Generate clean 2D game sprites & animation atlases — component-row pipeline: state rows, alpha cleanup, frame extraction, runtime atlases. Codex/Claude skill.
Commits on main
Merge pull request #172 from aldegad/plan/prop-plate-side-face-1107
35a27e1 darkest_alex committed 12 days ago
docs(changelog): 4.3.0's upgrade note says runs without --prop-plate keep 4.2.0's bytes but for the engine fingerprint implementation_sha256, which hashes every file in sprite_gen/video
c284e1b Soohong Kim committed 12 days ago
docs(prop-plate): the plate's shaded side and top go with it, read as shades of prop_plate.rgb joined to the front, the edge rule measured from what went, the new report rows and warning, and CHANGELOG Unreleased (v4.3.0)
e1d3a39 Soohong Kim committed 12 days ago
Merge pull request #171 from aldegad/plan/cycle-source-on-fill-1107
b9b7024 darkest_alex committed 12 days ago
docs(cycle-source): video-loop keeps a filled loop's cut with the fill off as cycle.source/ and the alignment aligns from it without the clip's frames, its cycle_source record and stderr lines, what to carry when aligning elsewhere, and CHANGELOG Unreleased (v4.2.0)
1c8ef55 Soohong Kim committed 12 days ago
fix(cycle-source): video-loop keeps a filled loop's cut with the fill off as cycle.source/, so video-cycle-align aligns it from there without the clip's frames, the same alignment as cutting it again
0d1a71c Soohong Kim committed 12 days ago
Merge pull request #170 from aldegad/plan/prop-plate-track-1107
6a3f4c4 darkest_alex committed 12 days ago
docs(prop-plate): the plate found in each frame and erased there with the line its edge blends into the key, the frames report's rows and the two warnings, the palette learned without it, and CHANGELOG Unreleased (v4.1.0)
c889b85 Soohong Kim committed 13 days ago
fix(prop-plate): video-frames finds the plate in each frame, moved or turned, and erases it and the line its edge blends into the key there, names the frames it was not found in or meets the subject in, and learns the palette without it
8a9bd0f Soohong Kim committed 13 days ago
feat(decontam): the palette can be learned without pixels the caller marks as not the subject's (`not_subject`, `decontam_not_subject`), None keeping every output as it was
2af587d Soohong Kim committed 13 days ago
Merge pull request #167 from aldegad/plan/no-repeat-by-default-1101
c30322b darkest_alex committed 13 days ago
docs(changelog): the v4.0.0 section shows the before and after GIF of a loop's wrap and lists the two alignment tests' timeout, pins and slow mark
e192c38 Soohong Kim committed 13 days ago
test(held-fill): the two alignment tests the fill at the end sends to the binary pin its calls by model, and the one left in the fast bundle is marked slow
24213df Soohong Kim committed 13 days ago
test(held-fill): two alignment tests the fill at the end sends to the binary take a timeout of their own with their count of calls
8f40ade Soohong Kim committed 17 days ago
test(no-repeat): the set tests film their front and back walks from the base as given, so no provider is called
5c739d2 Soohong Kim committed 17 days ago
docs(video-chain): section 9 of the video pipeline with what it takes from sprite-gen, that which faults refuse a frame is the caller's to read off the material, the routing table's video-chain and inspect-holds rows, the verb lists, and the CHANGELOG's Unreleased (v4.0.0) line
e22ff0e Soohong Kim committed 17 days ago
feat(video-chain): pieces of keyed clips on one canvas played as one move, where they join and when each frame shows being the caller's, the frames between, no picture shown again and the record the engine's
f2dba1a Soohong Kim committed 17 days ago
docs(no-repeat): asset-tools "Frames shown again" (the speck floor never above half a frame's largest part, a box that holds nothing, `steps` and `inspected`, a reading of nothing exits 2), video.md's clip of the canvas alone, and the CHANGELOG's Unreleased (v4.0.0) line
a65a421 Soohong Kim committed 17 days ago
fix(video): a published clip whose every frame is the canvas alone is not read (`held` read false, with why), where it read as one hold the motion means and said nothing
b0ab67a Soohong Kim committed 17 days ago
fix(inspect-holds): a figure drawn in a few pixels is read, and a reading that reads no frame against the frame before is an error, never a gate passed
d2e1574 Soohong Kim committed 17 days ago
test(no-repeat): RIFE calls pinned again where the default fill now makes real repeats; three tests read the fill
1532deb Soohong Kim committed 17 days ago
perf(cycle-align): the fill at the end takes a frame between the loop's own resampling made, not RIFE again
5dcc23a Soohong Kim committed 17 days ago
fix(held-fill): the read after the repairs keeps a run where the cut as it now plays reads drawn on twos or threes
014e61d Soohong Kim committed 17 days ago