import json import requests import io import base64 import uuid import sys, os from PIL import Image, PngImagePlugin from model_lists import * import time import random def call_txt2img(passingprompt,size,upscale,debugmode,filename="",model = "currently selected model",samplingsteps = "40",cfg= "7",hiressteps ="0",denoisestrength="0.6",samplingmethod="DPM++ SDE Karras", upscaler="R-ESRGAN 4x+",hiresscale="2",apiurl="http://127.0.0.1:7860", qualitygate=False,quality="7.6",runs="5",negativeprompt="",qualityhiresfix = False, qualitymode = "highest", qualitykeep="keep used", basesize="512"): #set the prompt! prompt = passingprompt checkprompt = passingprompt.lower() #set the URL for the API url = apiurl #rest of prompt things sampler_index = samplingmethod steps = samplingsteps if(debugmode==1): steps="10" cfg_scale = cfg originalsize = size #size sizes = setsize(size, basesize,originalsize) width = sizes[0] height = sizes[1] #upscaler enable_hr = upscale if(debugmode==1 or qualityhiresfix == True): enable_hr="False" #defaults hr_scale = hiresscale denoising_strength = denoisestrength hr_second_pass_steps = hiressteps #hr_upscaler = "LDSR" # We have the time, why not use LDSR if(upscaler != "automatic"): hr_upscaler = upscaler else: upscalerlist = get_upscalers() # on automatic, make some choices about what upscaler to use # photos, prefer 4x ultrasharp # anime, cartoon or drawing, go for R-ESRGAN 4x+ Anime6B # else, R-ESRGAN 4x+" if("hoto" in checkprompt and "4x-UltraSharp" in upscalerlist): hr_upscaler = "4x-UltraSharp" elif("anime" in checkprompt or "cartoon" in checkprompt or "draw" in checkprompt or "vector" in checkprompt or "cel shad" in checkprompt or "visual novel" in checkprompt): hr_upscaler = "R-ESRGAN 4x+ Anime6B" else: hr_upscaler = "R-ESRGAN 4x+" if(hiressteps==0): hiressteps = samplingsteps hr_second_pass_steps = int(hiressteps/2) hr_scale = 2 if(hr_upscaler== "4x-UltraSharp"): denoising_strength = "0.35" if(hr_upscaler== "R-ESRGAN 4x+ Anime6B+"): denoising_strength = "0.6" # 0.6 is fine for the anime upscaler if(hr_upscaler== "R-ESRGAN 4x+"): denoising_strength = "0.5" # default 0.6 is a lot and changes a lot of details #params to stay the same script_dir = os.path.dirname(os.path.abspath(__file__)) # Script directory outputTXT2IMGfolder = os.path.join(script_dir, "./automated_outputs/txt2img/") outputTXT2IMGfolder.replace("./", "/") if(filename==""): filename = str(uuid.uuid4()) outputTXT2IMGpng = '.png' #outputTXT2IMGFull = '{}{}{}'.format(outputTXT2IMGfolder,filename,outputTXT2IMGpng) outputTXT2IMGtxtfolder = os.path.join(script_dir, "./automated_outputs/prompts/") outputTXT2IMGtxtfolder.replace("./", "/") outputTXT2IMGtxt = '.txt' outputTXT2IMGtxtFull = '{}{}{}'.format(outputTXT2IMGtxtfolder,filename,outputTXT2IMGtxt) # params for quality gate isGoodNumber = float(quality) foundgood = False MaxRuns = int(runs) Runs = 0 scorelist = [] scoredeclist = [] imagelist = [] pnginfolist = [] seedlist = [] widthlist = [] heightlist = [] usedwidht = width usedheight = height usedseed = -1 imagethatiskept = "" # flow things continuewithnextpart = True # starting seed of -1 seed = -1 #call TXT2IMG payload = { "prompt": prompt, "sampler_index": sampler_index, "steps": steps, "cfg_scale": cfg_scale, "width": width, "height": height, "enable_hr": enable_hr, "denoising_strength": denoising_strength, "hr_scale": hr_scale, "hr_upscaler": hr_upscaler, "hr_second_pass_steps": hr_second_pass_steps, "seed": seed, "hr_prompt": prompt } if(model != "currently selected model"): payload.update({"sd_model": model}) if(negativeprompt != ""): payload.update({"negative_prompt": negativeprompt}) payload.update({"hr_negative_prompt": negativeprompt}) while Runs < MaxRuns: # make the filename unique for each run _0, _1, etc. addrun = "_" + str(Runs) filenamefull = filename + addrun outputTXT2IMGFull = '{}{}{}'.format(outputTXT2IMGfolder,filenamefull,outputTXT2IMGpng) r = [] # randomize the seed ( A number between 0 and 4,294,967,295 ) seed = random.randrange(1, 4294967295) payload["seed"] = seed # do we need to randomize the size? if(originalsize=='all' or originalsize == 'wild'): sizes = setsize(size, basesize, originalsize) width = sizes[0] height = sizes[1] payload["width"] = width payload["height"] = height # If we don't get an image back, we want to retry a few times. Max 3 times for i in range(4): response = requests.post(url=f'{url}/sdapi/v1/txt2img', json=payload) r = response.json() if('images' in r): break # this means if we have the images object, then we "break" out of the for loop. else: if(i == 3): print("If this keeps happening: Is WebUI started with --api enabled?") print("") raise ValueError("API has not been responding after several retries. Stopped processing.") print("") print("We haven't received an image from the API. Maybe something went wrong. Will retry after waiting a bit.") time.sleep(10 * (i+1) ) # incremental waiting time for i in r['images']: image = Image.open(io.BytesIO(base64.b64decode(i.split(",",1)[0]))) png_payload = { "image": "data:image/png;base64," + i } response2 = requests.post(url=f'{url}/sdapi/v1/png-info', json=png_payload) pnginfo = PngImagePlugin.PngInfo() pnginfo.add_text("parameters", response2.json().get("info")) image.save(outputTXT2IMGFull, pnginfo=pnginfo) if(qualitygate==True): # check if the file exists in the parent directory imagescorer_path = os.path.abspath(os.path.join(os.path.dirname(__file__), '..', 'stable-diffusion-webui-aesthetic-image-scorer', 'scripts')) #print(imagescorer_path) if imagescorer_path not in sys.path: sys.path.append(imagescorer_path) try: import image_scorer print("Found aesthetic-image-scorer! Using this to measure the results...") score = image_scorer.get_score(image) scoredeclist.append(score) score = round(score,1) seedlist.append(seed) widthlist.append(width) heightlist.append(height) scorelist.append(score) imagelist.append(outputTXT2IMGFull) pnginfolist.append(pnginfo) print("This image has scored: "+ str(score) + " out of " + str(isGoodNumber)) if(score >= isGoodNumber or debugmode == 1): foundgood = True print("Yay its good! Keeping this result.") else: runstodo = MaxRuns - Runs - 1 print("Not a good result. Retrying for another " + str(runstodo) + " times or until the image is good enough.") except ImportError: foundgood = True # just continue :) # handle the case where the module doesn't exist print("Could not find the stable-diffusion-webui-aesthetic-image-scorer extension.") print("Install this extension via the WebUI to use Quality Gate") pass else: foundgood = True # If there is no quality gate, then everything is good. So we escape this loop Runs += 1 if(foundgood == True): break #Break the loop if we found something good. Or if we set it to good :) if(len(imagelist) > 0): if(foundgood == True): if(qualitykeep == "keep used"): print("Removing any other images generated this run (if any).") else: if(qualitymode == "highest"): print("") print("Stopped trying, keeping the best image we had so far.") print("") else: print("") print("Eh, its all pretty bad. Not going forward with any image.") print("") # Get the index of the first occurrence of the maximum value in the list #if(qualitymode == "highest" or (qualitymode != "highest" and foundgood == True)): indexofimagetokeep = scoredeclist.index(max(scoredeclist)) outputTXT2IMGFull = imagelist[indexofimagetokeep] #store the image to keep in here, so we can pass it along pnginfo = pnginfolist[indexofimagetokeep] usedseed = seedlist[indexofimagetokeep] usedwidht = widthlist[indexofimagetokeep] usedheight = heightlist[indexofimagetokeep] imagethatiskept = imagelist[indexofimagetokeep] imagelist.pop(indexofimagetokeep) #remove all other images if(qualitykeep == "keep used"): for imagelocation in imagelist: os.remove(imagelocation) if(foundgood == False and qualitymode != "highest"): continuewithnextpart = False if(imagethatiskept != "" and qualitykeep == "keep used"): os.remove(imagethatiskept) # We have done everything, but if we want to run Hires fix from the quality gate, we are going to have to do it again. But this time a little easier. # We do have the check wether we want to run hiresfix first if(qualityhiresfix == True and upscale == False and continuewithnextpart == True): print("Quality Gate hires fix was enabled, but no hires fix settings were given.") if(qualityhiresfix == True and upscale == True and continuewithnextpart == True): print("Going to run the chosen image with hiresfix") payload["seed"] = usedseed payload["width"] = usedwidht payload["height"] = usedheight payload["enable_hr"] = "True" # make the filename unique for hiresfix addrun = "_hiresfix" filenamefull = filename + addrun outputTXT2IMGFull = '{}{}{}'.format(outputTXT2IMGfolder,filenamefull,outputTXT2IMGpng) # If we don't get an image back, we want to retry a few times. Max 3 times for i in range(4): response = requests.post(url=f'{url}/sdapi/v1/txt2img', json=payload) r = response.json() if('images' in r): break # this means if we have the images object, then we "break" out of the for loop. else: if(i == 3): print("If this keeps happening: Is WebUI started with --api enabled?") print("") raise ValueError("API has not been responding after several retries. Stopped processing.") print("") print("We haven't received an image from the API. Maybe something went wrong. Will retry after waiting a bit.") time.sleep(10 * (i+1) ) # incremental waiting time for i in r['images']: image = Image.open(io.BytesIO(base64.b64decode(i.split(",",1)[0]))) png_payload = { "image": "data:image/png;base64," + i } response2 = requests.post(url=f'{url}/sdapi/v1/png-info', json=png_payload) pnginfo = PngImagePlugin.PngInfo() pnginfo.add_text("parameters", response2.json().get("info")) image.save(outputTXT2IMGFull, pnginfo=pnginfo) with open(outputTXT2IMGtxtFull,'w',encoding="utf8") as txt: json_object = json.dumps(payload, indent = 4) txt.write(json_object) return [outputTXT2IMGFull,pnginfo,continuewithnextpart] def setsize(ratio,basesize, originalsize): # prompt + size if(originalsize == "all"): sizelist = ["portrait", "wide", "square"] ratio = random.choice(sizelist) # from base ratio if(ratio=='wide' and basesize != "1024"): width = str(int(basesize) + 256) height = basesize elif(ratio=='wide' and basesize == "1024"): width = "1152" height = "896" elif(ratio=='portrait' and basesize != "1024"): width = basesize height = str(int(basesize) + 256) elif(ratio=='portrait' and basesize == "1024"): width = "896" height = "1152" elif(ratio=='ultrawide'): width = "1280" height = "360" elif(ratio=='ultraheight'): width = "360" height = "1280" elif(ratio=='wild'): width = str(round((random.randint(0,4) * 128) + (int(basesize) /2) ) ) # random value of 0 to 512 in steps of 128 + half of base size height = str(round( (random.randint(0,4) * 128) + (int(basesize) /2) ) ) # random value of 0 to 512 in steps of 128 + half of base size else: width = basesize height = basesize return [width, height]