Zengtudor/Wan2.1
1
0
Fork 0
mirror of https://github.com/Wan-Video/Wan2.1.git synced 2025-11-04 14:16:57 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

fixed simple annoying bugs

Browse Source
This commit is contained in:
DeepBeepMeep 2025-09-16 22:38:45 +02:00
parent 7bcd724621
commit ae93d71217
4 changed files with 6 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@ -24,12 +24,12 @@ WanGP supports the Wan (and derived models), Hunyuan Video and LTV Video models
- The long awaited **Vace for Wan 2.2** is at last here or maybe not: it has been released by the *Fun Team* of *Alibaba* and it is not official. You can play with the vanilla version (**Vace Fun**) or with the one accelerated with Loras (**Vace Fan Cocktail**)
- **First Frame / Last Frame for Vace** : Vace model are so powerful that they could do *First frame / Last frame* since day one using the *Injected Frames* feature. However this required to compute by hand the locations of each end frame since this feature expects frames positions. I made it easier to compute these locations by using the "L" alias :
- **First Frame / Last Frame for Vace** : Vace models are so powerful that they could do *First frame / Last frame* since day one using the *Injected Frames* feature. However this required to compute by hand the locations of each end frame since this feature expects frames positions. I made it easier to compute these locations by using the "L" alias :
For a video Gen from scratch *"1 L L L"* means the 4 Injected Frames will be injected like this: frame no 1 at the first position, the next frame at the end of the first window, then the following frame at the end of the next window, and so on ....
If you *Continue a Video* , you just need *"L L L"* since the the first frame is the last frame of the *Source Video*. In any case remember that numeral frames positions (like "1") are aligned by default to the beginning of the source window, so low values such as 1 will be considered in the past unless you change this behaviour in *Sliding Window Tab/ Control Video, Injected Frames aligment*.
If you *Continue a Video* , you just need *"L L L"* since the first frame is the last frame of the *Source Video*. In any case remember that numeral frames positions (like "1") are aligned by default to the beginning of the source window, so low values such as 1 will be considered in the past unless you change this behaviour in *Sliding Window Tab/ Control Video, Injected Frames aligment*.
- **Qwen Inpainting** exist now in two versions: the original version of the previous release and a Lora based version. Each version has its pros and cons. For instance the Lora version supports also **Outpainting** ! However it tends to change slightly the original image even outside the outpainted area.
- **Qwen Edit Inpainting** exists now in two versions: the original version of the previous release and a Lora based version. Each version has its pros and cons. For instance the Lora version supports also **Outpainting** ! However it tends to change slightly the original image even outside the outpainted area.
- **Better Lipsync with all the Audio to Video models**: you probably noticed that *Multitalk*, *InfiniteTalk* or *Hunyuan Avatar* had so so lipsync when the audio provided contained some background music. The problem should be solved now thanks to an automated background music removal all done by IA. Don't worry you will still hear the music as it is added back in the generated Video.

1
models/qwen/qwen_main.py
View File

@ -217,6 +217,7 @@ class model_factory():
def get_loras_transformer(self, get_model_recursive_prop, model_type, model_mode, **kwargs):
if model_mode == 0: return [], []
preloadURLs = get_model_recursive_prop(model_type, "preload_URLs")
if len(preloadURLs) == 0: return [], []
return [os.path.join("ckpts", os.path.basename(preloadURLs[0]))] , [1]

2
models/wan/any2video.py
View File

@ -146,7 +146,7 @@ class WanAny2V:
from mmgp.safetensors2 import torch_load_file
else:
if self.transformer_switch:
if 0 in submodel_no_list[2:] and 1 in submodel_no_list:
if 0 in submodel_no_list[2:] and 1 in submodel_no_list[2:]:
raise Exception("Shared and non shared modules at the same time across multipe models is not supported")
if 0 in submodel_no_list[2:]:

2
wgp.py
View File

@ -62,7 +62,7 @@ AUTOSAVE_FILENAME = "queue.zip"
PROMPT_VARS_MAX = 10
target_mmgp_version = "3.6.0"
WanGP_version = "8.6"
WanGP_version = "8.61"
settings_version = 2.35
max_source_video_frames = 3000
prompt_enhancer_image_caption_model, prompt_enhancer_image_caption_processor, prompt_enhancer_llm_model, prompt_enhancer_llm_tokenizer = None, None, None, None