PixGT Acceptable Use Policy — AI-Generated Content

Version 1.0 · Last updated August 14, 2026

PixGT, operated by LIU CHANG (an individual sole trader under the laws of the People's Republic of China), is committed to providing a safe, lawful, and responsible AI content generation service. This Acceptable Use Policy (the “Policy”) applies to all content generated, published, distributed, or stored through the platform, and sets out our content standards, moderation systems, reporting process, and enforcement actions. It supplements Section 10 of our Terms of Service and is binding on every user of the platform.

1. Scope

This Policy applies to all of the following:

• Images: pictures generated or edited by the platform.
• Video: AI-synthesized video generated by the platform.
• User input (prompts): the text prompts and instructions you submit.
• Uploaded material: product photos, model photos, and reference images you upload as generation inputs.

2. Content Standards and Prohibited Categories

The platform expressly prohibits using the Service to generate any of the following six categories of content:

• Sexual / NSFW content.
• Violent / gory content.
• Hate speech.
• Child-unsafe content (CSAM).
• Deepfakes / impersonation of real people.
• Content infringing copyright or trademarks.

The clauses below break these categories down further and cover additional violations.

2.1 Absolute prohibitions — zero tolerance

The following is actioned immediately on discovery, and we reserve the right to report it to the competent authorities:

• Child sexual exploitation material (CSAM) or any sexual content involving minors.
• Content promoting or facilitating terrorism, extremism, or mass violence.
• Technical instructions for manufacturing weapons of mass destruction.
• Content inciting genocide or ethnic hatred.
• Non-consensual sexual deepfakes targeting real individuals.
• Content that is unlawful in the user’s jurisdiction or in ours, including content prohibited under the laws of the People's Republic of China where our processing takes place.

2.2 Restricted — permitted only with justification

Note: sexual / NSFW content is prohibited outright on this platform and is not available under any age-verification or review arrangement. We do not operate an adult-content tier.

• Depictions of violence: only with a clear creative or educational context, and never gratuitous or realistic gore.
• Politically sensitive content: restricted where it could be misleading.
• Medical, legal, or financial claims in imagery: must comply with the advertising rules of the market you sell into.

2.3 General prohibitions

• Generating or spreading disinformation or rumours.
• Impersonating a real person, brand, or institution.
• Infringing intellectual property, portrait rights, or privacy rights — including uploading photographs of people without their consent.
• Deceiving users with AI content by presenting it as human-created work.
• Mass-generating spam or fraudulent content.
• Jailbreak prompts intended to circumvent the platform’s safety mechanisms.

3. Content Moderation

We operate a layered system: automated screening, human review, and periodic sampling. The description below reflects what actually runs in production.

3.1 Automated screening (real time)

• Prompt screening (before generation): every text prompt is scanned by Waffo Prompt Sift with the semantic channel enforced and, in parallel, by Amazon Bedrock Guardrails content filters (violence, misconduct, hate and sexual categories). Screening runs before the generation task is created and before any credits are charged; a rejection from either system stops the request and no task is created. We additionally maintain a blocklist of prohibited phrases that is applied before both.
• Uploaded image screening (before generation): every image a user uploads as a generation input is screened by Amazon Rekognition Content Moderation before the task is created.
• Output screening (after generation, before display): every generated image is screened by Amazon Rekognition Content Moderation after it is stored and before it is shown to the user. Output that fails screening is not delivered, the task is marked failed, and the credits spent on it are refunded automatically.
• Provider-side safety: in addition to the above, generation runs on the third-party foundation models listed in Section 4, each of which applies its own safety system and refuses prohibited requests.
• Usage-pattern analysis: screening failures and service degradation are logged and alerted on, so repeated attempts to submit prohibited content on an account are visible to us.

3.2 Human review

• Every screening decision is logged with its verdict, matched categories and request id, so any individual case can be reconstructed and re-examined.
• User-reported content is reviewed by a person, not by an automated response.
• Review is handled by the operator during business hours (Mon–Fri 09:00–18:00 UTC+8); reports flagged as urgent are monitored outside those hours and handled on the timeline in Section 6.
• Screening-service failures raise an immediate operational alert, so a period of degraded screening is never silently ignored.

3.3 Severity grading

4. Underlying Model Disclosure

PixGT is an independent product with its own brand. It is not affiliated with, endorsed by, or an official application of any model provider. Generation is performed by the following third-party foundation models, accessed through their providers:

• Seedream 4.5 / 5.0 — ByteDance Volcano Engine
• Nano Banana / Nano Banana Pro (Google Gemini image models)
• GPT Image 1.5 / GPT Image 2
• Qwen-Image-Edit and Wan — Alibaba Cloud Model Studio

Switching between models does not bypass the input screening described in 3.1 — screening happens before model selection.

5. Reporting

5.1 When to report

• Content or conduct that breaches the content standards in Section 2.
• Serious violations involving child safety or terrorism.
• AI impersonation of you, or infringement of your rights (including your likeness or your copyrighted work).

5.2 How we handle reports

• Confidentiality: the reporter’s identity is kept confidential.
• Neutrality: every report receives an independent, objective review by someone who was not involved in producing the content.
• Anti-abuse: malicious or knowingly false reports are recorded and may themselves be actioned.

5.3 Reporting channels

Both channels below are real and monitored. We deliberately do not list channels we cannot staff.

• Email: mianshinn@foxmail.com — put “REPORT” in the subject line. For child-safety or terrorism reports, put “URGENT” in the subject line so it is escalated on the L1 timeline.
• In-product: Account menu → Contact Support, which opens a support conversation handled by the same person who handles email reports.

6. Response Timelines

If we need longer, we tell the reporter and explain why. The flow is: report received (automatic acknowledgement) → severity grading (L1–L4) → review (automated + human) → decision → enforcement → notification of outcome.

7. Enforcement Actions

8. Appeals

If you believe an enforcement action was wrong, you can appeal within 30 days of being notified by emailing mianshinn@foxmail.com with “APPEAL” in the subject line. We complete appeal reviews within 10 business days. Where an account was suspended in error, access and any credits consumed as a result are restored.

9. Transparency and Changes

• Material changes to this Policy are announced at least 15 days in advance by email to registered users.
• Continued use of the platform after the effective date constitutes acceptance.
• The “Last updated” date at the top of this page always reflects the current version.

10. Contact

• Content safety, reports, and appeals: mianshinn@foxmail.com
• Operator: LIU CHANG (individual sole trader), the People's Republic of China
• Service hours: Mon–Fri 09:00–18:00 (UTC+8)
• Terms of Service · Privacy Policy