Ethanfel

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-28 10:37:13 +01:00

2f626d8a96 fix: use videoprism_lvt_public_v1_large with joint video-text forward

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 23:57:08 +01:00

1d8b9b59e0 debug: add DIT velocity diagnostic at t=1 to isolate DIT vs VAE quality issue

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 22:40:00 +01:00

8bf4a0c3fc debug: log conditioner output stats and T2A text feature stats

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 22:28:10 +01:00

477fe0f08f debug: add latent and audio stats logging to V2A sampler

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 22:13:25 +01:00

c0b7ccbcee fix: substitute empty_clip_feat for video features when no video present

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 22:06:42 +01:00

45633788a4 debug: add latent and audio stats logging to T2A node

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:56:08 +01:00

11457fc27a debug: fix VAE load_state_dict diagnostic — load into .model directly

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:53:27 +01:00

f2705b3063 debug: log weight load stats for diffusion and VAE checkpoints

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:40:36 +01:00

83a7f2787b feat: add debug_zero_video/sync toggles and feature stats logging to sampler

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:28:22 +01:00

140cc5ee9a feat: implement real Synchformer visual encoder (TimeSformer ViT-B/16)

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:21:41 +01:00

f99d2666e8 fix: interpolate sync_cond to match audio sequence length in transformer

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:20:02 +01:00

934a401633 perf: replace PIL+PNG frame files with direct ffmpeg stdin pipe

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:19:27 +01:00

b3ac9ab22f feat: log MP4 conversion time before subprocess spawn

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:13:43 +01:00

ca87c41a2e feat: add per-step timing to feature extraction logs

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 21:07:18 +01:00

63bd999dfa fix: switch to VideoPrism large (1024-dim) and fix Synchformer output shape

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 20:56:55 +01:00

20fb766ad2 fix: cast tensors to float32 before numpy() in feature save

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 20:49:57 +01:00

93120eb6b9 feat: auto-resolve synchformer checkpoint from prismaudio models dir

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 20:38:02 +01:00

b1a2ee594e fix: correct VideoPrism import (videoprism.models, not videoprism); add flax dep

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 20:33:47 +01:00

0f46e8359d feat: switch managed venv to jax[cuda13] for GPU feature extraction

Ethanfel pushed to master at Ethanfel/ComfyUI-SelVA

2026-03-27 20:27:33 +01:00

06f8dbbab4 feat: add hf_token input and HF_TOKEN env forwarding to feature extractor