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
ab06ac2f64
Closer to original EchoShot implementation, bugfixes
...
Turns out I never had the right way of using EchoShot and was always just accidentally using my existing rope splitting... which just worked with the EchoShot weights, this commit adds proper EchoShot and it's used when their prompt format is used, the previous behaviour can be restored by using my splitting method with " | " between the prompts. EchoShot example has been updated to reflect that.
2025-08-07 12:08:42 +03:00
kijai
6191451e2c
support qwen 2.5 7b
2025-08-06 21:02:39 +03:00
kijai
2a21c4f6ae
VAE cleanup and adjustments, add node to allow using native VAE
2025-08-06 17:28:50 +03:00
kijai
69dd4689bf
Make prompt extender use sampling and add seed option
2025-08-05 21:17:11 +03:00
kijai
92eac2a9e1
typo
2025-08-05 18:25:42 +03:00
kijai
9129c8cc1d
Revert dimension check until I learn how to math
2025-08-05 14:26:45 +03:00
kijai
bacb2a2053
Update nodes_model_loading.py
2025-08-05 01:44:42 +03:00
kijai
8e71286b6d
Allow torch.compile VAE decoder
...
Slight speedup especially for 5B...
2025-08-05 01:31:31 +03:00
kijai
0e9b1973b0
Add error indicating incompatible input size
...
More relevant now with 5B VAE requiring things to be divisible by 32
2025-08-05 00:35:52 +03:00
kijai
6ec463288f
Fix LoRA scheduling with multiple LoRAs
2025-08-04 23:47:02 +03:00
kijai
1b73c10ed5
Update nodes.py
2025-08-04 23:18:50 +03:00
kijai
fc7f504e85
Fix empty latents bug from previous commit
2025-08-04 23:16:25 +03:00
kijai
fe7e124329
More 5B features
2025-08-04 21:27:36 +03:00
kijai
78a7e20838
Fix lora schedule with multilora node
2025-08-04 16:51:53 +03:00
kijai
7862749cfe
Fix single text encode
2025-08-04 01:25:06 +03:00
kijai
267e86704b
Update gguf.py
2025-08-04 00:49:53 +03:00
kijai
7a00a875e7
LoRA scheduling on GGUFs too
2025-08-04 00:49:29 +03:00
kijai
3552478d02
Update nodes_model_loading.py
2025-08-04 00:04:21 +03:00
kijai
801f543a7a
Allow LoRA timestep scheduling
...
LoRA strength can now be a list of floats that represents the strength at any given step
2025-08-04 00:02:19 +03:00
kijai
88b5feb1ed
Fix VRAM management (again)
2025-08-03 22:35:24 +03:00
Jukka Seppänen
1654741cdf
Merge pull request #955 from SD-inst/error_propagation
...
Propagate errors to frontend
2025-08-03 22:03:30 +03:00
kijai
5cde6f2216
Fix T5 double RAM use
2025-08-03 12:28:25 +03:00
kijai
26c1911e41
Allow loading 5B controlnet
...
inference doesn't work yet
2025-08-03 12:17:10 +03:00
kijai
309a93c221
Update nodes_model_loading.py
2025-08-02 21:13:29 +03:00
rkfg
00875a65a6
Propagate errors to frontend
2025-08-02 20:29:55 +03:00
kijai
c93e69fb9a
Merge branch 'main' of https://github.com/kijai/ComfyUI-WanVideoWrapper
2025-08-02 19:53:26 +03:00
kijai
b5199ec7e0
Fix control lora
2025-08-02 19:53:20 +03:00
Jukka Seppänen
158b62dafa
Merge pull request #952 from SD-inst/tdqm-fix
...
Reduce tdqm spam
2025-08-02 19:46:40 +03:00
kijai
e9b537f23c
Update nodes.py
2025-08-02 19:35:02 +03:00
kijai
7074688ced
Add prompt extension with local Qwen
2025-08-02 19:27:26 +03:00
rkfg
ed247d78b6
Reduce tdqm spam
2025-08-02 16:38:27 +03:00
kijai
ce4ac89659
Add shift to DummyComfyWanModelObject
2025-08-02 10:40:42 +03:00
kijai
8035353cc0
Update pyproject.toml
2025-08-02 02:09:54 +03:00
kijai
99d012230f
Catch missing taew2_1.safetensors better
2025-08-02 02:09:42 +03:00
kijai
a9925496d5
update some workflows
2025-08-02 02:03:01 +03:00
kijai
907d1b0b0f
5B + context windows
2025-08-02 00:34:10 +03:00
kijai
25fe9ea642
handle cancellation better
2025-08-01 21:57:08 +03:00
kijai
168f3ccbee
Update nodes.py
2025-08-01 21:24:35 +03:00
kijai
d9be7b9c26
Allow other schedulers for 5B
2025-08-01 21:21:55 +03:00
kijai
e05864e751
Update nodes.py
2025-08-01 18:31:40 +03:00
kijai
83bb0e7f35
fp8 text encoder fixes
2025-08-01 17:42:32 +03:00
kijai
8321e19009
Fix single text encoder after previous change
2025-08-01 14:45:29 +03:00
kijai
5406a72f62
Improve text encoder cache and add alternative node which leaves nothing to memory after encoding
...
WanVideoTextEncodeCached -node will load T5 when the prompt is not found in the cache, then unload it completely.
2025-08-01 12:36:19 +03:00
kijai
9edab74562
Default non_blocking to False
...
too many RAM issues
2025-08-01 02:29:07 +03:00
kijai
7eebde487f
typo
2025-08-01 00:54:54 +03:00
kijai
945700d56e
5B fixes
2025-07-31 22:41:56 +03:00
kijai
5bacc50088
Add safety check for context windows to avoid crash on tensor slicing
2025-07-31 17:09:05 +03:00
kijai
b60814fb6b
Fix split prompting, get rid of diffusers on model class
2025-07-31 15:11:47 +03:00
kijai
10b67de349
Fix loading TAEW from extra paths
2025-07-31 13:47:59 +03:00
kijai