v0.4.1

Changed

• Downloads say what each zip holds. The panel leads with "Line art · all series" (line art and photos, chips with
  the counts), then a new shisei-complete zip with every format (69 MB), one zip per map format
  (shisei-clay, shisei-depth, shisei-normal: depth and normal were one shisei-controlnet zip, but people pick
  one ControlNet and the normals are twice the size), the series zips marked line art + photos, and the originals
  marked line art + photos at full size, with a note that the maps have no originals. The hero download button no
  longer shows the line-art zip's size as if it were everything. README download tables gain a contents column.
• Chinese copy: 白模 now only names the clay format; the mannequin line art is 素体 (hero title 姿势线稿参考, page
  title, share card, pose-page descriptions, README). Every language's hero line mentions the three map formats.

v0.4.0

Added

• Clay, depth and normal maps for 944 line views (270 sets), made from the line art and pixel-aligned with it
  (Qwen-Image-Edit clay, Depth Anything V2 depth, Marigold normals; docs/GENERATION.md). Catalog maps: {view: file},
  images/{clay,depth,normal}/ in the three tiers, QC round 5 in qc.json maps: an alignment score
  (tools/map_align.py), a five-panel review (review_sheets.py --mode maps, docs/qc/maps_prompt.md) and a clay
  close-up for nipple marks (--mode mapclay, maps_nipple_prompt.md); build_images.py maps=…, verify.py and
  qc_merge.py handle them. Of 1,212 reviewed views, 268 fail (misaligned upper-body renders, modelled faces or nipples).
• Lightbox: the line / photo / compare switch adds clay, depth and normal, each with a one-line note; a slider wipes the
  line art in over a map. Downloads get _clay / _depth / _normal names, deep links #P001-diag-clay. The home
  marquee mixes the three formats in, labelled. The grid and its filters are unchanged. Strings in all four languages.
• Release zips shisei-clay and shisei-controlnet (depth + normal), all series, w900; the download panel and the
  READMEs list them under "Other formats".
• Costume library data and images (D-10, D-11; the gallery pages come later): 494 costume cards in 12 categories in
  data/costumes.json, generated by Muse as ghost-mannequin product photos in four views, matted locally and QC'd
  per view. 1870 of 1976 views pass (482 items); WebP on the house grey #D9D9D9 in three tiers plus a transparent
  w900 (images/costume/). The lingerie category LG is all suggestive. Tools: costume_cards.py, costume_brief.py,
  costume_intake.py, costume_sheets.py, costume_qc.py, costume_build.py; docs/costume/PLAN.md.
• Phones: the lightbox download button reads "Save to Photos" and hands the w900 image to the system share sheet
  (its "Save Image" goes to the photo album; <a download> only reached Files / Downloads). Browsers without file
  sharing, such as WeChat, get a long-press hint over the image instead. Strings in all four languages.
• Single-view fills from the original generator (Muse, round fill-r1): import_fills.py stages them, catalog
  fills: {view: source} points each view at its fill source, and build, review, dups, merge, pack and verify resolve
  the source per view. Batches 01–19: 79 of 126 delivered views passed QC; std now has 300 of 365 views, upper 359
  of 400, sexy 352 of 400, duo 201 of 275 (docs/GENERATION.md).
• Pose descriptions (desc_en, the copyable prompt in the lightbox) for all 215 sexy, upper and duo poses, taken
  verbatim from the descriptions the views were generated from (Muse prompt export); every pose now has one.
  import_descs.py fills them, import_series.py keeps them on re-import and verify.py requires one per pose.
• Home hero: a "making of" badge above the title links the X article on how the library was built; it replaces the
  SHISEI · 姿勢 eyebrow, strings in all four languages.

v0.3.0

280 pose sets in four series, each drawn from up to five angles: 1,133 line-art views and 89 photo renders, every one QC-passed. New since v0.2.0: 43 sets redrawn as whole sets (std now shows 71 of 73, upper 80 of 80), a line-art / photo / compare lightbox, and blurred instead of hidden suggestive poses.

Images and data are now CC0 1.0 (LICENSE-images); code stays Apache-2.0.

Downloads

• shisei-all-v0.3.0.zip — every series, 900×1200 WebP (recommended, 34.1 MB)
• shisei-std / upper / sexy / duo — one series each (sexy and duo are suggestive)
• shisei-original-v0.3.0.zip — every series at source resolution (144.8 MB)

Each zip includes LICENSE, LICENSE-images and NOTICE; .sha256 files sit next to them. Full notes in CHANGELOG.md.

Gallery: https://shisei.indaloop.app

v0.2.0

256 pose sets in four series, each drawn from up to five angles. Every image passed QC.

Downloads

• shisei-all-v0.2.0.zip — every series, 900×1200 WebP (recommended, about 1/4 of the originals)
• shisei-std / upper / sexy / duo — one series each
• shisei-original-v0.2.0.zip — every series at source resolution

Each zip includes LICENSE and NOTICE (Apache-2.0); .sha256 files sit next to them. Full notes in CHANGELOG.md.

Gallery: https://shisei.indaloop.app