app.py

from diffusers import KandinskyV22PriorPipeline, KandinskyV22Pipeline
import torch
import cv2
import numpy as np
from transformers import pipeline
import gradio as gr
from PIL import Image
from diffusers.utils import load_image
import os, random, gc, re, json, time, shutil, glob
import PIL.Image
import tqdm
from accelerate import Accelerator
from huggingface_hub import HfApi, InferenceClient, ModelCard, RepoCard, upload_folder, hf_hub_download, HfFileSystem
HfApi=HfApi()
HF_TOKEN=os.getenv("HF_TOKEN")
HF_HUB_DISABLE_TELEMETRY=1
DO_NOT_TRACK=1
HF_HUB_ENABLE_HF_TRANSFER=0
accelerator = Accelerator(cpu=True)
InferenceClient=InferenceClient()

apol=[]

pope_prior = accelerator.prepare(KandinskyV22PriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-2-prior", torch_dtype=torch.float32))
pope_prior.prior.to(memory_format=torch.channels_last)
pope_prior = pope_prior.to("cpu")
pope = accelerator.prepare(KandinskyV22Pipeline.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float32))
pope.unet.to(memory_format=torch.channels_last)
pope = pope.to("cpu")

def chdr(apol,prompt,modil,stips,fnamo,gaul):
    try:
        type="KNDSK22_INTERP"
        los=""
        tre='./tmpo/'+fnamo+'.json'
        tra='./tmpo/'+fnamo+'_0.png'
        trm='./tmpo/'+fnamo+'_half.png'
        flng=["yssup", "sllab", "stsaerb", "sinep", "selppin", "ssa", "tnuc", "mub", "kcoc", "kcid", "anigav", "dekan", "edun", "slatineg", "xes", "nrop", "stit", "ttub", "bojwolb", "noitartenep", "kcuf", "kcus", "kcil", "elttil", "gnuoy", "thgit", "lrig", "etitep", "dlihc", "yxes"]
        flng=[itm[::-1] for itm in flng]
        ptn = r"\b" + r"\b|\b".join(flng) + r"\b"
        if re.search(ptn, prompt, re.IGNORECASE):
            print("onon buddy")
        else:
            dobj={'img_name':fnamo,'model':modil,'lora':los,'prompt':prompt,'steps':stips,'type':type}
            with open(tre, 'w') as f:
                json.dump(dobj, f)
            HfApi.upload_folder(repo_id="JoPmt/hf_community_images",folder_path="./tmpo",repo_type="dataset",path_in_repo="./",token=HF_TOKEN)
        dobj={'img_name':fnamo,'model':modil,'lora':los,'prompt':prompt,'steps':stips,'type':type,'haed':gaul,}
        with open(tre, 'w') as f:
            json.dump(dobj, f)
        HfApi.upload_folder(repo_id="JoPmt/Tst_datast_imgs",folder_path="./tmpo",repo_type="dataset",path_in_repo="./",token=HF_TOKEN)
        try:
            for pgn in glob.glob('./tmpo/*.png'):
                os.remove(pgn)
            for jgn in glob.glob('./tmpo/*.json'):
                os.remove(jgn)
            del tre
            del tra
            del trm
        except:
            print("cant")
    except:
        print("failed to make obj")

def plax(gaul,req: gr.Request):
    gaul=str(req.headers)
    return gaul

def plex(cook, img, neg_prompt, stips, prior_stps, itr_stps, one, two, three, nut, wit, het, gaul, progress=gr.Progress(track_tqdm=True)):
    gc.collect()
    apol=[]
    modil="kandinsky-community/kandinsky-2-2-prior,kandinsky-community/kandinsky-2-2-decoder"
    goof = load_image(img).resize((wit, het))
    prompt = cook
    negative_prior_prompt = neg_prompt
    nm=0
    fnamo=""+str(int(time.time()))+""
    if nut == 0:
        nm = random.randint(1, 2147483616)
        while nm % 32 != 0:
            nm = random.randint(1, 2147483616)
    else:
        nm=nut
    generator = torch.Generator(device="cpu").manual_seed(nm)
    
    img_emb = pope_prior(prompt=prompt, guidance_scale=one, num_inference_steps=prior_stps, generator=generator)
    negative_emb = pope_prior(prompt=negative_prior_prompt, guidance_scale=1, num_inference_steps=prior_stps)
    imags = pope(image_embeds=img_emb.image_embeds,negative_image_embeds=negative_emb.image_embeds,num_inference_steps=stips,generator=generator,height=het,width=wit).images[0]
    images_texts = [cook, goof, imags]
    weights = [one, two, three]
    primpt = ""
    prior_out = pope_prior.interpolate(images_texts, weights, num_inference_steps=itr_stps)
    imas = pope(**prior_out, height=het, width=wit, num_inference_steps=stips)
    for i, imge in enumerate(imas["images"]):
        apol.append(imge)
        imge.save('./tmpo/'+fnamo+'_'+str(i)+'.png', 'PNG')
    imags.save('./tmpo/'+fnamo+'_half.png', 'PNG')
    apol.append(imags)
    chdr(apol,prompt,modil,stips,fnamo,gaul)
    return apol

def aip(ill,api_name="/run"):
    return
def pit(ill,api_name="/predict"):
    return

with gr.Blocks(theme=random.choice([gr.themes.Monochrome(),gr.themes.Base.from_hub("gradio/seafoam"),gr.themes.Base.from_hub("freddyaboulton/dracula_revamped"),gr.themes.Glass(),gr.themes.Base(),]),analytics_enabled=False) as iface:
    ##iface.description="Running on cpu, very slow! by JoPmt."
    out=gr.Gallery(label="Generated Output Image", columns=1)
    inut=gr.Textbox(label="Prompt")
    mput=gr.Image(label="drop", type="filepath")
    gaul=gr.Textbox(visible=False)
    btn=gr.Button("GENERATE")
    with gr.Accordion("Advanced Settings", open=False):
        inet=gr.Textbox(label="Negative_prompt", value="lowres,text,bad quality,low quality,jpeg artifacts,ugly,bad hands,bad face,blurry,bad eyes,watermark,signature")
        inyt=gr.Slider(label="Num inference steps",minimum=1,step=1,maximum=30,value=10)
        ihop=gr.Slider(label="Num prior inference steps",minimum=1,step=1,maximum=10,value=5)
        ihip=gr.Slider(label="Num prior interpolation steps",minimum=1,step=1,maximum=10,value=5)
        inat=gr.Slider(label="Text Guide",minimum=0.01,step=0.01,maximum=0.99,value=0.5)
        csal=gr.Slider(label="Your Image Guide",minimum=0.01,step=0.01,maximum=0.99,value=0.5)
        csbl=gr.Slider(label="Generated Image Guide",minimum=0.01,step=0.01,maximum=0.99,value=0.3)
        indt=gr.Slider(label="Manual seed (leave 0 for random)",minimum=0,step=32,maximum=2147483616,value=0)
        inwt=gr.Slider(label="Width",minimum=256,step=32,maximum=1024,value=768)
        inht=gr.Slider(label="Height",minimum=256,step=32,maximum=1024,value=768)
    
    btn.click(fn=plax,inputs=gaul,outputs=gaul).then(fn=plex, outputs=[out], inputs=[inut,mput,inet,inyt,ihop,ihip,inat,csal,csbl,indt,inwt,inht,gaul])

iface.queue(max_size=1,api_open=False)
iface.launch(max_threads=20,inline=False,show_api=False)