Furkan Gözükara
patreon
2 Pre-Processing Scripts And 3 Datasets Of Processed Class Images For Popular Models
🕑 Added 2023-06-09 01:07:46 +0000 UTCJoin discord and tell me your discord username to get a special rank : SECourses Discord
13 August 2023 Update : fast_post_process.py added to the attachments. The difference of this file is that, it downscales images to 1000px first then do processing for faster processing. Your saved images will be still original resolution. It will be only used to speed up image detection and processing.
Video Tutorial for this post : https://youtu.be/olX1mySE8HA
Batch preprocess images. Removes multiple-face, black&white, NSWF. Datasets (4k+ images) for ReVAnimated, Deliberate V2, DreamShaper V6.
The shared datasets are 512x768 and 768x1024 pixels photo of man for ReVAnimated, Deliberate V2, DreamShaper V6. Check below to see exactly how many images and how much space they require. Each one has more than 4,000 unique images.
The password for files (if asked) are simply : secourses
Used VAE : vae-ft-mse-840000-ema-pruned.ckpt
Used prompt is : photo of man - cfg 7 - no negatives - eular a
So the class is photo of man
These datasets are made for hopefully my upcoming styling DreamBooth training tutorials
The used models safetensors direct download links from Hugging Face
revAnimated_v122.safetensors (full FP - 5.1 GB)
DreamShaper_6_NoVae.safetensors (full FP - 5.6 GB)
Deliberate_v2.safetensors (FP16 - no full FP available - 2 GB)
512x768 Pixel class images (raw output and cleaned)
4238 imgs (512x768) deliberate_v2_post_processed_clean.zip (2.22 GB)
4263 imgs (512x768) DreamShaper_6_NoVae_post_processed_clean.zip (2 GB)
6055 imgs (512x768) revAnimated_v122_post_processed_clean (2.96 GB)
768x1024 Pixel class images (raw output and cleaned)
5954 imgs (768x1024) deliberate_v2_post_processed_clean.zip (6.58 GB)
6281 imgs (768x1024) revAnimated_v122_post_processed_clean.zip (6.65 GB)
4975 imgs (768x1024) DreamShaper_6_NoVae_post_processed_clean.rar (5.16 GB)
768x1024 Pixel class images (hd fix and cleaned - upcoming) (better than raw)
5443_imgs_768x1024_revAnimated_v122_high_res_fix_clean.zip
6181_imgs_768x1024_deliberate_v2_high_res_fix_clean.zip
5692_imgs_768x1024_DreamShaper_6_NoVae_high_res_fix_clean.zip
Shared Scripts
NSFW remover standalone APP (exe with DLLs does not require visual studio - 600 MB) : https://huggingface.co/MonsterMMORPG/SECourses/resolve/main/nsfw%20remover%20app.zip
Watch video to learn scripts.
Download attached scripts
To use these files unrunpod
You need to install 7zip
apt-get install p7zip-full
Download with
wget
e.g. :
Then use this command to extract them
7z x "6055 imgs (512x768) revAnimated_v122_post_processed_clean.zip"
It will ask you password
type secourses
typing will be invisible
type again and it works
Comments
Furkan Gözükara
i think if you can get 100k that can make huge difference. I am producing more images right now. will update this post. I am also planning a tutorial video for 1m images fine tuning hopefully.
diffusers
So these are the datasets used to train models like RevAnimated? How many images does one need to train a model effectively?
Furkan Gözükara
this is dataset. not something different than model. i generated them and post processed them. you can also do this on every model. so this is prepared for you if you use :)
Kevin momplaisir
how are this data set diferent then the civitai model how are this enhanced?
