What are your WAN Lora Training and Dataset Captioning Settings?
Hi malcolm, i was just interested in what your wan 2.1 character lora training settings were. Also, do you use videos in your datasets for these characer loras or only images? How do you caption your datasets? for some wan2.1 character loras, i see some people just write 'woman' for every image/video of the dataset, or write nothing at all.
Thanks in advance :D
interested as well. I wonder if training a character lora on i2v would help alot more when doing i2v where the character is not present on the image (i.e: a background image as starting image and the character enter the frame). When I use my t2v lora doing that, the character is less than 80% what is should look like.
Hey guys!
Nothing changed and what I wrote here still stands: https://civitai.com/articles/19686/wan-training-loras-workflows-thoughts
I do no captions for characters (only styles/concepts)
For WAN there are only images (25~).
Yes, the token is not really needed, you can write "woman", 'person' and it will work :)
@FireBurst699 I don't think I follow, my loras can be used in both t2v and i2v and it worked fine for both (on 2.1 and 2.2), you can take at my example workflows.