Aeontropy on Nostr: My theory in why hands gets wrong besides using a high number of steps. is that the ...
My theory in why hands gets wrong besides using a high number of steps. is that the database doesn't follow any particular order. let's say hands needs to be trained in batches sets 🖐️👌👍 first set with only extended palm, next one with ok sign, ect... but for that database needs only images with only one hand expression per set, if training contains other form will mess the number and position. I don't know if the lora perfect hands for sdxl follows that training idea 🤔
Published at
2024-09-15 21:59:03Event JSON
{
"id": "1ed22e175bcc540cd0d8bb9b18cbfe24bc71706bedd85d0125aeb9ffe2bd2f4f",
"pubkey": "b75b9a3131f4263add94ba20beb352a11032684f2dac07a7e1af827c6f3c1505",
"created_at": 1726437543,
"kind": 1,
"tags": [],
"content": "My theory in why hands gets wrong besides using a high number of steps. is that the database doesn't follow any particular order. let's say hands needs to be trained in batches sets 🖐️👌👍 first set with only extended palm, next one with ok sign, ect... but for that database needs only images with only one hand expression per set, if training contains other form will mess the number and position. I don't know if the lora perfect hands for sdxl follows that training idea 🤔",
"sig": "2a47439035e1a67d9e9b744283f660859b4bd5abae3ba11663e32c6ad1ec3dda10f6f85c85c9c6cab041c4621a25803b9dcae29d20a2ea24d361ee5d87585f60"
}
