← Back to Insights

A Publishing Checklist for AI Images, Synthetic Voices and Digital Humans

By the RAY STUDIO editorial team

Before release, review AI use, asset and personal permissions, visible and embedded labels, platform declarations, file versions and approval records.

室内拍摄现场的摄影机、三脚架与工作人员

Before publishing video with AI imagery, synthetic voices or digital humans, check more than visual quality. Use one checklist covering AI scope, asset and personal permissions, labels, platform declarations, output files and review records, with a named approver.

Whether footage is live action or AI-generated, review the version, images, sound, labels and approvals together before release.

1. Identify where AI was used

Record generated shots, people, narration, music, captions and virtual scenes, distinguishing them from real footage. Mixed productions need timecodes, tools and intended uses so review does not depend on memory.

Distinguish internal assistance from public outputs. AI research or storyboard references differ from generated imagery, sound or people in the finished film. Assess the final distribution file.

2. Verify assets, likenesses and voices

Generation does not automatically resolve rights. Check sources and permitted uses of inputs, fonts, music, trademarks, product photographs and references. For identifiable people or voices, confirm that agreements cover generation, editing and public distribution.

China's deep-synthesis regulations cover synthetic voices, generated or replaced faces, manipulated poses and digital people, and impose notification and consent requirements for editing biometric information such as faces and voices. Digital-human or voice tools do not remove the need to check permissions.

3. Check clear visible labels

The Measures for Labelling AI-Generated Synthetic Content, published in March 2025 and effective from 1 September 2025, distinguish visible and implicit labels and specify forms for text, audio, images, video and virtual scenes.

Check placement for each format. Video needs appropriate prominent notices at its beginning and around playback; audio may use spoken or rhythmic indications and interface notices. Assess digital humans and virtual scenes against their presentation and platform rules. An “AI version” filename alone is insufficient.

4. Preserve implicit labels through export and transcoding

Metadata can record synthetic-content attributes, provider details and content identifiers. Files may pass through editing, grading, voiceover, subtitling, compression and platform conversion, so check the final publishing file again.

Do not maliciously remove, alter, forge or hide required labels for a cleaner look. Retain each format, language and platform version with its check results to prevent notices or metadata disappearing during export.

5. Declare AI content when uploading

The measures require publishers to declare generated or synthetic content and use provider labelling functions. Check the upload page for AI or content-declaration options even when the film already has visible notices.

Websites, WeChat Channels, Douyin, Bilibili and other channels have different interfaces and rules. Record each account, version, declaration method, publication time and operator.

6. Archive files, evidence and approvals

Keep final films, captions or copy, AI-use notes, sources, likeness and voice permissions, declaration screenshots or records, reviewer names and dates. These support later corrections, withdrawals or explanations better than recollection.

For long-running or repeatedly edited advertisements, separate masters from platform versions and lock final filenames and version numbers. This improves later revisions, repeat campaigns and multi-channel distribution.

Commercial and product video production: https://raystudio.art/services/dongguan-guanggaopian

AI labelling checks: https://raystudio.art/news/ab764w9af49isk30h154188l

AI-video procurement contracts: https://raystudio.art/news/yzc9kmfp4fcddwktl34i3x5f

Project brief: https://raystudio.art/contact#project-brief-form

7. References and content note

CAC, MIIT, MPS and NRTA, Measures for Labelling AI-Generated Synthetic Content, published 14 March 2025, effective 1 September 2025: https://www.cac.gov.cn/2025-03/14/c_1743654684782215.htm

CAC, MIIT and MPS, Provisions on the Administration of Deep Synthesis of Internet Information Services, published 11 December 2022, effective 10 January 2023: https://www.cac.gov.cn/2022-12/11/c_1672221949354811.htm

National standard GB 45438–2025: https://openstd.samr.gov.cn/bzgk/std/newGbInfo?hcno=F32EA2A561F1886CD8D606513512D547&refer=outter

Prepared by Guangdong RAY STUDIO from official rules verifiable as of 28 August 2026 and common publishing workflows. This is a production checklist, not legal advice. Review each project's content, services, platforms and applicable rules separately.