feat: improve AI tool attribution from metadata signals#22
Merged
Conversation
Map unique metadata patterns to specific AI models so detection results identify WHICH tool generated the content, not just that it's AI-generated. Tool identification rate improves from 30% to 79% across aihubmix test files. - C2PA: Google claim_generator/info fallback → "google ai" (Imagen/Gemini) - XMP: photoshop:Credit "Made with Google AI" → "google ai" - MP4: AIGC ContentProducer ID mapping → "wan" (Wan videos) - EXIF: AIGC JSON in UserComment with ContentProducer prefix → "qwen" - known_tools: add "google ai", "wan", "qwen", "通义万相" patterns - i18n: add signal_xmp_credit and signal_exif_aigc_label to all 7 locales Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
MatrixA
changed the title
…idated generation script Add AI-generated media from 60+ models via aihubmix API as test fixtures, with integration tests covering all detection methods (C2PA, XMP, EXIF AIGC, MP4 AIGC, watermark, WAV TTS heuristic, filename). Also consolidates 7 generation scripts into one with all API route fixes from retry rounds. - 70 new fixtures: 39 images, 22 videos, 9 audio from Google Imagen/Gemini, OpenAI, Ideogram, Flux, Doubao Seedream/Seedance, Jimeng, Wan, Veo, Qwen, Sora, and various TTS models - 49 new integration tests in tests/aihubmix_detection.rs - .gitattributes: add LFS tracking for .wav, .png, .jpg, .jpeg in fixtures - scripts/: consolidated generate_aihubmix_all_models.py with all retry fixes (Flux polling, Gemini native API, AIGC producer mapping, multipart upload) - Cleaned up: removed 6 retry scripts, generated/ directory, __pycache__ Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Update AI tool count from 61 to 76 across all 9 README files - Document XMP Credit field detection (e.g. Google AI) - Document EXIF AIGC label detection (e.g. Qwen via ContentProducer) - Document MP4 AIGC ContentProducer→tool mapping (e.g. Wan videos) - Document C2PA claim_generator vendor inference (e.g. Google) - Add new tools to recognition table: Qwen, Wan, Google AI - Update Cargo.toml description to list all 10 detection methods Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Improve AI tool identification so detection results tell you WHICH model generated the content, not just "it's AI." Tool identification rate improves from 30% to 79% across 86 detected aihubmix-generated files.
Changes
claim_generator/claim_generator_infofallback →google ai(covers all Imagen & Gemini images)photoshop:Credit= "Made with Google AI" →google aiContentProducerID mapping →wan(all Wan video variants)UserCommentAIGC JSON withContentProducerprefix →qwen(Qianfan Qwen images)google ai,wan,qwen,通义万相signal_xmp_creditandsignal_exif_aigc_labelkeys to all 7 locale filesChecklist
cargo fmt -- --checkpassescargo clippy -- -D warningspassescargo testpasses