this lora helps create slightly more correct anatomical features for any woman.
this is probably one of the first lora trains for a 1.3b model. so there is no place to look for settings or better variations yet. this train is hastily done and may not give a perfect result. nevertheless, it improves generations.
I will continue experimenting with this. I will be glad to get any feedback, especially on your generation results.
If you wanna help, you can send me a link to any cloud file with short videos or photos on this topic. or maybe something else, it's you turn.
Description
FAQ
Comments (26)
Are you planning to release an image-to-video (I2V) version for 1.4, or is it already compatible with 1.4?
are you about 14b i2v? if so it's already works fine with any kind of pictures
@6116091735 i got the same problem like @momox
@6116091735 does it work on gguf ? i load it on gguf version and i get this error ERROR lora diffusion_model.blocks.17.cross_attn.v.weight shape '[5120, 5120]' is invalid for input of size 2359296
ERROR lora diffusion_model.blocks.17.cross_attn.o.weight shape '[5120, 5120]' is invalid for input of size 2359296
ERROR lora diffusion_model.blocks.17.ffn.0.weight shape '[13824, 5120]' is invalid for input of size 13762560
ERROR lora diffusion_model.blocks.17.ffn.2.weight shape '[5120, 13824]' is invalid for input of size 13762560
ERROR lora diffusion_model.blocks.18.self_attn.q.weight shape '[5120, 5120]' is invalid for input of size 2359296
ERROR lora diffusion_model.blocks.18.self_attn.k.weight shape '[5120, 5120]' is invalid for input of size 2359296
ERROR lora diffusion_model.blocks.18.self_attn.v.weight shape '[5120, 5120]' is invalid for input of size 2359296 i thought that the problem is the resolution but it was not because i tried also bigger ones but nothing i had i2v 14b 480p Q5 K_S
@6116091735 do you have a solution for this ?
I thought that the lora works in 14b image2video
this lora is for 1.3b only. wan 14b and wan 1.3b are not compatible
@6116091735 ok i thought it worked since you told me in the first comment this
are you about 14b i2v? if so it's already works fine with any kind of pictures
Can't make it work in comfy native and kijai's wan_wrapper with Wan2_1-T2V-1_3B_fp8_e4m3fn.safetensors. It prints errors: lora key not loaded: diffusion_model.blocks.30.cross_attn.k.lora_A.weight, ERROR lora diffusion_model.blocks.9.ffn.0.weight shape '[8960, 1536]' is invalid for input of size 70778880. Other 1.3b loras from civitai work for me
i run it on wan2.1_t2v_1.3B_fp16.safetensors without any errors
use official wrappers...
upd: sorry, was using wrong lora file =)
fp8 seems to be the magic word of many Lora's to be working like poo, I recommend using a gguf if you cant play with the original one. For some reason the gguf's always play nice with them.
Would love to see a 14b version, just saying.
try this lora (WAN2.1-14B-compatible) https://civitai.com/models/1307155?modelVersionId=1475095
hi,which lora models work on WAN 2.1?
depends of wan size 1.3b or 14b and they aren't similar
try this lora (WAN2.1-14B-compatible) https://civitai.com/models/1307155?modelVersionId=1475095
How do I set the Lora weight and prompt words? I am not getting the results I want.
I like female genital anatomy
does not work on pinokio в wan2.1
which wan did you use?
Any chance for getting a similar Lora, but with male anatomy?
high chance but it's a lil bit harder
is there a 14b version? or can this work well with 14b?
it's for 1.3b only
hello. Now Wan 1.3B have a propre section. Can you move your lora to Wan t2v 1.3B thx