modules/search_pipeline.py

import modules.async_worker as worker

import shared
import glob
from pathlib import Path
import datetime
import re
import json

from PIL import Image

# Copy this file, add suitable code and add logic to modules/pipelines.py to select it

def search_for_words(searchfor, prompt):
    searchfor = searchfor.lower()
    prompt = prompt.lower()
    words = searchfor.split(",")
    result = True
    for word in words:
        word = word.strip()
        if word not in prompt:
            result = False
            break
    return result

def search(search_string, maxresults=10, callback=None):
    images = []
    skip = 0

    folder=shared.path_manager.model_paths["temp_outputs_path"]
    current_time = datetime.datetime.now()
    daystr = current_time.strftime("%Y-%m-%d")

    # Parse search arguments
    searchfor = re.sub(r"search: *", "", search_string, count=1, flags=re.IGNORECASE)

    chomp = True # Do this until we can't chomp off any more options
    while chomp:
        chomp = False

        # Date
        matchstr = r"^[0-9]{4}-[0-9]{2}-[0-9]{2}\s?"
        match = re.match(matchstr, searchfor, re.IGNORECASE)
        if match is not None:
            daystr = match.group().strip()
            searchfor = re.sub(matchstr, "", searchfor)
            chomp = True

        # All Dates
        matchstr = r"^all:\s?"
        match = re.match(matchstr, searchfor, re.IGNORECASE)
        if match is not None:
            daystr = "*"
            searchfor = re.sub(matchstr, "", searchfor)
            chomp = True

        # Skip
        matchstr = r"^skip:\s?(?P<skip>[0-9]+)\s?"
        match = re.match(matchstr, searchfor, re.IGNORECASE)
        if match is not None:
            skip = int(match.group("skip"))
            searchfor = re.sub(matchstr, "", searchfor)
            chomp = True

        # Skip more
        matchstr = r"^\+(?P<skip>[0-9]+)\s?"
        match = re.match(matchstr, searchfor, re.IGNORECASE)
        if match is not None:
            skip += int(match.group("skip"))
            searchfor = re.sub(matchstr, "", searchfor)
            chomp = True

        # Set max
        matchstr = r"^max:\s?(?P<max>[0-9]+)\s?"
        match = re.match(matchstr, searchfor, re.IGNORECASE)
        if match is not None:
            maxresults = int(match.group("max"))
            searchfor = re.sub(matchstr, "", searchfor)
            chomp = True

    searchfor = searchfor.strip()

    # For all folder/daystr/*.(png|gif) ... match metadata
    globs = ["*.png", "*.gif"]
    pngs = set()
    for g in globs:
        for f in glob.glob(str(Path(folder) / daystr / g)):
            pngs.add(f)
    pngs = sorted(pngs)

    found = 0
    for file in pngs:
        im = Image.open(file)
        metadata = {"prompt": ""}
        if im.info.get("parameters"):
            metadata = json.loads(im.info["parameters"])

            # Show image if metadata is missing. (like for gifs)
        if searchfor == "" or "Prompt" not in metadata or search_for_words(searchfor, metadata["Prompt"]):
            # Return finished image to preview
            if callback is not None and found > skip:
                callback(found - skip, 0, 0, maxresults, None) # Returning im here is a bit much...
            images.append(file)
            found += 1
        if found >= (maxresults + skip):
            break

    return images[skip:]


class pipeline:
    pipeline_type = ["search"]

    model_hash = ""

    def parse_gen_data(self, gen_data):
        gen_data["original_image_number"] = gen_data["image_number"]
        gen_data["image_number"] = 1
        gen_data["show_preview"] = False
        return gen_data

    def load_base_model(self, name):
        # We're not doing models here
        return

    def load_keywords(self, lora):
        filename = lora.replace(".safetensors", ".txt")
        try:
            with open(filename, "r") as file:
                data = file.read()
            return data
        except FileNotFoundError:
            return " "

    def load_loras(self, loras):
        return

    def refresh_controlnet(self, name=None):
        return

    def clean_prompt_cond_caches(self):
        return


    def process(
        self,
        gen_data=None,
        callback=None,
    ):
        worker.add_result(
            gen_data["task_id"],
            "preview",
            (-1, f"Searching ...", None)
        )
        maxresults = gen_data["original_image_number"]
        maxresults = 100 if maxresults <= 1 else maxresults # 0 and 1 is 100 matches (max)

        images = search(gen_data["positive_prompt"], maxresults=maxresults, callback=callback)

        return images