John6666

/

pony-photo-real-merge-x3-v2sharper-sdxl

like 0

That's terrible. I wonder how far the regulations have gone...😓
I haven't seen it yet, so I don't have it.
I don't think it showed up on Civitai, probably because I'm an unpaid user.
Maybe it was an early access slot.

BTW, regarding v10, it looks like someone has a backup on HF, but it doesn't seem to be recognized by safetensors.
Another file in the same repository looks fine, so maybe this version is the only file that is slightly corrupted.
It's not out of the realm of possibility that something weird is mixed in...
https://huggingface.co/alexgenovese/reica_models/blob/main/fullyREALXL_v10Perfect10n.safetensors

BTW, regarding v10, it looks like someone has a backup on HF, but it doesn't seem to be recognized by safetensors.
Another file in the same repository looks fine, so maybe this version is the only file that is slightly corrupted.
It's not out of the realm of possibility that something weird is mixed in...
https://huggingface.co/alexgenovese/reica_models/blob/main/fullyREALXL_v10Perfect10n.safetensors

Its a header problem, i reverted it to ckpt as this usually fix heades but failed too, everything else is fine, maybe reverting safetensors to diffusers to manually fix

Error while deserializing header: HeaderTooLarge

Error: Failed to parse file fullyREALXL_v10Perfect10n.safetensors: safetensor header is too big. Maximum supported size is 25000000 bytes.

That's terrible. I wonder how far the regulations have gone...😓
I haven't seen it yet, so I don't have it.
I don't think it showed up on Civitai, probably because I'm an unpaid user.
Maybe it was an early access slot.

Users like Ai_photos, DiceAiDevelopment, and pAInCREAT0R had some of the most realistic models on the site. Sadly, they didn't back up most of their files on Hugging Face. Seamless realism can be controversial, but these models helped to shape and improve some of the new models that don't focus on realism. it's a terrible loss ...

The v3 of Ai_photos will indeed be difficult because the release period is too short, but there is no lack of hope that other models can be salvaged by user volunteers. If it is a famous model, it is more likely to be on a closed pay site or pirated, for example.
That said, I did succeed in loading one. I don't know if this model file works or not, or if this guy is real or not.

import torch
from safetensors.torch import load_file, save_file
weights = torch.load("fullyREALXL_v10Perfect10n.safetensors", weights_only=True)
state_dict = weights['state_dict']
save_file(state_dict, "fullyREALXL_v10Perfect10n_.safetensors")

Sorry. This way, it will be safetensors, but the internal structure seems to be broken.
The result of the processing is clearly that the number of layers is not enough.