from asyncio import constants import gradio as gr import requests import os import re import random from words import * from base64 import b64decode from PIL import Image import io import numpy as np # GPT-J-6B API API_URL = "https://api-inference.huggingface.co/models/EleutherAI/gpt-j-6B" #HF_TOKEN = os.environ["HF_TOKEN"] #headers = {"Authorization": f"Bearer {HF_TOKEN}"} prompt = """ Bilbo is a hobbit rogue who wears a brown cloak and carries a ring. Bremen is a human wizard, he wears a blue robe and carries a wand. """ examples = [["river"], ["night"], ["trees"],["table"],["laughs"]] def npc_randomize(): #name is a random combination of syllables name ="" for i in range(random.randint(2,4)): name += random.choice(constants) name += random.choice(vowels) if random.random()<0.5: name += random.choice(constants) if random.random()<0.1: name += random.choice(seperators) #capitalize first letter name = name[0].upper() + name[1:] race=random.choice(races) characterClass=random.choice(classes) pronoun=random.choices(["he","she","they"],weights=[0.45,0.45,0.1],k=1)[0] return name,race,characterClass,pronoun def genericDescription(): desc=" wears a {color} {outfit}".format(color=random.choice(colors),outfit=random.choice(outfits)) if random.random()<0.5: desc+=" and a {color} {outfit}".format(color=random.choice(colors),outfit=random.choice(outfits)) if random.random()<0.5: desc+=" and carries a {weapon}".format(weapon=random.choice(weapons)) elif random.random()<0.5: desc+=" and carries a {weapon} and a {object}".format(weapon=random.choice(weapons),object=random.choice(objects)) else: desc+=" and carries two {weapon}s".format(weapon=random.choice(weapons)) return desc def npc_generate(name,race,characterClass,pronoun): desc="{name} is a {race} {characterClass}, {pronoun}".format(name=name,race=race,characterClass=characterClass,pronoun=pronoun) p = prompt + "\n"+desc print(f"*****Inside desc_generate - Prompt is :{p}") json_ = {"inputs": p, "parameters": { "top_p": 0.9, "temperature": 1.1, "max_new_tokens": 50, "return_full_text": False, }} #response = requests.post(API_URL, headers=headers, json=json_) response = requests.post(API_URL, json=json_) output = response.json() print(f"If there was an error? Reason is : {output}") #error handling if "error" in output: print("using fallback description method!") #fallback method longDescription=genericDescription() else: output_tmp = output[0]['generated_text'] print(f"GPTJ response without splits is: {output_tmp}") if "\n\n" not in output_tmp: if output_tmp.find('.') != -1: idx = output_tmp.find('.') longDescription = output_tmp[:idx+1] else: idx = output_tmp.rfind('\n') longDescription = output_tmp[:idx] else: longDescription = output_tmp.split("\n\n")[0] # +"." longDescription = longDescription.replace('?','') print(f"longDescription being returned is: {longDescription}") return desc+longDescription def desc_to_image(desc): print("*****Inside desc_to_image") desc = " ".join(desc.split('\n')) desc = desc + ", character art, concept art, artstation" steps, width, height, images, diversity = '50','256','256','1',15 iface = gr.Interface.load("spaces/multimodalart/latentdiffusion") print("about to die",iface,dir(iface)) prompt = re.sub(r'[^a-zA-Z0-9 ,.]', '', desc) print("about to die",prompt) img=iface(desc, steps, width, height, images, diversity)[0] return img def desc_to_image_dalle(desc): print("*****Inside desc_to_image") desc = " ".join(desc.split('\n')) desc = desc + ", character art, concept art, artstation" steps, width, height, images, diversity = '50','256','256','1',15 #iface = gr.Interface.load("huggingface/flax-community/dalle-mini")#this isn't a real interface iface = gr.Interface.load("spaces/multimodalart/rudalle") print("about to die",iface,dir(iface)) prompt = re.sub(r'[^a-zA-Z0-9 ,.]', '', desc) print("about to die",prompt) model='Realism' aspect_ratio = 'Square' #img=iface(desc,model,aspect_ratio)[0] result=iface(desc,"Square","Realism") print(f"result is: {result}") return result[0] def desc_to_image_cf(desc): cf = gr.Interface.load("spaces/Gradio-Blocks/clip-guided-faces") print("about to die",cf) text=desc init_image=None skip_timesteps=0 clip_guidance_scale=600 tv_scale=0 range_scale=0 init_scale=0 seed=0 image_prompts=None timestep_respacing= 25 cutn=16 im_prompt_weight =1 result = cf.fns[0].fn(text, init_image, skip_timesteps, clip_guidance_scale, tv_scale, range_scale, init_scale, seed, image_prompts,timestep_respacing, cutn, im_prompt_weight) #convert result from dataurl to image img=result[0] header, encoded = img.split(",", 1) data = b64decode(encoded) image = Image.open(io.BytesIO(data)) image_np = np.array(image) return image_np demo = gr.Blocks() with demo: gr.Markdown("

NPC Generator

") gr.Markdown( "based on Gradio poetry generator." "
first input name, race and class (or generate them randomly)
" "
Next, use GPT-J to generate a short description
" "
Finally, Generate an illustration 🎨 provided by Latent Diffusion model.
" #"
Or using Rudalle model.
" "
Or using clip-guides faces.
" ) with gr.Row(): b0 = gr.Button("Randomize name,race and class") b1 = gr.Button("Generate NPC Description") b2 = gr.Button("Generate Portrait (latent diffusion)") b3 = gr.Button("Generate Portrait (clip-faces)") with gr.Row(): input_name = gr.Textbox(label="name",placeholder="Drizzt") input_race = gr.Textbox(label="race",placeholder="dark elf") input_class = gr.Textbox(label="class",placeholder="ranger") input_pronoun = gr.Textbox(label="pronoun",placeholder="he") with gr.Row(): desc_txt = gr.Textbox(label="description",lines=7) output_image = gr.Image(label="portrait",type="filepath", shape=(256,256)) b0.click(npc_randomize,inputs=[],outputs=[input_name,input_race,input_class,input_pronoun]) b1.click(npc_generate, inputs=[ input_name,input_race,input_class,input_pronoun], outputs=desc_txt) b2.click(desc_to_image, desc_txt, output_image) b3.click(desc_to_image_cf, desc_txt, output_image) #examples=examples demo.launch(enable_queue=True, debug=True)