0.2.0 — Earthshine

By the faint structural light a workflow leaves behind, chandra can now read images that carry no generation recipe at all.

Added

• ComfyUI images that aren't generations — background removers, pose detectors, upscalers, mask extractors, anything with no sampler — are now handled instead of rejected. There's no recipe to extract, so chandra show/inject describe the workflow's operation pipeline — one line per output image, since one workflow can write several (a background remover emits both the cut-out and its mask, by different paths):

  LoadImage → InspyrenetRembg → SaveImage
  LoadImage → InspyrenetRembg → MaskToImage → SaveImage
  

  That goes into both the parameters text and the XMP description, so SD Prompt Reader / CivitAI and general viewers (Pix) show something useful rather than nothing.

Fixed

• chandra inject no longer crashes on a ComfyUI image with no resolvable prompt (a non-generation workflow, or a generation whose conditioning can't be traced). The XMP-description step raised TypeError on the missing prompt text.

Full changelog: https://github.com/Technologicat/chandra/blob/main/CHANGELOG.md

chandra 0.1.1

Fixed

• chandra inject: the XMP description (the recipe general image viewers such as Pix show) now reports the same step count as the parameters chunk the SD tools read. A dynamic-steps chain that resolves to a fraction (e.g. 5.6) is truncated to the integer that actually ran (5); previously the parameters chunk truncated but the XMP showed the raw fraction, so the two layers disagreed.

Internal

• CI hardened: all GitHub Actions pinned to commit SHAs and GITHUB_TOKEN scoped to least privilege.
• chandra/xmp.py: the XMP packet's byte-order-mark is now written as an explicit \uFEFF escape in the source rather than a literal (invisible) character. Output bytes are unchanged.

Full changelog: https://github.com/Technologicat/chandra/blob/main/CHANGELOG.md