Repository Issue Activity (beta)

#45 v4-round6, built-in watermark detector in this project gave good results, but GEMINI command "@SynthID" still said "generated or edited by Google AI".

#52 Repo is not working due to model changes.

#49 Unable to detect synthID in validation images using reverse-SynthID/src/extraction/robust_extractor.py

#51 Gemini image downloads being weird

#41 My SynthID Findings

#12 Starlog published a deep-dive on aloshdenny/reverse-SynthID

#48 Critical: channel weight misalignment in detect_from_v4_codebook silently inverts Green/Blue priority, causing false positives

#47 -

#43 The watermark detector doesn't work with Gemini 2.5 and Gemini 3 pro

#24 Discussion: Theorised fixes and how they will be broken again

#20 Regarding pure black/white image generation

#15 Migrate image datasets to Hugging Face / Kaggle to reduce repo size

#3 Missing spectral_codebook.npz (119 MB) and reference images for V3 Bypass

#11 Does not work

#9 Bypass does not work at all

#10 Warning: Could not load extractor: persistent IDs in protocol 0 must be ASCII strings

#19 Multi-modal watermark dataset contribution: 1,113 files across image/audio/video (SynthID + C2PA)

#8 Bypass does not work on images generated by Google Labs Flow

#7 How to impove the detect accuracy?

#2 Could this process be used to add synthid to images

#6 Clarifications on Input Image Requirements (Size/Aspect Ratio) for Codebook Generation and High-Resolution Support