Commit History
Yesterday
-
Add WanGP PTD RefAsymOPD adapters and output heads
Verified signaturePublish eight scheduled RefAsymOPD adapters and the prepared output heads at the paths used by the full and pruned WanGP defaults. Include source attribution, license, and validation status.de1028e
2 days ago
-
Add latest H3 LightX2V Turbo LoRAs with explicit alpha tensors
Verified signatureAdd FL2V 4-step v1.3, FL2V 8-step v1.1, Ref2V 4-step v1.0 and Ref2V 8-step v1.1. Preserve original BF16 weights and add per-module alpha=8 tensors from the upstream metadata.8b3f78b
5 days ago
-
Add FastH3-Omni-8-Step-V1-pruned_rank8_bf16.safetensors
Verified signatureb575601 -
Add FastH3-Omni-8-Step-V1-pruned_rank8_int8_convrot.safetensors
Verified signatured0bc6d8 -
Add FastH3-Omni-NVFP4-QAD-pruned_rank8_nvfp4.safetensors
Verified signature007f3fe
11 days ago
-
Add TaoMate FL2VA 2026-10-15 and LightVAE Lite BF16/INT8
Verified signature630975d
14 days ago
-
Remove DMAD adapter and notice
Verified signature62e5db0 -
Add MiniMax H3 DMAD full-critic 4-step LoRA
Verified signatured785fd3
15 days ago
-
Add H3 Veda T2VA and R2VA predictors in BF16 and INT8 ConvRot
Verified signaturePrepared WanGP predictor modules for full and pruned H3 variants. BF16 weights preserve upstream dequantization; INT8 files use the shared ConvRot format. Source revisions and tile plans are retained in checkpoint metadata.49fe56a -
Add ConvRot file-level format metadata to prepared checkpoints
Verified signatureAdd quantization_format=int8_convrot and convrot_format=comfy_quant_v1. Preserve every tensor payload, dtype, shape and filename. Verified identical quantization detection, maps and runtime tensors; detection and conversion each perform zero per-layer comfy_quant JSON reads with the header markers.98b5288 -
Consolidate main history while preserving all current files
8d90934