Skip to content

v0.10.0

Choose a tag to compare

View all tags
@wiltodelta wiltodelta released this 09 Jun 20:24
· 37 commits to main since this release
v0.10.0
This tag was signed with the committer’s verified signature.
wiltodelta Victor Kuznetsov
SSH Key Fingerprint: O249fX54NelFGDT+oau8iOrSSN72WqsLtBEa7MuSnbg
Verified
Learn about vigilant mode.
826cfdb
This commit was signed with the committer’s verified signature.
wiltodelta Victor Kuznetsov
SSH Key Fingerprint: O249fX54NelFGDT+oau8iOrSSN72WqsLtBEa7MuSnbg
Verified
Learn about vigilant mode.

Highlights

Removal pipeline

  • ControlNet is now the default pipeline (text/face structure preservation). --pipeline sdxl opts down to plain SDXL for inputs without text/faces.
  • Unified vendor-adaptive strength (the certified ControlNet floors, same ladder for both pipelines): OpenAI 0.20 / Google 0.30 / unknown 0.30.
  • all and batch now remove the Doubao / Jimeng / Samsung visible text marks, not just the Gemini sparkle (they route through the watermark registry).
  • --adaptive-polish is ON by default (it self-gates to a no-op where there is no detail deficit). Disable with --no-adaptive-polish.

New CLI knobs

  • --model (HuggingFace model id) on invisible/all/batch.
  • --guidance-scale (CFG) on all diffusion commands.

Deprecations / renames (backward-compatible)

  • The plain-SDXL pipeline profile default was renamed to sdxl; default still works as a deprecated alias (warns).
  • --auto is deprecated and now a no-op (the adaptive polish it enabled is on by default).

Internal

  • Consolidated the Doubao/Jimeng/Samsung engines into a shared config-driven base (byte-exact behavior).
  • Fewer redundant file reads in identify, single decode/encode in the invisible post-processing, and a batch full-project code-review pass.

Full changelog: v0.9.0...v0.10.0

Assets 2
Loading