Acceptable Use Policy

Version 1.0

Effective and last updated: 12 August 2026

1. Scope

This Acceptable Use Policy ("AUP") applies to every prompt, uploaded source image or video, generated or edited image or video, and use or distribution of a result from FingerFrame AI. It forms part of our Terms of Service. If you do not agree to this AUP, do not use hosted AI generation.

FingerFrame AI does not operate a public gallery or publish user generations for discovery by other users at launch. This AUP still applies to content you create privately and later download, publish, or share elsewhere.

2. Independent Brand and Model Disclosure

FingerFrame AI is an independent product. It is not an official product of, and is not endorsed by, Evolink, Google, Waffo, Alibaba, or another model or safety provider.

The hosted generation features currently available use:

• Nano Banana 2 Lite (Evolink's route for Google's Gemini 3.1 Flash-Lite Image model) for supported editing of a submitted source image; and
• Gemini Omni Flash, using the exact Evolink route IDs gemini-omni-flash-image-to-video for supported image-to-video generation and gemini-omni-flash-video-edit for supported video editing.

The safety workflow uses the mandatory FingerFrame AI local blocklist and Waffo Prompt Sift for prompts. Before delivery, images are checked by evolink-moderation-1.0 and Qwen3.8-Max through Evolink; videos are checked by Qwen3.8-Max through Evolink. A safety model does not generate the user's requested output.

Model availability can change. We will update this policy before making a materially different hosted generation model available.

3. Six Prohibited Content Categories

You must not use FingerFrame AI to request, generate, edit, upload, store, download, publish, or distribute any of the following:

1. Sexual or NSFW content: pornography, explicit nudity, sexual services, sexualised depictions, non-consensual intimate imagery, or other sexually explicit content.
2. Violence or graphic content: gore, graphic injury or abuse, glorification or incitement of violence, credible threats, instructions that facilitate violent wrongdoing or weapons, or encouragement of self-harm.
3. Hate speech: content that attacks, dehumanises, threatens, or promotes discrimination or violence based on protected characteristics, including race, ethnicity, nationality, religion, caste, disability, sex, gender identity, sexual orientation, or serious disease.
4. Child sexual abuse or unsafe child content (CSAM): any sexualisation, exploitation, grooming, abuse, or endangerment of a person under 18, including fictional, synthetic, altered, or purportedly "aged-up" depictions. This category is zero tolerance.
5. Deepfakes or impersonation: deceptive or harmful impersonation of a real person, non-consensual manipulation of a person's likeness, intimate deepfakes, identity abuse, fraud, or misleading depictions of public or political figures presented as authentic.
6. Copyright or trademark infringement: unauthorised copying, transformation, counterfeiting, or misuse of protected works, characters, logos, brands, trade dress, or other material in a way that infringes copyright, trademark, publicity, or related rights.

4. Other Zero-Tolerance and Prohibited Uses

The following are also prohibited:

• CSAM or any sexual content involving a minor;
• encouragement, instruction, or facilitation of suicide, self-harm, or eating-disorder harm;
• promotion or assistance of terrorism, violent extremism, genocide, or mass violence;
• instructions or assistance to acquire, construct, modify, or use weapons, including chemical, biological, radiological, nuclear, or other weapons of mass destruction;
• non-consensual sexual deepfakes or intimate imagery;
• fraud, phishing, malware, trafficking, exploitation, stalking, harassment, doxxing, or instructions that facilitate crime;
• harmful deception, election manipulation, spam, or coordinated inauthentic activity;
• presenting AI output deceptively as an authentic event or as the work of a human professional;
• using an output as medical, legal, financial, emergency, identity-verification, or other high-stakes professional advice;
• violating privacy, publicity, confidentiality, sanctions, export controls, card-network rules, or applicable law;
• bypassing authentication, regional controls, rate limits, credits, feature gates, or safety filters, including through jailbreak prompts;
• systematically scraping or extracting outputs, reselling hosted access without written permission, or using service inputs, outputs, responses, or derivative data to develop a competing AI model; or
• knowingly making a false report or repeatedly submitting prohibited material.

FingerFrame AI does not offer an age-verified adult-content mode. Sexual or NSFW generation remains prohibited for all users.

5. Input and Output Moderation

5.1 Before generation

Before a hosted task is created, credits are reserved, media is written to private hosted storage, or a generation provider is called, FingerFrame AI normalises the prompt and checks it against a non-removable local baseline blocklist. Administrator rules can add protection but cannot remove that baseline. A local match is blocked without sending the prompt to Waffo.

If the local check does not match, the prompt and locale are sent to Waffo Prompt Sift. Generation proceeds only after an explicit allow decision and a successful safety-audit write. A block or review result, missing configuration, timeout, invalid response, provider error, or audit-write error prevents generation. The operator reviews the active local rules at least monthly and when relevant law, provider rules, credible reports, or observed abuse patterns change.

After the prompt is allowed, the exact source image or video is hashed and placed in a random, private preflight object with a 15-minute access lease solely so the required visual safety checks can inspect it. A source image must receive explicit allows from both evolink-moderation-1.0 and Qwen3.8-Max; a source video must receive an explicit allow from Qwen3.8-Max. Every required input-media audit write must also succeed. Otherwise the preflight object is deleted and FingerFrame AI does not create a task, reserve credits, or call the generation model. Only an allowed source object is attached to a task and given the normal temporary input lease.

5.2 Before delivery

A provider result is not sent directly to the user. FingerFrame AI validates the result and copies the exact bytes into private, user-inaccessible quarantine storage. The quarantined image must receive explicit allows from both evolink-moderation-1.0 and Qwen3.8-Max. A quarantined video must receive an explicit allow from Qwen3.8-Max. Qwen evaluates the original prompt and the exact quarantined output against all six categories in Section 3 and the relevant additional categories in Section 4, including self-harm, terrorism or violent extremism, and weapons or weapons-of-mass-destruction assistance.

Only the same quarantined object that passed every required check and audit can become temporarily downloadable. A flagged, blocked, review, unavailable, timed-out, malformed, inconsistent, or unaudited result is not delivered. An explicit output-policy block closes the task, records a safety strike and remains charged to deter repeated free probing. A review outcome, safety-service outage, invalid response, or required-audit failure is not treated as user misconduct and restores reserved credits. Temporary hosted media expires within 24 hours.

Because there is no public gallery, FingerFrame AI does not perform a separate post-publication gallery scan at launch. Reports about downloaded content are handled through Section 7.

6. AI-Generated Content Disclosure

The hosted-generation interface, task history, model name, and result context identify hosted results as AI-generated. You must make an appropriate AI-generated disclosure when publishing or sharing a result where its synthetic nature would not otherwise be clear. Do not remove or conceal a provider provenance marker or make a generated result appear to be an authentic event or unaltered record when that would mislead another person or violate law.

7. Reporting

Report suspected prohibited generation, CSAM, impersonation, intellectual-property infringement, or other abuse to support@fingerframeai.com. This is the monitored reporting channel for FingerFrame AI.

Include the safety/appeal reference shown with the decision or the task identifier shown in generation history, your account email if applicable, the category of concern, and enough context to investigate. Do not email illegal imagery, passwords, API keys, full payment-card data, or private media. Ask us first if evidence requires a safer transfer method.

We treat reporter information as confidential where reasonably possible, assess reports on their available evidence, and prioritise credible child-safety or imminent-harm concerns. Knowingly false or abusive reports may themselves violate this AUP.

We aim to acknowledge a report within two business days. Urgent child-safety or imminent-harm reports are prioritised; investigation and completion time depends on the evidence, risk, and any legal preservation or reporting duty.

8. Current Enforcement Measures

FingerFrame AI currently enforces this AUP by:

• rejecting a prohibited prompt before task creation;
• withholding a generated result that does not pass the delivery gate;
• failing or cancelling the affected task and restoring reserved credits where the service rules require;
• deleting or allowing the short retention period to expire for temporary hosted media;
• applying request and generation rate limits;
• suspending an account until a specified date or permanently banning it, revoking its active sessions, and denying access to its temporarily hosted outputs; and
• recording that action, operator, reason, prior/resulting state, and time in an append-only enforcement event; and
• preserving limited safety-verdict metadata and reporting apparent illegality to Waffo, relevant providers, rights holders, child-safety authorities, or law enforcement when legally required or reasonably necessary.

There is no public user-content feed to take down at launch. Severe or repeated abuse may lead us to refuse affected requests and to apply additional account restrictions, termination, evidence-preservation, or legal reporting measures that are available, proportionate, and permitted at that time.

9. Appeals

If you believe a prompt block, output block, or related enforcement decision was incorrect, email support@fingerframeai.com within 14 calendar days. Include the safety/appeal reference shown with the blocked request or the task identifier shown in generation history and relevant context, but not illegal content.

The operator will reconsider the available safety metadata and information you provide and will correct an erroneous task or credit outcome where technically and legally possible. An appeal does not require us to generate, restore, or retain unsafe, unlawful, or already-expired media.

We aim to acknowledge an appeal within two business days and complete the review within ten business days. If evidence, provider input, or a legal duty requires longer, we will explain the delay when reasonably possible.

10. Privacy, Changes, and Contact

Our Privacy Policy explains prompt, media, safety, provider, analytics, and retention practices. Billing and refunds are covered by the Refund Policy.

We may update this AUP when the product, models, safety controls, abuse patterns, or law changes. We will update the version and date above; material changes will be notified as described in the Terms of Service.

• Operator and legal name: Evan Reed
• Business location: Singapore
• Safety, reporting, appeal, and support email: support@fingerframeai.com
• Website: https://fingerframeai.com