From 742a0dccabdea7afde200a14b939f891eb282a73 Mon Sep 17 00:00:00 2001 From: Victor Kuznetsov Date: Sun, 27 Sep 2026 21:24:59 -0700 Subject: [PATCH] Align the remaining klein and color-restoration wording with the results module-internals still called FLUX.2 klein pending; the sweep still counted three inconclusive Gemini rows where only one was an explicit tool verdict, and the removal page stated a Google chroma leak the data only warns of. Co-Authored-By: Claude Opus 5.5 --- docs/module-internals.md | 8 +++++--- docs/research-sweep-2026-09.md | 11 ++++++----- docs/synthid-removal-research.md | 10 ++++++---- 3 files changed, 17 insertions(+), 12 deletions(-) diff --git a/docs/module-internals.md b/docs/module-internals.md index 0abd02d..ef07106 100644 --- a/docs/module-internals.md +++ b/docs/module-internals.md @@ -2238,9 +2238,11 @@ runs the same two-stage recipe on a Chroma1-HD (`lodestones/Chroma1-HD`, Apache-2.0) global pass through diffusers' `ChromaImg2ImgPipeline`. It is the answer to issue #88's FLUX.2 request through the model that actually exposes strength-controlled img2img in that family. diffusers has since added -`Flux2KleinInpaintPipeline` with a real strength path, so the earlier "FLUX.2 -is not integrable" conclusion is superseded pending a measured FLUX.2 klein -stage; the full research record is +`Flux2KleinInpaintPipeline` with a real strength path. A FLUX.2 klein 4B global +stage was measured on 2026-09-27 and not adopted: the stock pipeline feeds the +clean source back as reference tokens and keeps SynthID, and without them klein +clears both vendors only at lower face and CJK-text fidelity than `qwen-zimage`. +The full research record is [`chroma1-engine-research.md`](chroma1-engine-research.md). The profile also implements the shared verified-text donor hook through its already diff --git a/docs/research-sweep-2026-09.md b/docs/research-sweep-2026-09.md index 20898f4..5e78280 100644 --- a/docs/research-sweep-2026-09.md +++ b/docs/research-sweep-2026-09.md @@ -56,16 +56,17 @@ This page is an index. Each finding lives in its canonical page. the content, in `ReservedCode1` and `ReservedCode2`. The library accepts those fields as label evidence but does not verify the signature. - **Experiments.** Chroma restoration after regeneration (arXiv:2508.21072) was - measured on 2026-09-27 and is not shipped: transplanted source chroma left - three of four Gemini checks inconclusive and ghosts redrawn objects + measured on 2026-09-27 and is not shipped: transplanted source chroma drew one + explicit "inconclusive" and two unclear Gemini answers where the untouched + outputs were all clean, and ghosts redrawn objects ([removal research](synthid-removal-research.md#color-restoration-after-regeneration-2026-09-27)). FLUX.2 klein 4B was measured the same day: stock inpainting copies the source through its reference tokens and keeps OpenAI SynthID, and without - them it clears both vendors only at lower fidelity than `qwen-zimage` once - its Google operating point carries the usual cross-source margin + them it clears both vendors only at lower face and CJK-text fidelity than + `qwen-zimage` ([Chroma1 research](chroma1-engine-research.md#flux2-klein-as-a-global-stage-2026-09-27)). Still open: - Z-Image-Turbo as global stages (with an Apple Silicon arm through mflux), + Z-Image-Turbo as a global stage (with an Apple Silicon arm through mflux), linear probes on newer frozen backbones for `classify`, Resemble Perth and Meta PixelSeal as benchmark oracles, re-watermarking as a removal primitive (arXiv:2605.16796, MIT code), RAVEN view-synthesis removal diff --git a/docs/synthid-removal-research.md b/docs/synthid-removal-research.md index fe3f131..3e1f580 100644 --- a/docs/synthid-removal-research.md +++ b/docs/synthid-removal-research.md @@ -276,10 +276,12 @@ output `a,b`) and `AB` (output `L`, source `a,b`). - At 0.35 the regeneration redraws objects, and transplanted source chroma paints the old objects' colors onto the new ones as visible ghosts. -Not shipped. Source chroma is not a safe channel for Google's mark. The -Y/Cb/Cr scramble splits above placed OpenAI's mark in luma; for Google either -the decoder reads more than luma or the Lab-to-RGB conversion carries source -structure back into it, and these arms do not separate the two. A +Not shipped. Source chroma is not shown safe for Google's mark: one explicit +"inconclusive" against four clean untouched outputs is a warning, not a +measured leak. The Y/Cb/Cr scramble splits above placed OpenAI's mark in luma; +if Google's does come back, either its decoder reads more than luma or the +Lab-to-RGB conversion carries source structure back into it, and these arms +do not separate the two. A structure-gated transplant restricted to OpenAI is the only arm left open, and its gain would be color fidelity at a floor that already clears with margin. Hash-bound batch and verdicts: