@Aiwood check dm

@Aiwood 4 小段视频吗？ 好神奇

@Aiwood 感谢~~！

@Aiwood wan每秒影格不是16?用24格有特別用意?

@Aiwood 最近兩天練的都失敗了 謝謝你提供參數

@Aiwood 训练连贯动作是不是full模式比较好？请教下full和head一般分别在什么场景下用

I put it in the i2v folder but I think it doesn't detect it correctly, it doesn't appear for me to select this lora.

@solracnauj Wan2GP has been updated and supports now safetensors formats and replicate format

@Pleasant_Strain_2515491 Thanks for the info.

我也遇到一樣的問題

你应该贴出你的配置

@Aiwood I'm an idiot. I didn't choose Lora.

@Aiwood 模型 、cfg、shift、precision这些应该 怎么选 啊，我生出来的视频总是会变成灰蒙蒙的。

you need good gaming machine with good GPU(lotsa RAM).

download comfyui windows (ready made) unzip it and launch(mayby need update so wan works).

many examples to start

https://comfyanonymous.github.io/ComfyUI_examples

look wan and lora page

download workflow JSON or drag and drop image from that page to comfyui

download and move all model files and loras/vae/texencoders/etc.. where they need go

if image is saved with workflow comfyui understand it

if use examples from wan page only need add lora between model and positive/negative clip

in this case its phun hun dance lora

basically its that easy.

i havent try this but many other

@Aiwood thanks!

Is that landscape or portrait?

@amida_d it's always written as X, Y so 512 is X (width) and 320 is Y (height) so it's longest side would be top and bottom meaning that it's landscape.

Have you found a solution to this issue? I've tried settings from 1e-5 to 5e-5 but none worked well. I've tested many times, and this problem has been bothering me for 3 days. I'm using the diffusion-pipe model to train dance videos, but the results are very poor, completely incomparable to this

I was trying out with 2.2 and ran into lots of hallucinations. Lucky read this message and switch to 2.1. :) Thanks mate!

No need anymore just use animate. Some real good workflows out there, i got one that's specific for 12-16vram but actually runs on 8. Example https://civitai.com/images/101926778?postId=22611948

@1548431473200 maybe give me a downloadable link? using https://send.vis.ee/ ?

Look, some people just want to download the lora, bing bam boom, generate. All I can say is who ever makes a Wan2.2 version WILL get a ton of generations off of it, there's nothing on the market for the phut hon dance. Unless you go back to Wan2.1

@vgbestly no clue what your talking about, anyone with a dance video such as this one can throw it into animate and get the result in wan2.2. You literally pic a video and a person that should do the dance and bing bam boom generate. I'll even do a few just to show you and link it here using the workflow i shared above
https://civitai.com/posts/25564507

https://civitai.com/images/101920009

https://civitai.com/images/101926778

you cant get much simpler than this person do this dance.