653ceab414
Reduce Peak WAN inference VRAM usage - part II ( #10062 )
...
* flux: math: Use _addcmul to avoid expensive VRAM intermediate
The rope process can be the VRAM peak and this intermediate
for the addition result before releasing the original can OOM.
addcmul_ it.
* wan: Delete the self attention before cross attention
This saves VRAM when the cross attention and FFN are in play as the
VRAM peak.
2025-09-27 18:14:16 -04:00
160698eb41
convert nodes_qwen.py to V3 schema ( #10049 )
2025-09-27 12:25:35 -07:00
7eca95657c
convert nodes_photomaker.py to V3 schema ( #10017 )
2025-09-27 02:36:43 -07:00
ad5aef2d0c
convert nodes_pixart.py to V3 schema ( #10019 )
2025-09-27 02:34:32 -07:00
bcfd80dd79
convert nodes_luma.py to V3 schema ( #10030 )
2025-09-27 02:28:11 -07:00
6b4b671ce7
convert nodes_bfl.py to V3 schema ( #10033 )
2025-09-27 02:27:01 -07:00
a9cf1cd249
convert nodes_hidream.py to V3 schema ( #9946 )
2025-09-26 23:13:05 -07:00
255572188f
Add workflow templates version tracking to system_stats ( #9089 )
...
Adds installed and required workflow templates version information to the
/system_stats endpoint, allowing the frontend to detect and notify users
when their templates package is outdated.
- Add get_installed_templates_version() and get_required_templates_version()
methods to FrontendManager
- Include templates version info in system_stats response
- Add comprehensive unit tests for the new functionality
2025-09-26 21:29:13 -07:00
0572029fee
Update template to 0.1.88 ( #10046 )
2025-09-26 21:18:16 -07:00
196954ab8c
Add 'input_cond' and 'input_uncond' to the args dictionary passed into sampler_cfg_function ( #10044 )
2025-09-26 19:55:03 -07:00
1e098d6132
Don't add template to qwen2.5vl when template is in prompt. ( #10043 )
...
Make the hunyuan image refiner template_end 36.
2025-09-26 18:34:17 -04:00
cd66d72b46
convert CLIPTextEncodeSDXL nodes to V3 schema ( #9716 )
2025-09-26 14:15:44 -07:00
2103e39335
convert nodes_post_processing to V3 schema ( #9491 )
2025-09-26 14:14:42 -07:00
d20576e6a3
convert nodes_sag.py to V3 schema ( #9940 )
2025-09-26 14:13:52 -07:00
a061b06321
convert nodes_tcfg.py to V3 schema ( #9942 )
2025-09-26 14:13:05 -07:00
80718908a9
convert nodes_sdupscale.py to V3 schema ( #9943 )
2025-09-26 14:12:38 -07:00
7ea173c187
convert nodes_fresca.py to V3 schema ( #9951 )
2025-09-26 14:12:04 -07:00
76eb1d72c3
convert nodes_rebatch.py to V3 schema ( #9945 )
2025-09-26 14:10:49 -07:00
c4a46e943c
Add @kosinkadink as code owner ( #10041 )
...
Updated CODEOWNERS to include @kosinkadink as a code owner.
2025-09-26 17:08:16 -04:00
2b7f9a8196
Fix the failing unit test. ( #10037 )
2025-09-26 14:12:43 -04:00
ce4cb2389c
Make LatentCompositeMasked work with basic video latents. ( #10023 )
2025-09-25 17:20:13 -04:00
c8d2117f02
Fix memory leak by properly detaching model finalizer ( #9979 )
...
When unloading models in load_models_gpu(), the model finalizer was not
being explicitly detached, leading to a memory leak. This caused
linear memory consumption increase over time as models are repeatedly
loaded and unloaded.
This change prevents orphaned finalizer references from accumulating in
memory during model switching operations.
2025-09-24 22:35:12 -04:00
fccab99ec0
Fix issue with .view() in HuMo. ( #10014 )
2025-09-24 20:09:42 -04:00
fd79d32f38
Add new audio nodes ( #9908 )
...
* Add new audio nodes
- TrimAudioDuration
- SplitAudioChannels
- AudioConcat
- AudioMerge
- AudioAdjustVolume
* Update nodes_audio.py
* Add EmptyAudio -node
* Change duration to Float (allows sub seconds)
2025-09-24 18:59:29 -04:00
341b4adefd
Rodin3D - add [Rodin3D Gen-2 generate] api-node ( #9994 )
...
* update Rodin api node
* update rodin3d gen2 api node
* fix images limited bug
2025-09-24 14:05:37 -04:00
b8730510db
ComfyUI version 0.3.60
2025-09-23 11:50:33 -04:00
e808790799
feat(api-nodes): add wan t2i, t2v, i2v nodes ( #9996 )
2025-09-23 11:36:47 -04:00
145b0e4f79
update template to 0.1.86 ( #9998 )
...
* update template to 0.1.84
* update template to 0.1.85
* Update template to 0.1.86
2025-09-23 11:22:35 -04:00
707b2638ec
Fix bug with WanAnimateToVideo. ( #9990 )
2025-09-22 17:34:33 -04:00
8a5ac527e6
Fix bug with WanAnimateToVideo node. ( #9988 )
2025-09-22 17:26:58 -04:00
e3206351b0
add offset param ( #9977 )
2025-09-22 17:12:32 -04:00
1fee8827cb
Support for qwen edit plus model. Use the new TextEncodeQwenImageEditPlus. ( #9986 )
2025-09-22 16:49:48 -04:00
27bc181c49
Set some wan nodes as no longer experimental. ( #9976 )
2025-09-21 19:48:31 -04:00
d1d9eb94b1
Lower wan memory estimation value a bit. ( #9964 )
...
Previous pr reduced the peak memory requirement.
2025-09-20 22:09:35 -04:00
7be2b49b6b
Fix LoRA Trainer bugs with FP8 models. ( #9854 )
...
* Fix adapter weight init
* Fix fp8 model training
* Avoid inference tensor
2025-09-20 21:24:48 -04:00
9ed3c5cc09
[Reviving #5709 ] Add strength input to Differential Diffusion ( #9957 )
...
* Update nodes_differential_diffusion.py
* Update nodes_differential_diffusion.py
* Make strength optional to avoid validation errors when loading old workflows, adjust step
---------
Co-authored-by: ThereforeGames <eric@sparknight.io>
2025-09-20 21:10:39 -04:00
66241cef31
Add inputs for character replacement to the WanAnimateToVideo node. ( #9960 )
2025-09-20 02:24:10 -04:00
e8df53b764
Update WanAnimateToVideo to more easily extend videos. ( #9959 )
2025-09-19 18:48:56 -04:00
852704c81a
fix(seedream4): add flag to ignore error on partial success ( #9952 )
2025-09-19 16:04:51 -04:00
9fdf8c25ab
api_nodes: reduce default timeout from 7 days to 2 hours ( #9918 )
2025-09-19 16:02:43 -04:00
dc95b6acc0
Basic WIP support for the wan animate model. ( #9939 )
2025-09-19 03:07:17 -04:00
711bcf33ee
Bump frontend to 1.26.13 ( #9933 )
2025-09-19 03:03:30 -04:00
24b0fce099
Do padding of audio embed in model for humo for more flexibility. ( #9935 )
2025-09-18 19:54:16 -04:00
1ea8c54064
make kernel of same type as image to avoid mismatch issues ( #9932 )
2025-09-18 19:51:16 -04:00
8d6653fca6
Enable fp8 ops by default on gfx1200 ( #9926 )
2025-09-18 19:50:37 -04:00
dd611a7700
Support the HuMo 17B model. ( #9912 )
2025-09-17 18:39:24 -04:00
9288c78fc5
Support the HuMo model. ( #9903 )
2025-09-17 00:12:48 -04:00
e42682b24e
Reduce Peak WAN inference VRAM usage ( #9898 )
...
* flux: Do the xq and xk ropes one at a time
This was doing independendent interleaved tensor math on the q and k
tensors, leading to the holding of more than the minimum intermediates
in VRAM. On a bad day, it would VRAM OOM on xk intermediates.
Do everything q and then everything k, so torch can garbage collect
all of qs intermediates before k allocates its intermediates.
This reduces peak VRAM usage for some WAN2.2 inferences (at least).
* wan: Optimize qkv intermediates on attention
As commented. The former logic computed independent pieces of QKV in
parallel which help more inference intermediates in VRAM spiking
VRAM usage. Fully roping Q and garbage collecting the intermediates
before touching K reduces the peak inference VRAM usage.
2025-09-16 19:21:14 -04:00
a39ac59c3e
Add encoder part of whisper large v3 as an audio encoder model. ( #9894 )
...
Not useful yet but some models use it.
2025-09-16 01:19:50 -04:00
1a85483da1
Fix depending on asserts to raise an exception in BatchedBrownianTree and Flash attn module ( #9884 )
...
Correctly handle the case where w0 is passed by kwargs in BatchedBrownianTree
2025-09-15 20:05:03 -04:00
rattus128
Alexander Piskun
Christian Byrne
ComfyUI Wiki
Jedrzej Kosinski
comfyanonymous
Yoland Yan
Guy Niv
Jukka Seppänen
Changrz
comfyanonymous
Kohaku-Blueleaf
Jodh Singh
DELUXA
blepping