Update app.py

app.py

@@ -1,110 +1,121 @@
 import gradio as gr
+import torch
 import numpy as np
 import modin.pandas as pd
-import torch
-import os
-from glob import glob
-from pathlib import Path
-from typing import Optional
-from diffusers import LTXImageToVideoPipeline
-from diffusers.utils import load_image, export_to_video
 from PIL import Image
-
-import uuid
-import random
+from diffusers import DiffusionPipeline #, StableDiffusion3Pipeline
 from huggingface_hub import hf_hub_download
 
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
 torch.cuda.max_memory_allocated(device=device)
 torch.cuda.empty_cache()
-pipe = LTXImageToVideoPipeline.from_pretrained("Lightricks/LTX-Video", torch_dtype=torch.bfloat16)
-pipe.to(device)
-torch.cuda.empty_cache()
-max_64_bit_int = 2**63 - 1
 
-def sample(
-    image: Image, prompt, negative_prompt,
-    seed: Optional[int] = 42,
-    randomize_seed: bool = True,
-    num_inference_steps: int = 25,
-    num_frames: int = 60,
-    fps_id: int = 12,
-    decode_noise_scale: float = 0.3,
-    max_sequence_length: int = 512,
-    decoding_t: int = 3,  # Number of frames decoded at a time! This eats most VRAM. Reduce if necessary.
-    device: str = "cuda",
-    output_folder: str = "outputs",
-):
-    if image.mode == "RGBA":
-        image = image.convert("RGB")
-        
-    if(randomize_seed):
-        seed = random.randint(0, max_64_bit_int)
-    torch.cuda.max_memory_allocated(device=device)
-    torch.cuda.empty_cache()
-    generator = torch.manual_seed(seed)
+def genie (Model, Prompt, negative_prompt, height, width, scale, steps, seed, refine, high_noise_frac, upscale):
+    generator = np.random.seed(0) if seed == 0 else torch.manual_seed(seed)
+       
+    if Model == "PhotoReal":
+        pipe = DiffusionPipeline.from_pretrained("circulus/canvers-real-v3.9.1", torch_dtype=torch.float16, safety_checker=None) if torch.cuda.is_available() else DiffusionPipeline.from_pretrained("circulus/canvers-real-v3.8.1")
+        pipe.enable_xformers_memory_efficient_attention()
+        pipe = pipe.to(device)
+        torch.cuda.empty_cache()
+        if refine == "Yes":
+            refiner = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0", use_safetensors=True, torch_dtype=torch.float16, variant="fp16") if torch.cuda.is_available() else DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0")
+            refiner.enable_xformers_memory_efficient_attention()
+            refiner = refiner.to(device)
+            torch.cuda.empty_cache()
+            int_image = pipe(Prompt, negative_prompt=negative_prompt, height=height, width=width, num_inference_steps=steps, guidance_scale=scale).images
+            image = refiner(Prompt, negative_prompt=negative_prompt, image=int_image, denoising_start=high_noise_frac).images[0]
+            torch.cuda.empty_cache()
+            return image
+        else:
+            image = pipe(Prompt, negative_prompt=negative_prompt, height=height, width=width, num_inference_steps=steps, guidance_scale=scale).images[0]
+            torch.cuda.empty_cache()
+            return image
     
-    os.makedirs(output_folder, exist_ok=True)
-    base_count = len(glob(os.path.join(output_folder, "*.mp4")))
-    video_path = os.path.join(output_folder, f"{base_count:06d}.mp4")
-    video = pipe(image=image, prompt=prompt, negative_prompt=negative_prompt, width=768, height=512, num_frames=num_frames, num_inference_steps=num_inference_steps, guidance_scale=3.5, decode_noise_scale=0.3, max_sequence_length=512).frames[0]
-    export_to_video(video, video_path, fps=fps_id)
-    torch.manual_seed(seed)
-    torch.cuda.empty_cache()
-    return video_path, seed
-
-def resize_image(image, output_size=(768, 512)):
-    # Calculate aspect ratios
-    target_aspect = output_size[0] / output_size[1]  # Aspect ratio of the desired size
-    image_aspect = image.width / image.height  # Aspect ratio of the original image
+    if Model == "Animagine XL 3.0":
+        animagine = DiffusionPipeline.from_pretrained("cagliostrolab/animagine-xl-3.0", torch_dtype=torch.float16, safety_checker=None) if torch.cuda.is_available() else DiffusionPipeline.from_pretrained("cagliostrolab/animagine-xl-3.0")
+        animagine.enable_xformers_memory_efficient_attention()
+        animagine = animagine.to(device)
+        torch.cuda.empty_cache()
+        if refine == "Yes":
+            torch.cuda.empty_cache()
+            torch.cuda.max_memory_allocated(device=device)
+            int_image = animagine(Prompt, negative_prompt=negative_prompt, height=height, width=width, num_inference_steps=steps, guidance_scale=scale, output_type="latent").images
+            torch.cuda.empty_cache()
+            animagine = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0", use_safetensors=True, torch_dtype=torch.float16, variant="fp16") if torch.cuda.is_available() else DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0")
+            animagine.enable_xformers_memory_efficient_attention()
+            animagine = animagine.to(device)
+            torch.cuda.empty_cache()
+            image = animagine(Prompt, negative_prompt=negative_prompt, image=int_image, denoising_start=high_noise_frac).images[0]
+            torch.cuda.empty_cache()
+            return image    
+        else:
+            image = animagine(Prompt, negative_prompt=negative_prompt, height=height, width=width, num_inference_steps=steps, guidance_scale=scale).images[0]
+            torch.cuda.empty_cache()
+            return image
 
-    # Resize then crop if the original image is larger
-    if image_aspect > target_aspect:
-        # Resize the image to match the target height, maintaining aspect ratio
-        new_height = output_size[1]
-        new_width = int(new_height * image_aspect)
-        resized_image = image.resize((new_width, new_height), Image.LANCZOS)
-        # Calculate coordinates for cropping
-        left = (new_width - output_size[0]) / 2
-        top = 0
-        right = (new_width + output_size[0]) / 2
-        bottom = output_size[1]
-    else:
-        # Resize the image to match the target width, maintaining aspect ratio
-        new_width = output_size[0]
-        new_height = int(new_width / image_aspect)
-        resized_image = image.resize((new_width, new_height), Image.LANCZOS)
-        # Calculate coordinates for cropping
-        left = 0
-        top = (new_height - output_size[1]) / 2
-        right = output_size[0]
-        bottom = (new_height + output_size[1]) / 2
-
-    torch.cuda.empty_cache()
-    cropped_image = resized_image.crop((left, top, right, bottom))
-    return cropped_image
-
-with gr.Blocks() as demo:
-  
-  with gr.Row():
-    with gr.Column():
-        image = gr.Image(label="Upload your image", type="pil")
-        generate_btn = gr.Button("Generate")
+    if Model == "SDXL 1.0":
         
-    video = gr.Video()
-  with gr.Accordion("Advanced options", open=False):
-      seed = gr.Slider(label="Seed", value=42, randomize=True, minimum=0, maximum=max_64_bit_int, step=1)
-      randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-      prompt=gr.Textbox(label="Prompt")
-      negative_prompt=gr.Textbox(label="Negative Prompt", value="Blur, Motion Blur, Artifacts, Motion Artifacts, Poor Quality, Low Resolution, Smudging, Streaks, Motion Streaks, Pixels, Pixelated, Ugly, Mutation, Mutated")
-      num_inference_steps=gr.Slider(label="Number of Inference Steps", value=25, minimum=25, maximum=100, step=25)
-      num_frames=gr.Slider(label = "Number of Frames", value=60, minimum=30, maximum=120)
-      fps_id = gr.Slider(label="Frames per second", info="The length of your video in seconds will be frames/fps", value=12, minimum=5, maximum=30, step=1)
-      
-  image.upload(fn=resize_image, inputs=image, outputs=image, queue=False)
-  generate_btn.click(fn=sample, inputs=[image, prompt, negative_prompt, seed, randomize_seed, num_inference_steps, num_frames, fps_id], outputs=[video, seed], api_name="video")
-  
+        torch.cuda.empty_cache()
+        torch.cuda.max_memory_allocated(device=device)
+        sdxl = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True)
+        sdxl.enable_xformers_memory_efficient_attention()
+        sdxl = sdxl.to(device)   
+        torch.cuda.empty_cache()
+    
+        if refine == "Yes":
+            torch.cuda.max_memory_allocated(device=device)
+            torch.cuda.empty_cache()
+            image = sdxl(Prompt, negative_prompt=negative_prompt, height=height, width=width, num_inference_steps=steps, guidance_scale=scale, output_type="latent").images
+            torch.cuda.empty_cache()
+            sdxl = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0", use_safetensors=True, torch_dtype=torch.float16, variant="fp16") if torch.cuda.is_available() else DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0")
+            sdxl.enable_xformers_memory_efficient_attention()
+            sdxl = sdxl.to(device)
+            torch.cuda.empty_cache()
+            refined = sdxl(Prompt, negative_prompt=negative_prompt, image=image, denoising_start=high_noise_frac).images[0]
+            torch.cuda.empty_cache()
+            return refined
+        else:   
+            image = sdxl(Prompt, negative_prompt=negative_prompt, height=height, width=width, num_inference_steps=steps, guidance_scale=scale).images[0]
+            torch.cuda.empty_cache()
+            return image
+        
+    if Model == 'FusionXL':
+        torch.cuda.empty_cache()
+        torch.cuda.max_memory_allocated(device=device)
+        pipe = DiffusionPipeline.from_pretrained("circulus/canvers-fusionXL-v1", torch_dtype=torch.float16, safety_checker=None) if torch.cuda.is_available() else DiffusionPipeline.from_pretrained("circulus/canvers-real-v3.8.1")
+        pipe.enable_xformers_memory_efficient_attention()
+        pipe = pipe.to(device)
+        torch.cuda.empty_cache()
+        if refine == "Yes":
+            torch.cuda.empty_cache()
+            torch.cuda.max_memory_allocated(device=device)
+            int_image = pipe(Prompt, negative_prompt=negative_prompt, height=height, width=width, num_inference_steps=steps, guidance_scale=scale, output_type="latent").images
+            pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0", use_safetensors=True, torch_dtype=torch.float16, variant="fp16") if torch.cuda.is_available() else DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0")
+            pipe.enable_xformers_memory_efficient_attention()
+            pipe = pipe.to(device)
+            torch.cuda.empty_cache()
+            image = pipe(Prompt, negative_prompt=negative_prompt, image=int_image, denoising_start=high_noise_frac).images[0]
+            torch.cuda.empty_cache()
+            return image
+        else:
+            image = pipe(Prompt, negative_prompt=negative_prompt, height=height, width=width, num_inference_steps=steps, guidance_scale=scale).images[0]
+            torch.cuda.empty_cache()
+            return image
 
-if __name__ == "__main__":
-    demo.queue(max_size=20, api_open=False)
-    demo.launch(show_api=False)
+    return image
+    
+gr.Interface(fn=genie, inputs=[gr.Radio(['PhotoReal', 'Animagine XL 3.0', 'SDXL 1.0', 'FusionXL',], value='PhotoReal', label='Choose Model'),
+                               gr.Textbox(label='What you want the AI to generate. 77 Token Limit.'), 
+                               gr.Textbox(label='What you Do Not want the AI to generate. 77 Token Limit'),
+                               gr.Slider(512, 1024, 768, step=128, label='Height'),
+                               gr.Slider(512, 1024, 768, step=128, label='Width'),
+                               gr.Slider(1, maximum=15, value=5, step=.25, label='Guidance Scale'), 
+                               gr.Slider(5, maximum=100, value=50, step=5, label='Number of Iterations'), 
+                               gr.Slider(minimum=0, step=1, maximum=9999999999999999, randomize=True, label='Seed: 0 is Random'), 
+                               gr.Radio(["Yes", "No"], label='SDXL 1.0 Refiner: Use if the Image has too much Noise', value='No'),
+                               gr.Slider(minimum=.9, maximum=.99, value=.95, step=.01, label='Refiner Denoise Start %')],
+             outputs=gr.Image(label='Generated Image'), 
+             title="Manju Dream Booth V2.1 with SDXL 1.0 Refiner - GPU", 
+             description="<br><br><b/>Warning: This Demo is capable of producing NSFW content.", 
+             article = "If You Enjoyed this Demo and would like to Donate, you can send any amount to any of these Wallets. <br><br>SHIB (BEP20): 0xbE8f2f3B71DFEB84E5F7E3aae1909d60658aB891 <br>PayPal: https://www.paypal.me/ManjushriBodhisattva <br>ETH: 0xbE8f2f3B71DFEB84E5F7E3aae1909d60658aB891 <br>DOGE: D9QdVPtcU1EFH8jDC8jhU9uBcSTqUiA8h6<br><br>Code Monkey: <a href=\"https://huggingface.co/Manjushri\">Manjushri</a>").launch(debug=True, max_threads=80)