π΄Patreon | Stable difussion and Content Creation | PANyZHAL
ββββββββββββ β¦ β β¦ ββββββββββββ
MY HONEST RECOMMENDATION
Clip skip 2 (also works in 1)
You can use this negatives:
lowres photo,blurry,film grain,costume,gloves,tattoo,big hand.weights 1 or 0.9 were good in my tests.
For avoid the color skin you can put in negatives one of these according to the base model: black man, black boy, black male
Get fun and if you have some comments let me know and if you like let me a review i will keep posting models :3
Description
FAQ
Comments (9)
Genuinely curious about how you even go about training something like this as a concept - ie, how to tag and what to tag/input as training images. For a meme.. this works extremely well haha
thanks you for your comment :3, for expressions or faces, i try to use just 4 or 5 images,(in this case 2 realistic and 2 draws) the simplest one for avoid contamination,and don't tag the expression, the person tag put it in a uncommon phrase eg: instead tag 1boy,boy. i write dark skin boy. also tag clothes, hair, background, and style(photo or drawing) for helping the flexibility.
And for this i think i trained with 50epochs, 1rep. dadaptadam. with this argument: decouple=True weight_decay=0.6 betas=0.9,0.99 (you can watch the complete config train in the file :D).
And do tests in the loras around 35,50 epoch(normally 30,35 are the best), the point were it can merged good with other loras,models and poses, because higher contaminate a lot. I hope this helps and sorry for my english
lol youtuber thumbnails
Error occurred when executing CheckpointLoaderSimple: 'model.diffusion_model.input_blocks.0.0.weight' (help pls)
how did you solved this, sorry for the delay and my english xd :c
@PANyZHALΒ I added a Lora loader and put the model inside, instead of checkpoint loader
train it on pony xl or illustrious
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.