
Warning: Highly experimental, does not work for a lot of things, often it's not the 1st person holding the item but the person in the picture, garbled hands, needs multiple tries etc (but it's my best attempt after many attempts). The Pony LoRA is considerably less capable than the Animagine LoRA.
Trigger: povhldg
<item>
Helpful tag: pov hands
(for Animagine), pov holding
or pov hand
(for Pony)
Notes on Pony version
The Pony LoRA works somewhat differently:
-
The
povhldg
token has a weaker effect -
Using
pov hands
may lead to multiple hands appearing, often focuses on hand holding -
pov holding
seems to be a better supporting tag in Pony, but can lead to you using both hands -
Update: Just found out that
pov hand
is better thanpov hands
for Pony -
Sometimes the person is also holding the same item,
holding
in negatives may help -
Unlike the Animagine LoRA, resizing it did not lead to improvements, so I left it at the original dim (8/4). But it also doesn't seem to hurt it (too much) so feel free to resize yourself
The Pony LoRA was trained with a higher lr and more steps.
描述:
训练词语: povhldg,pov hands
名称: concept_povholding_res.safetensors
大小 (KB): 28218
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success