Update app.py

app.py

@@ -6,32 +6,28 @@ import gradio as gr
 import numpy as np
 import torch
 from PIL import Image
-from diffusers import StableDiffusionXLPipeline, EDMEulerScheduler, StableDiffusionXLInstructPix2PixPipeline, AutoencoderKL, EulerAncestralDiscreteScheduler
-from custom_pipeline import CosStableDiffusionXLInstructPix2PixPipeline
-from huggingface_hub import hf_hub_download
-from huggingface_hub import InferenceClient
+from diffusers import StableDiffusionXLImg2ImgPipeline, StableDiffusionXLPipeline, EDMEulerScheduler, StableDiffusionXLInstructPix2PixPipeline, AutoencoderKL, DPMSolverMultistepScheduler
+from huggingface_hub import hf_hub_download, InferenceClient
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-dtype = torch.float16
 vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+pipe = StableDiffusionXLPipeline.from_pretrained("SG161222/RealVisXL_V4.0", torch_dtype=torch.float16, vae=vae)
+pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True, algorithm_type="sde-dpmsolver++")
+pipe.to("cuda")
 
-repo = "fluently/Fluently-XL-Final"
+refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0", vae=vae, torch_dtype=torch.float16, use_safetensors=True, variant="fp16")
+refiner.to("cuda")
 
-pipe_best = StableDiffusionXLPipeline.from_pretrained(repo, torch_dtype=torch.float16, vae=vae)
-pipe_best.load_lora_weights("ehristoforu/dalle-3-xl-v2", weight_name="dalle-3-xl-lora-v2.safetensors", adapter_name="dalle")
-pipe_best.load_lora_weights("KingNish/Better-Image-XL-Lora", weight_name="example-03.safetensors", adapter_name="lora")
-pipe_best.set_adapters(["lora","dalle"], adapter_weights=[1.5, 0.5])
-pipe_best.to("cuda")
+pipe_fast = StableDiffusionXLPipeline.from_pretrained("SG161222/RealVisXL_V4.0_Lightning", torch_dtype=torch.float16, vae=vae, use_safetensors=True)
+pipe_fast.to("cuda")
 
-pipe_3D = StableDiffusionXLPipeline.from_pretrained(repo, torch_dtype=torch.float16, vae=vae)
-pipe_3D.load_lora_weights("artificialguybr/3DRedmond-V1", weight_name="3DRedmond-3DRenderStyle-3DRenderAF.safetensors", adapter_name="3D")
-pipe_3D.set_adapters(["3D"])
-pipe_3D.to("cuda")
-
-pipe_logo = StableDiffusionXLPipeline.from_pretrained(repo, torch_dtype=torch.float16, vae=vae)
-pipe_logo.load_lora_weights("artificialguybr/LogoRedmond-LogoLoraForSDXL", weight_name="LogoRedmond_LogoRedAF.safetensors", adapter_name="logo")
-pipe_logo.set_adapters(["logo"])
-pipe_logo.to("cuda")
+help_text = """
+To optimize image results:
+- Adjust the **Image CFG weight** if the image isn't changing enough or is changing too much. Lower it to allow bigger changes, or raise it to preserve original details.
+- Modify the **Text CFG weight** to influence how closely the edit follows text instructions. Increase it to adhere more to the text, or decrease it for subtler changes.
+- Experiment with different **random seeds** and **CFG values** for varied outcomes.
+- **Rephrase your instructions** for potentially better results.
+- **Increase the number of steps** for enhanced edits.
+"""
 
 def set_timesteps_patched(self, num_inference_steps: int, device = None):
     self.num_inference_steps = num_inference_steps
@@ -50,55 +46,86 @@ def set_timesteps_patched(self, num_inference_steps: int, device = None):
 # Image Editor
 edit_file = hf_hub_download(repo_id="stabilityai/cosxl", filename="cosxl_edit.safetensors")
 EDMEulerScheduler.set_timesteps = set_timesteps_patched
-pipe_edit = StableDiffusionXLInstructPix2PixPipeline.from_single_file(
-    edit_file, num_in_channels=8, is_cosxl_edit=True, vae=vae, torch_dtype=torch.float16,
-)
+pipe_edit = StableDiffusionXLInstructPix2PixPipeline.from_single_file( edit_file, num_in_channels=8, is_cosxl_edit=True, vae=vae, torch_dtype=torch.float16 )
 pipe_edit.scheduler = EDMEulerScheduler(sigma_min=0.002, sigma_max=120.0, sigma_data=1.0, prediction_type="v_prediction")
 pipe_edit.to("cuda")
 
+client1 = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+system_instructions1 = "<|system|>\nAct as Image Prompt Generation expert, Your task is to modify prompt by USER to more better prompt for Image Generation in Stable Diffusion XL. \n Modify the user's prompt to generate a high-quality image by incorporating essential keywords and styles according to prompt if none style is mentioned than assume realistic. The optimized prompt may include keywords according to prompt for resolution (4K, HD, 16:9 aspect ratio, , etc.), image quality (cute, masterpiece, high-quality, vivid colors, intricate details, etc.), and desired art styles (realistic, anime, 3D, logo, futuristic, fantasy, etc.). Ensure the prompt is concise, yet comprehensive and choose keywords wisely, to generate an exceptional image that meets the user's expectations. \n Your task is to reply with final optimized prompt only. If you get big prompt make it concise. and Apply all keyword at last of prompt. Reply with optimized prompt only.\n<|user|>\n"
+
+def promptifier(prompt):
+    formatted_prompt = f"{system_instructions1}{prompt}\n<|assistant|>\n"
+    stream = client1.text_generation(formatted_prompt, max_new_tokens=100)
+    return stream
+
 # Generator
-@spaces.GPU(duration=45, queue=False)
+@spaces.GPU(duration=60, queue=False)
 def king(type ,
         input_image ,
         instruction: str ,
-        steps: int = 8,
-        randomize_seed: bool = False,
-        seed: int = 25,
-        text_cfg_scale: float = 7.3,
-        image_cfg_scale: float = 1.7,
+        negative_prompt: str ="",
+        enhance_prompt: bool = True,
+        steps: int = 25,
+        randomize_seed: bool = True,
+        seed: int = 2404,
         width: int = 1024,
         height: int = 1024,
-        style="BEST",
-        use_resolution_binning: bool = True,
-        progress=gr.Progress(track_tqdm=True),
+        guidance_scale: float = 6,
+        fast=True,
+        progress=gr.Progress(track_tqdm=True)
     ):
     if type=="Image Editing" :
+        input_image = Image.open(input_image).convert('RGB')
         if randomize_seed:
-            seed = random.randint(0, 99999)
-        text_cfg_scale = text_cfg_scale
-        image_cfg_scale = image_cfg_scale
-        input_image = input_image
-
-        steps=steps
+            seed = random.randint(0, 999999)
         generator = torch.manual_seed(seed)
         output_image = pipe_edit(
-            instruction, image=input_image,
-            guidance_scale=text_cfg_scale, image_guidance_scale=image_cfg_scale,
-            num_inference_steps=steps, generator=generator).images[0]
-        return seed, output_image
+            instruction, negative_prompt=negative_prompt, image=input_image,
+            guidance_scale=guidance_scale, image_guidance_scale=1.5,
+            width = input_image.width, height = input_image.height,
+            num_inference_steps=steps, generator=generator, output_type="latent",
+        ).images
+        refine = refiner(
+            prompt=f"{instruction}, 4k, hd, high quality, masterpiece",
+            negative_prompt = negative_prompt,
+            guidance_scale=7.5,
+            num_inference_steps=steps,
+            image=output_image,
+            generator=generator,
+        ).images[0]  
+        return seed, refine
     else :
         if randomize_seed:
-            seed = random.randint(0, 99999)
+            seed = random.randint(0, 999999)
         generator = torch.Generator().manual_seed(seed)
-        if style=="3D":
-            instruction = f"3DRenderAF, 3D Render, {instruction}"
-            image = pipe_3D( prompt = instruction, guidance_scale = 5, num_inference_steps = steps, width = width, height = height, generator = generator).images[0]         
-        elif style=="Logo":
-            instruction = f"LogoRedAF, {instruction}"
-            image = pipe_logo( prompt = instruction, guidance_scale = 5, num_inference_steps = steps, width = width, height = height, generator = generator).images[0]         
-        else:
-            image = pipe_best( prompt = instruction, guidance_scale = 5, num_inference_steps = steps, width = width, height = height, generator = generator).images[0]         
-        return seed, image
+        if enhance_prompt:
+            print(f"BEFORE: {instruction} ")
+            instruction = promptifier(instruction)
+            print(f"AFTER: {instruction} ")
+        guidance_scale2=(guidance_scale/2)
+        if fast:
+            refine = pipe_fast(prompt = instruction,
+            guidance_scale = guidance_scale2, 
+            num_inference_steps = int(steps/2.5),
+            width = width, height = height,
+            generator = generator,
+            ).images[0]
+        else:            
+            image = pipe_fast( prompt = instruction,
+            negative_prompt=negative_prompt,
+            guidance_scale = guidance_scale, 
+            num_inference_steps = steps, 
+            width = width, height = height,
+            generator = generator, output_type="latent",
+            ).images 
+
+            refine = refiner( prompt=instruction,
+                    negative_prompt = negative_prompt,
+                    guidance_scale = 7.5,
+                    num_inference_steps=  steps,
+                    image=image, generator=generator,
+            ).images[0]        
+        return seed, refine
 
 client = InferenceClient()
 # Prompt classifier
@@ -106,19 +133,25 @@ def response(instruction, input_image=None ):
     if input_image is None:
         output="Image Generation"
     else:
-        text = instruction
-        labels = ["Image Editing", "Image Generation"]
-        classification = client.zero_shot_classification(text, labels, multi_label=True)
-        output = classification[0]
-        output = str(output)
-        if "Editing" in output:
-            output = "Image Editing"
-        else:
-            output = "Image Generation"
+        try:
+            text = instruction
+            labels = ["Image Editing", "Image Generation"]
+            classification = client.zero_shot_classification(text, labels, multi_label=True)
+            output = classification[0]
+            output = str(output)
+            if "Editing" in output:
+                output = "Image Editing"
+            else:
+                output = "Image Generation"
+        except:
+            if input_image is None:
+                output="Image Generation"
+            else:
+                output="Image Editing"
     return output
 
 css = '''
-.gradio-container{max-width: 600px !important}
+.gradio-container{max-width: 700px !important}
 h1{text-align:center}
 footer {
     visibility: hidden
@@ -147,47 +180,50 @@ examples=[
         [
             "Image Generation",
             None,
-            "Ironman fighting with hulk, wall painting",
-
+            "An alien grasping a sign board contain word 'ALIEN' with Neon Glow, neon, futuristic, neonpunk, neon lights",
+        ],
+        [
+            "Image Generation",
+            None,
+            "Beautiful Eiffel Tower at Night",
         ],
         [
             "Image Generation",
             None,
             "Beautiful Eiffel Tower at Night",
-
         ],
     ]
 
 with gr.Blocks(css=css, theme="Nymbo/Nymbo_Theme") as demo:
-    gr.HTML("<center><h1>Image Gen Pro</h1></center>")
+    gr.HTML("<center><h1>Image Gen & Auto Edit</h1></center>")
     with gr.Row():
-        with gr.Column(scale=4):
-            instruction = gr.Textbox(lines=1, label="Instruction", interactive=True)
-        with gr.Column(scale=1):
-            type = gr.Dropdown(["Image Generation","Image Editing"], label="Task", value="Image Generation",interactive=True, info="AI will select option based on your query, but if it selects wrong, please choose correct one.")
-        with gr.Column(scale=1):
-            generate_button = gr.Button("Generate")
+        instruction = gr.Textbox(lines=1, label="Instruction", interactive=True)
+        generate_button = gr.Button("Run", scale=0)
     with gr.Row():
-        style = gr.Radio(choices=["BEST","3D","Logo"],label="Style", value="BEST", interactive=True)
+        type = gr.Dropdown(["Image Generation","Image Editing"], label="Task", value="Image Generation",interactive=True)
+        enhance_prompt = gr.Checkbox(label="Enhance prompt", value=False, scale=0)
+        fast = gr.Checkbox(label="FAST Generation", value=True, scale=0)
+        
     with gr.Row():
-        input_image = gr.Image(label="Image", type="pil", interactive=True)
+        input_image = gr.Image(label="Image", type='filepath', interactive=True)
 
     with gr.Row():
-        width = gr.Number(value=1024, step=16,label="Width", interactive=True)
-        height = gr.Number(value=1024, step=16,label="Height", interactive=True)
-    
-    with gr.Row():
-        text_cfg_scale = gr.Number(value=7.3, step=0.1, label="Text CFG", interactive=True)
-        image_cfg_scale = gr.Number(value=1.7, step=0.1,label="Image CFG", interactive=True)
-        steps = gr.Number(value=25, precision=0, label="Steps", interactive=True)
-        randomize_seed = gr.Radio(
-                ["Fix Seed", "Randomize Seed"],
-                value="Randomize Seed",
-                type="index",
-                show_label=False,
-                interactive=True,
-            )
-        seed = gr.Number(value=1371, precision=0, label="Seed", interactive=True)
+        guidance_scale = gr.Number(value=6.0, step=0.1, label="Guidance Scale", interactive=True)
+        steps = gr.Number(value=25, step=1, label="Steps", interactive=True)
+
+    with gr.Accordion("Advanced options", open=False):
+        with gr.Row():
+            negative_prompt = gr.Text(
+                    label="Negative prompt",
+                    max_lines=1,
+                    value="(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, ugly, disgusting, blurry, amputation,(face asymmetry, eyes asymmetry, deformed eyes, open mouth)",
+                    visible=True)
+        with gr.Row():
+            width =  gr.Slider( label="Width", minimum=256, maximum=2048, step=64, value=1024)
+            height =  gr.Slider( label="Height", minimum=256, maximum=2048, step=64, value=1024)
+        with gr.Row():
+            randomize_seed = gr.Checkbox(label="Randomize Seed", value = True, interactive=True )
+            seed = gr.Number(value=2404, step=1, label="Seed", interactive=True)
 
     gr.Examples(
         examples=examples,
@@ -196,8 +232,11 @@ with gr.Blocks(css=css, theme="Nymbo/Nymbo_Theme") as demo:
         outputs=[input_image],
         cache_examples=False,
     )
-    
+
+    # gr.Markdown(help_text)
+
     instruction.change(fn=response, inputs=[instruction,input_image], outputs=type, queue=False)
+
     input_image.upload(fn=response, inputs=[instruction,input_image], outputs=type, queue=False)
     
     gr.on(triggers=[
@@ -208,16 +247,19 @@ with gr.Blocks(css=css, theme="Nymbo/Nymbo_Theme") as demo:
             inputs=[type,
                 input_image,
                 instruction,
+                negative_prompt,
+                enhance_prompt,
                 steps,
                 randomize_seed,
                 seed,
-                text_cfg_scale,
-                image_cfg_scale,
                 width,
                 height,
-                style
+                guidance_scale,
+                fast,
             ],
             outputs=[seed, input_image],
+          api_name = "image_gen_pro",
+          queue=False
         )
 
-demo.queue(max_size=99999).launch()
+demo.queue(max_size=500).launch()