kijai
9f84cd2254
Fix offloading when using stand-in lora
2025-08-31 18:45:13 +03:00
kijai
b9b46638ac
Fix FantasyPortrait + GGUF main model
2025-08-31 18:41:08 +03:00
kijai
b1c8b8a280
remove flex attention as redundant
2025-08-31 18:14:38 +03:00
kijai
79395b872c
Update pyproject.toml
2025-08-31 18:10:07 +03:00
kijai
cedac626e2
Update __init__.py
2025-08-31 18:09:24 +03:00
kijai
4870b8f081
Update __init__.py
2025-08-31 18:06:20 +03:00
kijai
81dac31735
Merge branch 's2v'
2025-08-31 18:05:30 +03:00
kijai
26a11c3044
Update nodes.py
2025-08-31 18:01:53 +03:00
kijai
10d3a0fbef
Update model.py
2025-08-30 23:34:14 +03:00
kijai
40308f1c70
Allow basic FantasyPortrait + S2V
2025-08-30 22:33:32 +03:00
kijai
df0b8c419a
Fix
2025-08-30 18:01:05 +03:00
kijai
45f129b287
Revert this for now to avoid issues on some systems
...
Prefetching still works, but using cuda stream caused issue on some systems, investigate reasons later
2025-08-30 14:09:16 +03:00
kijai
a7cf6d4ed1
Update nodes.py
2025-08-29 21:33:12 +03:00
kijai
7dd0e1ac61
Update nodes_model_loading.py
2025-08-29 21:21:24 +03:00
kijai
f36fa45c29
Loading fixes when using merged loras
2025-08-29 20:57:28 +03:00
kijai
3a290cfd15
Update nodes_model_loading.py
2025-08-29 20:41:40 +03:00
kijai
8f6507fa64
Update nodes.py
2025-08-29 18:10:18 +03:00
kijai
1e4cc84a30
Fix previous
2025-08-29 17:35:27 +03:00
kijai
88cf36936b
skip this if not on cuda
2025-08-29 17:29:25 +03:00
kijai
876b5bda5c
Update nodes.py
2025-08-29 17:17:37 +03:00
kijai
5f7a5d533b
Reduce memory use in the Framepack loop
2025-08-29 16:59:23 +03:00
kijai
6a2053c9d1
Update nodes_model_loading.py
2025-08-29 15:26:47 +03:00
kijai
0c8a883a36
Update nodes.py
2025-08-29 15:25:31 +03:00
kijai
044ff9f25e
don't use model management for cuda stream...
2025-08-29 15:21:49 +03:00
kijai
6978686272
Update model.py
2025-08-29 15:16:02 +03:00
kijai
740157bd61
Fix MultiTalk when using the Multi/InfiniteTalk node
...
InfiniteTalk was also mistakenly using only first frame mask... but honestly after trying the reference method, the result was worse, so I'll default to single frame mask in all cases for now.
2025-08-29 14:48:12 +03:00
kijai
704cca215b
Fix control lora when not last lora to be loaded
2025-08-29 14:45:03 +03:00
kijai
999a314652
Fix GGUF offloading
2025-08-29 12:58:36 +03:00
kijai
f9be754980
cleanup
2025-08-29 00:17:05 +03:00
kijai
a21e4b3210
Update model.py
2025-08-28 20:54:36 +03:00
kijai
469d9168da
Update wanvideo2_2_S2V_framepack_pose_testing.json
2025-08-28 19:31:47 +03:00
kijai
a053336c1a
Update nodes.py
2025-08-28 18:52:10 +03:00
kijai
6faa24b7f2
Implement Framepack long geneneration method
...
RoPE handling is comfyanon's code
2025-08-28 18:48:39 +03:00
kijai
a5621b8739
Add pose input
2025-08-27 22:43:28 +03:00
kijai
5266959a93
Create wanvideo2_2_S2V_context_window_testing.json
2025-08-27 21:41:34 +03:00
kijai
a37b12235f
Add loudness norm node
2025-08-27 20:51:08 +03:00
kijai
90c3bbb6c2
better context window indices, cleanup
2025-08-27 18:47:31 +03:00
kijai
4bc1ffee24
Update wanvideo_Fun_2_2_control_example_03.json
2025-08-27 16:12:31 +03:00
kijai
56bd2ec5f6
Update WanVideoScheduler -node
2025-08-27 13:19:07 +03:00
kijai
a507a5763d
update WanVideoScheduler -node
2025-08-27 13:03:24 +03:00
kijai
c89b82e479
Merge branch 'main' into s2v
2025-08-27 12:24:08 +03:00
kijai
e241998f53
Fix Uni3C
2025-08-27 12:21:25 +03:00
kijai
0538f5d600
Update nodes.py
2025-08-27 02:41:24 +03:00
kijai
a84d142cb2
Update nodes.py
2025-08-27 02:40:27 +03:00
kijai
5d17484cc5
continue
2025-08-27 02:25:39 +03:00
kijai
636b252c7e
Merge branch 'main' into s2v
2025-08-26 22:41:37 +03:00
kijai
2a3ca69c83
Fix offloading when using merged loras
2025-08-26 22:34:12 +03:00
kijai
e85e1ff9e6
Update model.py
2025-08-26 22:05:22 +03:00
kijai
8e65fae3c5
Update model.py
2025-08-26 22:02:36 +03:00
kijai
96f7f6accd
fix normal model loading
2025-08-26 22:02:02 +03:00
kijai
3c79851230
ref latent
2025-08-26 21:30:10 +03:00
kijai
63d4b6aada
Update nodes.py
2025-08-26 19:07:38 +03:00
kijai
748ec89aa8
init
2025-08-26 19:02:24 +03:00
kijai
a1ca0985ec
Merge branch 'dev'
2025-08-26 16:39:26 +03:00
kijai
30f185a19b
Update pyproject.toml
2025-08-26 16:39:17 +03:00
kijai
4f13aa46d4
Update nodes.py
2025-08-26 01:12:22 +03:00
kijai
dfb2a59211
Update nodes.py
2025-08-26 01:11:20 +03:00
kijai
5f5f8ae30a
Merge branch 'main' into dev
2025-08-26 00:33:49 +03:00
kijai
82d2ab44e5
update workflows
2025-08-26 00:27:08 +03:00
kijai
efcef40b00
fix name
2025-08-26 00:18:05 +03:00
kijai
6fce0e2d3b
Add Wav2VecModelLoader to load wav2vec2 from single .safetensors
...
https://huggingface.co/Kijai/wav2vec2_safetensors/
2025-08-26 00:09:46 +03:00
kijai
665ddd6811
Merge branch 'main' into dev
2025-08-25 19:35:13 +03:00
kijai
e836134b90
UniAnimate fixes and optimization
2025-08-25 19:35:05 +03:00
kijai
17f02a91bf
Merge branch 'main' into dev
2025-08-25 17:29:04 +03:00
kijai
4d52c10148
Faster Multi/InfiniteTalk model load
2025-08-25 15:06:31 +03:00
kijai
9c63d8d04f
Fix temporal_mask dtype
2025-08-25 14:59:43 +03:00
kijai
288a7e96a9
Add experimental scheduler node with sigma plot
2025-08-25 14:53:36 +03:00
kijai
d9def84332
Fix UniAnimate and MultiTalk loading
2025-08-25 14:09:08 +03:00
kijai
42df277488
Merge branch 'main' into dev
2025-08-25 12:51:46 +03:00
kijai
66d4c5b90d
Slice possible alpha channel away before encoding
2025-08-25 12:50:41 +03:00
kijai
bc22008b66
Fix InfiniteTalk v2v
2025-08-24 22:58:10 +03:00
kijai
c65f81d089
Handle missing faces better in FantasyPortrait detection
2025-08-24 20:50:11 +03:00
kijai
71c93f87fd
Update nodes.py
2025-08-24 20:28:26 +03:00
kijai
0c3030a7d8
Add a node to easier split sigmas, EasyCache fixes
2025-08-24 20:07:34 +03:00
kijai
3d9b52cb10
Fix: can't del the end image at this point...
2025-08-24 14:18:54 +03:00
kijai
61aa0e2ffd
Drastically reduce VRAM usage on I2V encode node.
2025-08-24 11:39:56 +03:00
kijai
8a151b5402
Multi/InfiniteTalk sampling loop cleanup and optimizations, support FantasyPortrait within the loop
2025-08-24 11:39:18 +03:00
Jukka Seppänen
54f745bb63
Update nodes.py
2025-08-24 02:03:32 +03:00
kijai
d39155213e
This can be float for compatibility
2025-08-24 01:06:53 +03:00
kijai
9b98413e1e
Update nodes.py
2025-08-23 21:00:48 +03:00
kijai
11f3e41e9a
Merge branch 'main' into dev
2025-08-23 21:00:45 +03:00
kijai
9076a3aae8
bump version
2025-08-23 20:57:41 +03:00
kijai
b7d7f9afe5
cleanup unianimate stuff
2025-08-23 18:59:21 +03:00
kijai
77d79f6878
This is redundant
2025-08-23 17:55:12 +03:00
kijai
9a2a13498a
Cleanup MultiTalk model code
2025-08-23 17:49:04 +03:00
kijai
73cff6ebab
Cleanup Multi/InfiniteTalk sampling loop some
2025-08-23 16:27:10 +03:00
kijai
5f521dc169
Add generic CreateScheduleFloatList to assist with LoRA scheduling
2025-08-23 13:04:19 +03:00
kijai
4eeaf1ea19
Update nodes.py
2025-08-23 01:06:57 +03:00
kijai
d0b9f2c907
Allow UniAnimate to work with unmerged LoRAs
2025-08-23 01:06:22 +03:00
kijai
948fdd369d
cleanup
2025-08-23 01:05:41 +03:00
kijai
b1ceaaeb2e
Support Wan22 VAE in WanVideoLatentReScale
2025-08-22 15:04:29 +03:00
kijai
bb5503c9bb
remove prints
2025-08-22 12:57:57 +03:00
kijai
ff1fe75919
RoPE ntk scaling
2025-08-22 12:48:12 +03:00
kijai
49951a85b6
Update nodes.py
2025-08-22 01:51:16 +03:00
kijai
bfee623d77
Pass original image to 2nd sampler for 2.2 diff diff
2025-08-22 01:48:35 +03:00
kijai
c2923068e6
Differential diffusion for Multi/InfiniteTalk long I2V as well
2025-08-22 01:24:16 +03:00
kijai
644f9aeaa3
Fix in multitalk sampling
2025-08-21 22:14:37 +03:00
kijai
65973b0251
gguf loading fixes
2025-08-21 22:13:07 +03:00
kijai
2dfaeb0093
Update nodes_model_loading.py
2025-08-21 19:07:11 +03:00
kijai
b4d2290859
Merge branch 'main' into dev
2025-08-21 18:42:47 +03:00
kijai