license: mit
pretty_name: MAGICK Transparency (WSM v1 raw)
task_categories:
- image-segmentation
- image-to-image
tags:
- transparency
- alpha-matting
- layer-diffuse
- magick
- wsm
MAGICK Transparency — raw WSG samples
Raw (pre-VAE) samples for the WSM v1 transparency-estimation pilot
(magick_transparency_v1): predict a foreground alpha / transparency layer from an RGB
image. The matching FLUX.2 latents live in the companion latent repo
World-Snapshot/WSM_v1_Latent_PT_WSG_v1.8_for_Flux2
(custom_pilots/magick_transparency_v1/cache_named/).
Source
Built from the MAGICK RGBA asset set (transparent foreground objects). Each foreground is composited over random backgrounds so the model must recover the soft alpha of the object from a realistic RGB image.
How each sample is produced
Each self-describing named-key .npz holds two modalities (uint8, H×W×3; the trainer maps
pixels with x/127.5 − 1):
| key | meaning |
|---|---|
0-2 |
RGB — the foreground composited over a random background. |
LayerDiffuse_Transparency |
target — the RGB-replicated soft alpha / transparency of the foreground (1-channel alpha stored replicated to 3 channels). |
This is an estimation task: RGB → transparency. No caption, no reference image.
Layout
wsg_named/ the named-key .npz samples (0-2 + LayerDiffuse_Transparency)
split_manifest.jsonl records the deterministic train/val/test assignment used by WSM v1 training
(see split_summary.json); the split is per-sample (npz path), reproducing the trainer's hash split
at val_split_ratio=0.05.
Packaging
Streamed as multi-part MAGICK_Transparency_partNNN.tar.zst (each < 15 GB) with
..._parts_manifest.jsonl (path → part index) and ..._parts_summary.json.
Extract
mkdir -p data/custom_pilots/magick_transparency_v1
for f in MAGICK_Transparency_part*.tar.zst; do
tar -xf "$f" -C data/custom_pilots/magick_transparency_v1/
done
# -> data/custom_pilots/magick_transparency_v1/wsg_named/
License
MIT for the packaging/processing; underlying MAGICK assets follow their original terms.