examples, pin version

.gitignore

@@ -0,0 +1 @@
+envs

app.py

@@ -1,3 +1,4 @@
+import glob
 import os
 import random
 
@@ -6,8 +7,12 @@ import numpy as np
 import PIL.Image
 import spaces
 import torch
-from diffusers import (AutoencoderKL, DDIMInverseScheduler, DDIMScheduler,
-                       StableDiffusionXLPipeline)
+from diffusers import (
+    AutoencoderKL,
+    DDIMInverseScheduler,
+    DDIMScheduler,
+    StableDiffusionXLPipeline,
+)
 from torchvision.transforms import ToTensor
 
 # pyright: reportPrivateImportUsage=false
@@ -15,7 +20,8 @@ from torchvision.transforms import ToTensor
 
 DESCRIPTION = f"""
     # 🎨 Inversion-InstantStyle 🎨  
-    This is an interactive demo of noisy DDIM inversion capabilities on top of Instant-Style styling method. This was developed in the context of a style benchmark : [style-rank](https://gojasper.github.io/style-rank-project) by *Eyal Benaroche, Clément Chadebec, Onur Tasar, and Benjamin Aubin* from [Jasper Research](https://www.jasper.ai/) in the context of Eyal's internship with Ecole Polytechnique.
+    This is an interactive demo of [Inversion-InstantStyle](https://gojasper.github.io/style-rank-project/#inversion_instantstyle), which combines DDIM inversion and renoising with the [Instant-Style](https://instantstyle.github.io/) styling method.
+    It was proposed in the context of our [Style-Rank](https://gojasper.github.io/style-rank-project) benchmark which evaluates training-free styling methods, by *Eyal Benaroche, Clément Chadebec, Onur Tasar, and Benjamin Aubin* from [Jasper Research](https://www.jasper.ai/) and [Ecole Polytechnique](https://www.polytechnique.edu/).
     """
 
 OPEN_SOURCE_PROMO = f"""
@@ -57,6 +63,8 @@ if gr.NO_RELOAD:
             "stabilityai/stable-diffusion-xl-base-1.0", subfolder="scheduler"
         )
         invert_scheduler = DDIMInverseScheduler(**forward_scheduler.config)
+    else:
+        raise ValueError("This demo does not work on CPU.")
 
 css = """
     h1 {
@@ -175,15 +183,20 @@ def generate(
 
 examples_prompts = [
     "Astronaut in a jungle, detailed, 8k",
-    "A Bird",
-    "A Tiger",
-    "A Cat",
-    "cactus",
-    "A Panda",
+    "A bird",
+    "A tiger",
+    "A cat",
+    "A cactus",
+    "A panda",
+    "A duck",
+    "An elephant",
+    "A dragon head",
 ]
 
-examples_images = [f"./images/{i}.png" for i in range(6)]
-
+examples_images = glob.glob("./images/*.png")
+assert len(examples_images) == len(
+    examples_prompts
+), "Number of example images and prompts should match"
 examples = [[prompt, image] for prompt, image in zip(examples_prompts, examples_images)]
 
 with gr.Blocks(css=css) as demo:
@@ -197,7 +210,17 @@ with gr.Blocks(css=css) as demo:
 
             with gr.Column():
 
-                style_image = gr.Image()
+                with gr.Row():
+                    prompt = gr.Text(
+                        label="Target prompt",
+                        show_label=False,
+                        max_lines=1,
+                        placeholder="Enter your prompt to generate new content",
+                        container=False,
+                    )
+                    run_button = gr.Button("Run", scale=0)
+
+                style_image = gr.Image(label="Style reference image")
 
                 noise_scale = gr.Slider(
                     label="Noise Scale",
@@ -210,16 +233,6 @@ with gr.Blocks(css=css) as demo:
         with gr.Blocks():
 
             with gr.Column():
-                with gr.Row():
-                    prompt = gr.Text(
-                        label="Prompt",
-                        show_label=False,
-                        max_lines=1,
-                        placeholder="Enter your prompt",
-                        container=False,
-                    )
-                    run_button = gr.Button("Run", scale=0)
-
                 result = gr.Image(label="Result", show_label=False)
 
     with gr.Accordion("Advanced options", open=False):

requirements.txt

@@ -1,12 +1,14 @@
 accelerate
 diffusers
+fastapi==0.112.2
+ftfy
 gradio==4.37.2
 numpy==1.26.4
+optimum
+peft==0.6.0
+pydantic
 spaces
-torch==2.0.1
+torch>=2.0.1
 torchvision
 transformers >= 4.34.0
-xformers
-ftfy
-peft==0.6.0
-optimum
+xformers