us. Users can report content, and third parties can request removal of content that concerns them, under the Content Removal Policy. Reports are handled on the same enforcement ladder as anything our own systems flag.

9. When screening finds a violation

If pre-screening or post-screening turns up a violation, we may remove the content, take action on the account (a warning, suspension, or permanent ban, as described in the Content Moderation Policy), and preserve the content and related evidence. Apparent child sexual abuse material, or an attempt to create it, is reported to the National Center for Missing and Exploited Children. See the Underage Policy for how we handle anything involving minors specifically.

10. Limits, false positives, and appeals

Automated screening is not perfect. It may occasionally block or rewrite a prompt, or reject a reference image, that would in fact have been legitimate adult content. If you believe screening got a decision wrong, you can appeal it under the Complaint Policy.

11. Applies to everything you create

Screening applies the same way to private and public content, because what we control is the act of generation itself, not who ends up seeing the result afterward. We do not use your content to train our AI models; see the Terms of Service for that commitment in full.

12. No bypassing screening

Attempting to evade, trick, or probe our screening systems, whether by altering your wording, splitting a request across multiple prompts, or any other means, violates the Acceptable Use Policy and is treated as a deliberate violation.

13. Contact

Questions about this policy can be sent to legal@ember.xxx.