Commit History
-
Add precompute_wd_tags.py and precompute_embeddings.py
Two-step pipeline for training data prep: 1. precompute_wd_tags.py — WD tagger (eva02-large-v3) → wd_tags.json 2. precompute_embeddings.py — SigLIP pools + Gemma embeddings → memmap files No TrueStream imports; settings as variables at top. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>db700ca -
Add ControlNet + IP-Adapter weights, HED detector, training scripts
- controlnet.safetensors — combined ControlNet (7 blocks) + IP-Adapter (step 10000) - ip_adapter.safetensors — IP-Adapter only weights - hed_detector.safetensors — HED edge detector (Apache-2.0) - config.json — model metadata - train.py — joint ControlNet + IP-Adapter training script - precompute_hed.py — dataset HED edge precomputation - control_maps.py, hed.py — utility modules - convert_to_safetensors.py — checkpoint conversion tool1d1bd5d -
initial commit
Verified signaturef1c36ea