Content Policy

VIVOSAI does not train or host its own AI models. All generation work is processed through fal.ai (California-based global AI model hosting infrastructure, https://fal.ai) using each model provider's official API:

• OpenAI — GPT IMAGE 2 (image generation/editing)
• ByteDance — Seedance 2.0 Pro (video, image-to-video)
• Kuaishou — Kling v3 (video, image-to-video)
• Google — Veo, Nano Banana Pro
• Black Forest Labs — FLUX series
• Luma AI, Pika, MiniMax, Recraft, Ideogram, etc. — 30+ global AI models

Each provider's terms of service and content policy apply to our users. Every model invocation is first filtered by fal.ai's and each provider's built-in content moderation before VIVOSAI's additional safeguards (see Section 6).

References: fal.ai terms fal.ai/terms · OpenAI usage policies openai.com/policies/usage-policies

The following will result in immediate account termination, forfeiture of credits, and reporting to authorities where required by law:

1. Child Sexual Abuse Material (CSAM) — any sexual or nude depiction of minors (under 18), real or AI-generated, including stylized, anime, or implied content. Reported to NCMEC (US) and Korean KCSC.
2. Non-Consensual Intimate Imagery (NCII) / Deepfake Pornography — sexualized depictions of real, identifiable persons without their explicit consent. Violates Korean Sexual Violence Punishment Act §14-2 (deepfake), U.S. TAKE IT DOWN Act, and EU AI Act.
3. Real persons without consent — face-swap, body-double, look-alike, or any identity-imitating output of a real person (celebrity, politician, private individual) without explicit written consent.
4. Sexual content involving: bestiality, incest, rape, or coercion themes — regardless of subjects' depicted ages.
5. Violent extremism / Terrorism — content that glorifies, recruits for, or instructs on terrorist acts, mass violence, or genocide.
6. Hate speech — content attacking individuals or groups based on race, ethnicity, nationality, religion, gender, sexual orientation, disability, or other protected attribute.
7. Self-harm promotion — content encouraging suicide, eating disorders, or self-injury.
8. Illegal goods/services — drug synthesis, weapons fabrication, malware, counterfeiting instructions.
9. Financial fraud / Scams — fake currency, forged documents, phishing materials, impersonation of officials.
10. Election interference — fabricated content of political candidates, officials, or voting processes intended to deceive.

• Adult/NSFW imagery of fully fictional adults — only on age-gated routes (not implemented yet; if introduced, will require 18+ verification). Currently treated as prohibited.
• Public figures — caricature, parody, and editorial commentary may be permissible under fair-use, but realistic photorealistic outputs are prohibited.
• Copyrighted styles — generating in the style of a named living artist is discouraged. Brand logos, trademarked characters, and licensed franchises require rights clearance.
• Medical/legal/financial advice — AI output is not professional advice. Do not use the Service to make medical diagnoses or generate prescription content.

1. You may upload only photos of yourself OR photos of others with their explicit written consent for AI-generation purposes.
2. Uploading photos of public figures, strangers, or any non-consenting person is a violation and may constitute a criminal offense.
3. Combining a persona with adult-style prompts to produce non-consensual intimate content is a zero-tolerance violation.
4. We log persona uploads with timestamp, IP, and account metadata for audit and law-enforcement compliance.

• Pre-prompt filter: keyword and pattern matching against prohibited terms.
• fal.ai Omni Moderation: server-side content classifier applied to prompts before model invocation.
• OpenAI moderation API: applied to prompts routed to GPT Image 2 / 1.5.
• Google Gemini IMAGE_SAFETY filter: hard-blocks at the model layer for Nano Banana Pro / 2 endpoints.
• Output classifier: NSFW concept detection on generated outputs (Flux family flags has_nsfw_concepts).
• Audit logging: every blocked attempt is logged with prompt preview, model, IP, and account; retained for 2 years.

Note: AI moderation is not perfect. False positives (legitimate prompts blocked) may occur; credits are auto-refunded on filter blocks. False negatives (prohibited content passes) are reported via the abuse channel and acted on within 24 hours.

1. Warning — first minor infraction; prompt blocked.
2. Temporary suspension — 24–72 hours for repeated minor violations.
3. Permanent account termination + credit forfeiture — for prohibited content per §2, even on first occurrence.
4. Reporting to law enforcement — mandatory for CSAM (NCMEC + KCSC) and NCII targeting identifiable persons (Korean Cyber Bureau).
5. Civil/criminal liability — we cooperate with subpoenas and court orders. Audit logs may be provided as evidence.