This is the first time I have synthesized wan2.2's lora. There might be many problems, but some decent results can also be achieved. (I think the effect is quite good QWQ)
The demonstration effect is that this model was mounted in both the high-sound-diffusion and low-noise modes in the comfyui. The high-sound-diffusion intensity was set to 1.0 and the low-noise intensity to 1.5. (I'm not sure if this is the best effect, but I found it quite satisfactory after using it.)
After my test, both 16:9 and 9:16 have a good effect.
(The displayed prompt words might be a little off and deviate too much, which might cause mistakes. Of course, Wan2.2 supports English. So, let's use Chinese for the trigger words.)
The workflow I'm using is shown here with the arrow ↓ https://civarchive.com/articles/18232/wan22lightning-workflow
Perhaps this model is not perfect. If there's time, we can try to update it again.
Thank you for using it. Have a great time! : )
这是我第一次炼制的wan2.2的lora，可能会存在许多问题，但是也能出现一些还不错的效果。（我感觉效果还不错QWQ）
演示效果是在comfyui中高燥和低噪中都挂载了此模型，高燥强度为1.0 低噪强度为1.5（我不确定这是最好的效果但是我使用下来感觉还不错）
我测试下来16:9和9:16都有一个不错的效果
（展示的提示词可能会有点偏差跑太多可能记错了，当然wan2.2支持英语，触发词还是用中文吧）
我使用的工作流我在这里箭头↓
https://civarchive.com/articles/18232/wan22lightning-workflow
可能这个模型并不完善，有时间的话再重新尝试更新一下:）
感谢你的使用，祝你玩的开心:）
Description
FAQ
Comments (7)
支持
This is absolutely amazing! I was really surprised—wan2.2's t2v has always been a bit weak when it comes to proper anime output, but this LoRA completely overcomes that limitation. If the character exists in the dataset, it can be perfectly reproduced in anime style, with camera movement and facial expressions flowing naturally. Truly a game changer!
Thank you so much for creating such an incredible LoRA!
PS: May I ask how you created this LoRA? That information would be extremely valuable!
Thank you for your appreciation. The result was obtained by using 730 image materials and running for about 18 epochs, with approximately 8,000 steps.
I'm having trouble getting it to be consistently anime, unpredictably realistic. Could you tell me how you set it up to get reliable results?
I'm struggling to get this to be consistent. I'm using the trigger words and I've tested with some of the examples and those work but then other descriptions jump back to realistic. Any suggestions?
I haven't tested this Lora, but in its current state I can halfway guess the source of the problem you describe. I'd say it's more the fact that it's only being one Lora, while WAN2.2 expects a High and a Low Lora variant to work properly.
I think the creator of this Lora was expecting that things work like in 2.1 still. Only one Lora needed there. Sadly there aren't a lot of anime loras around for 2.2 yet.
If you already tried his suggested workflow, there's little to add from my end, as I never tried it. Maybe ask the other commenters having had success, if they were willing to share their workflow.
效果很棒。提个小建议：请不要在文件名中包含非英文字符和空格，这会给很多人造成困扰。
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.