
If you like what I do here, please consider supporting my work via Ko-fi <3
https://ko-fi.com/pixelworldai
Version 2 "Act of Grog" !!
Larger dataset and better captioning.
Act of Grog was finetuned on over 7500 images, all captioned with @braintacles new brainblip model that provides much more natural language in captions versus what you might get with WD1.4 tagging.
This addresses the first version's primary issue of every image coming out super artsy/abstract, now the captions have marked those in the dataset and you can get normal images, or even better abstract ones now.
But what really makes this so special is that with this model you can prompt more naturally than before, so instead of something like "1girl, black hair, blue eyes, red dress, city, people, sidewalk" you can prompt more like "a woman with black hair and blue eyes, wearing a red dress, standing on a crowded sidewalk".
With the brainblip captioning also comes content ratings, this model features two:
pg_rating - mostly SFW
x_rating - NSFW
These ratings won't always make or break your prompt, it all depends on the weights. You can obviously still get a nude character with PG rating if nudity outweighs the rating in your prompt, but it still is pretty neat and I'm excited to see more models using it, especially after using the "rating_safe/questionable/explicit" in Pony Diffusion.
All example pics started at a base resolution of 768x768, I would highly recommend looking into Kohya deepshrink to go even higher than that.
Does not require VAE
Hope you enjoy! -J
Make sure to check out the AIpub Discord and Mastodon!
描述:
-
Larger dataset
-
Natural language captions
Larger dataset
Natural language captions
训练词语:
名称: talesOfGrog_v20ActOfGrog.safetensors
大小 (KB): 2082643
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success