Installing the custom script SD-Chad

[ghost]

Hi, i just tried installing the aesthetics scorer mentioned in the custom scripts Wiki (https://github.com/grexzen/SD-Chad) however, simply putting the files in the custom scripts folder does not work. The scripts readme doesn't mention anything about how to get it to run. I would be glad if someone could explain it to me. :)

Kind regards

[grexzen]

Hi, thanks for letting me know. I just checked and yeah something in the latest update changed the way scripts are visible. Need to find out what changed.

In the meantime, you can use the code directly in the custom code script text area (use --allow-code while starting SD).

You can change how many images you want to generate by editing the "n" at the bottom of the code "run(p, n)"

It will generate the image and put those above the score limit (that you can also edit, but those are statistically significant after generating 600K images) in your images folder, and those below in your grids folder (you an also edit the folders in the code).

The p.seed function as-is adds 1 to your seed for each new gen. If you edit that to = -1, then it will use random seeds every time.

Hope that helps. I'll try fixing it next.

---

#create aesthetic checker

from pathlib import Path
import torch
import torch.nn as nn
import torch.nn.functional as F
import clip
import numpy as np
from PIL import Image
import math
import os
import sys
import traceback
from modules.processing import Processed, process_images, StableDiffusionProcessing, create_infotext
import modules.images as images

state_name = "sac+logos+ava1-l14-linearMSE.pth"
if not Path(state_name).exists():
    url = f"https://github.com/christophschuhmann/improved-aesthetic-predictor/blob/main/{state_name}?raw=true"
    import requests
    r = requests.get(url)
    with open(state_name, "wb") as f:
        f.write(r.content)

class AestheticPredictor(nn.Module):
    def __init__(self, input_size):
        super().__init__()
        self.input_size = input_size
        self.layers = nn.Sequential(
            nn.Linear(self.input_size, 1024),
            nn.Dropout(0.2),
            nn.Linear(1024, 128),
            nn.Dropout(0.2),
            nn.Linear(128, 64),
            nn.Dropout(0.1),
            nn.Linear(64, 16),
            nn.Linear(16, 1)
        )

    def forward(self, x):
        return self.layers(x)

device = "cuda" if torch.cuda.is_available() else "cpu"
pt_state = torch.load(state_name, map_location=torch.device('cpu')) 

# CLIP embedding dim is 768 for CLIP ViT L 14
predictor = AestheticPredictor(768)
predictor.load_state_dict(pt_state)
predictor.to(device)
predictor.eval()

clip_model, clip_preprocess = clip.load("ViT-L/14", device=device)

def get_image_features(image, device=device, model=clip_model, preprocess=clip_preprocess):
    image = preprocess(image).unsqueeze(0).to(device)
    with torch.no_grad():
        image_features = model.encode_image(image)
        # l2 normalize
        image_features /= image_features.norm(dim=-1, keepdim=True)
    image_features = image_features.cpu().detach().numpy()
    return image_features

def get_score(image):
    image_features = get_image_features(image)
    score = predictor(torch.from_numpy(image_features).to(device).float())
    return score.item()

#generate scored images

def run(p, n):
      for x in range(int(n)):
            p.seed += 1
            print(p.seed)
            proc = process_images(p)
            gens = proc.images
            chad_score = round(get_score(gens[0]),1)
            print(chad_score)
            if chad_score >= 7.5:
                save_chad = images.save_image(gens[0], p.outpath_samples, "", p.seed, "-Top--" + str(chad_score), opts.samples_format)
                print("Top")
            elif 7 <= chad_score < 7.5:
                save_soy = images.save_image(gens[0], p.outpath_grids, "", p.seed, "-Chad--" + str(chad_score), opts.samples_format)
                print("Chad")
            else:
                print("Trash")
      return Processed(p, gens, p.seed, "")

run(p, 10)

[ghost]

I'll give it a try, thank you. :)

[pomee4]

Thanks! It works for me. But I have a question... How do I change output folder so it puts the images above limit to text2img_images/Good and below to text2img_images/Bad ?

[some9000]

I will just bump this a little to, hopefully, motivate @grexzen to look into making it into an extension, possibly (turns out it's not that terrible). It just seems too neat a function to just slowly fade.

[ClashSAN]

@tsngo has this as an extension https://github.com/tsngo/stable-diffusion-webui-aesthetic-image-scorer