kijai
c9daa36ef7
Update wanvideo_Stand-In_reference_example_01.json
2025-08-14 13:31:53 +03:00
kijai
5691ae5314
Refactor Fun control input
...
Allows using reference only
2025-08-14 13:03:51 +03:00
kijai
64a0c03673
Fix Fun 2.2 reference image input
...
Turns out the reference image doesn't work in the initial fp8_scaled models I shared due to the ref_conv layer ending up scaled as well, I've fixed the models and added error indicating the issue when trying to use reference image on such bugged model.
No change in behaviour when using start image instead.
2025-08-14 11:13:03 +03:00
kijai
e1c052ff43
Update attention.py
2025-08-14 00:11:02 +03:00
kijai
8db5521463
Possible bandaid for q k v dtype mismatch
...
Still no clue where or why it happens, but clearly is happening for some when using sage with Stand-In
2025-08-14 00:10:38 +03:00
kijai
54dbcacd57
Fix fun control
2025-08-13 21:20:05 +03:00
kijai
4805b70d3b
Fix end_image only for 2.2 I2V
2025-08-13 21:04:08 +03:00
kijai
68e18eabb2
Create wanvideo_Fun2_2_control_camera_example_01.json
2025-08-13 19:33:25 +03:00
kijai
db86a1b838
Update nodes.py
2025-08-13 19:26:48 +03:00
kijai
00bcd1ce8b
Support Fun 2.2 Control-Camera
...
https://huggingface.co/alibaba-pai/Wan2.2-Fun-A14B-Control-Camera
https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/upload/main/Fun
2025-08-13 18:20:00 +03:00
kijai
b08f10ffd8
Re-implement Stand-in kv_cache
2025-08-13 11:58:22 +03:00
kijai
d596901055
Allow stand-in with VACE
...
Don't know how well it can actually work, has some effect
2025-08-13 10:00:37 +03:00
kijai
f933ad0840
Fix stand-in with GGUF
2025-08-13 01:05:22 +03:00
kijai
083237051f
I suppose kv_cache isn't actually used for this...
2025-08-13 00:07:14 +03:00
kijai
8319c961c0
Update model.py
2025-08-12 22:10:57 +03:00
kijai
5420c5aad9
Update nodes.py
2025-08-12 22:10:50 +03:00
kijai
987e92cc43
Create wanvideo_Stand-In_reference_example_01.json
2025-08-12 22:10:46 +03:00
kijai
a5ae70624d
Support Stand-In LoRA
...
https://github.com/WeChatCV/Stand-In
2025-08-12 21:51:38 +03:00
kijai
2c854c53ee
version
2025-08-12 21:48:15 +03:00
kijai
06863fe2d2
Allow doing T2V with the I2V encode node when no images are connected and allow using native set latent noise mask -node
...
Latent masking still doesn't work properly
2025-08-12 11:47:33 +03:00
kijai
97da701379
Fix DF sampler
2025-08-11 18:46:46 +03:00
Jukka Seppänen
bf84e99dec
Merge pull request #1019 from UaRuairc/patch-1
...
Fix res_multistep.step to set sigma_next correctly when ending early
2025-08-11 16:24:38 +03:00
kijai
057ee642f6
Add async prefetch as option for block swapping
...
Can speed up block swapping on some systems, added debug option as well so you can see if enabling the option gives you any benefit.
2025-08-11 16:17:16 +03:00
FB
1e091d181b
Fix res_multistep.step to set sigma_next correctly when ending early
...
Use the next sigma from the schedule when sampler ends early, only assume `sigma_next = 0` when on the true final step.
2025-08-11 14:17:08 +01:00
kijai
7c93aea182
rename fp8 optimization in code for clarity as it does more than that with loras
2025-08-11 14:32:57 +03:00
kijai
341932b58f
Fix unmerged lora application on unquantized models
2025-08-11 01:34:25 +03:00
kijai
d6270ace0a
Merge branch 'pr/1013'
2025-08-10 21:35:41 +03:00
kijai
abd350bbf4
Actually catch interrupt to free memory
2025-08-10 21:07:13 +03:00
Mel Massadian
19e88b0293
Merge branch 'kijai:main' into feat/encode_progress
2025-08-10 15:37:02 +02:00
kijai
7e474b6b16
Fix pusa extension
2025-08-10 01:34:16 +03:00
kijai
53e740a1e6
Update nodes.py
2025-08-09 16:24:30 +03:00
kijai
8185e9b3cb
Better fp8 linear layer patching with torch.compile
2025-08-09 16:18:33 +03:00
denk
c4c06d2c94
Revert "renamed controlnet workflows"
...
This reverts commit 31ae7a6227
2025-08-09 16:09:23 +03:00
denk
31ae7a6227
renamed controlnet workflows
2025-08-09 14:56:52 +03:00
denk
af0b4d9a29
Fix controlnet bug for wan2.2 and add workflow examples
2025-08-09 13:38:19 +03:00
kijai
cf8e403f88
Fix gguf
2025-08-09 13:28:20 +03:00
kijai
0da602f8e9
silence the img key error spam when using 2.1 I2V loras on 2.2 or T2V
2025-08-09 12:17:55 +03:00
kijai
99c946ae43
Fix lora not always working with compile
2025-08-09 11:20:30 +03:00
kijai
f0e81d01c9
prints
2025-08-09 10:52:27 +03:00
kijai
b57e245beb
Update wanvideo_Fun_2_2_control_example_01.json
2025-08-09 10:49:52 +03:00
kijai
ceedc9e493
Update nodes_model_loading.py
2025-08-09 10:49:42 +03:00
kijai
48fa904ad8
fp8 matmul for scaled models
...
Fp8 matmul (fp8_fast) doesn't seem feasible with unmerged LoRAs as you'd need to first upcast, then apply LoRA, then downcast back to fp8 and that is too slow. Direct adding in fp8 is also not possible since that's just not something fp8 dtypes support.
2025-08-09 10:17:11 +03:00
kijai
1757847e5f
bump version
2025-08-09 10:13:05 +03:00
Mel Massadian
a4871802ad
feat: ✨ add progress for VACE_Encode
...
allows to cancel earlier
2025-08-08 16:25:31 +02:00
kijai
a64d8bb0df
Fix fun control 2.2 timestep scheduling
2025-08-08 17:03:01 +03:00
kijai
f82d47273c
Update wanvideo_Fun_2_2_control_example_01.json
2025-08-08 16:26:26 +03:00
kijai
9be0eb133c
Fix lora scheduling in multilora loader
2025-08-08 13:47:08 +03:00
kijai
db93edb665
Create wanvideo_Fun_2_2_control_example_01.json
2025-08-08 13:42:07 +03:00
kijai
e1acc2c4e7
Support Fun 2.2
2025-08-08 11:49:04 +03:00
kijai
d446e97309
Experimental RAAG (Ratio Aware Adaptive Guidance) implementation
...
https://arxiv.org/abs/2508.03442
Available in experimental_args, alpha > 0 = enabled, default value is 1.0
2025-08-08 00:17:32 +03:00
kijai