Spaces:
Runtime error
Runtime error
Multi image support
Browse files
app.py
CHANGED
|
@@ -6,16 +6,16 @@ import torch
|
|
| 6 |
import devicetorch
|
| 7 |
import gradio as gr
|
| 8 |
import numpy as np
|
| 9 |
-
# import spaces
|
| 10 |
from PIL import Image
|
| 11 |
-
|
|
|
|
| 12 |
from diffusers import FluxKontextPipeline
|
| 13 |
from diffusers.utils import load_image
|
| 14 |
-
from dfloat11 import DFloat11Model
|
| 15 |
|
|
|
|
| 16 |
MAX_SEED = np.iinfo(np.int32).max
|
| 17 |
|
| 18 |
-
pipe = FluxKontextPipeline.from_pretrained("fuliucansheng/FLUX.1-Kontext-dev-diffusers", torch_dtype=torch.bfloat16)
|
| 19 |
DFloat11Model.from_pretrained(
|
| 20 |
"DFloat11/FLUX.1-Kontext-dev-DF11",
|
| 21 |
device="cpu",
|
|
@@ -23,69 +23,116 @@ DFloat11Model.from_pretrained(
|
|
| 23 |
)
|
| 24 |
pipe.enable_model_cpu_offload()
|
| 25 |
|
| 26 |
-
|
| 27 |
-
def infer(input_image, prompt, seed=42, randomize_seed=False, guidance_scale=2.5, steps=28, progress=gr.Progress(track_tqdm=True)):
|
| 28 |
"""
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
This function takes an input image and a text prompt to generate a modified version
|
| 32 |
-
of the image based on the provided instructions. It uses the FLUX.1 Kontext model
|
| 33 |
-
for contextual image editing tasks.
|
| 34 |
|
| 35 |
Args:
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
prompt (str): Text description of the desired edit to apply to the image.
|
| 39 |
-
Examples: "Remove glasses", "Add a hat", "Change background to beach".
|
| 40 |
-
seed (int, optional): Random seed for reproducible generation. Defaults to 42.
|
| 41 |
-
Must be between 0 and MAX_SEED (2^31 - 1).
|
| 42 |
-
randomize_seed (bool, optional): If True, generates a random seed instead of
|
| 43 |
-
using the provided seed value. Defaults to False.
|
| 44 |
-
guidance_scale (float, optional): Controls how closely the model follows the
|
| 45 |
-
prompt. Higher values mean stronger adherence to the prompt but may reduce
|
| 46 |
-
image quality. Range: 1.0-10.0. Defaults to 2.5.
|
| 47 |
-
steps (int, optional): Controls how many steps to run the diffusion model for.
|
| 48 |
-
Range: 1-30. Defaults to 28.
|
| 49 |
-
progress (gr.Progress, optional): Gradio progress tracker for monitoring
|
| 50 |
-
generation progress. Defaults to gr.Progress(track_tqdm=True).
|
| 51 |
|
| 52 |
Returns:
|
| 53 |
-
|
| 54 |
-
- PIL.Image.Image: The generated/edited image
|
| 55 |
-
- int: The seed value used for generation (useful when randomize_seed=True)
|
| 56 |
-
- gr.update: Gradio update object to make the reuse button visible
|
| 57 |
-
|
| 58 |
-
Example:
|
| 59 |
-
>>> edited_image, used_seed, button_update = infer(
|
| 60 |
-
... input_image=my_image,
|
| 61 |
-
... prompt="Add sunglasses",
|
| 62 |
-
... seed=123,
|
| 63 |
-
... randomize_seed=False,
|
| 64 |
-
... guidance_scale=2.5
|
| 65 |
-
... )
|
| 66 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
if randomize_seed:
|
| 68 |
seed = random.randint(0, MAX_SEED)
|
| 69 |
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 88 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
gradio_temp_dir = os.environ.get('GRADIO_TEMP_DIR', tempfile.gettempdir())
|
| 90 |
temp_file_path = os.path.join(gradio_temp_dir, "image.png")
|
| 91 |
image.save(temp_file_path, format="PNG")
|
|
@@ -94,14 +141,7 @@ def infer(input_image, prompt, seed=42, randomize_seed=False, guidance_scale=2.5
|
|
| 94 |
gc.collect()
|
| 95 |
devicetorch.empty_cache(torch)
|
| 96 |
|
| 97 |
-
return image,
|
| 98 |
-
|
| 99 |
-
# @spaces.GPU
|
| 100 |
-
def infer_example(input_image, prompt):
|
| 101 |
-
image, temp_file_path, seed, _ = infer(input_image, prompt)
|
| 102 |
-
gc.collect()
|
| 103 |
-
devicetorch.empty_cache(torch)
|
| 104 |
-
return image,temp_file_path, seed
|
| 105 |
|
| 106 |
css="""
|
| 107 |
#col-container {
|
|
@@ -114,7 +154,6 @@ css="""
|
|
| 114 |
#row {
|
| 115 |
min-height: 40vh; !Important
|
| 116 |
}
|
| 117 |
-
|
| 118 |
#row-height {
|
| 119 |
height: 65px !important
|
| 120 |
}
|
|
@@ -123,17 +162,26 @@ css="""
|
|
| 123 |
with gr.Blocks(css=css) as demo:
|
| 124 |
|
| 125 |
with gr.Column(elem_id="col-container"):
|
| 126 |
-
gr.Markdown(f"""# FLUX.1 Kontext [dev]
|
| 127 |
-
|
| 128 |
""")
|
| 129 |
with gr.Row(equal_height=True):
|
| 130 |
with gr.Column():
|
| 131 |
-
|
| 132 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 133 |
with gr.Column():
|
| 134 |
result = gr.Image(label="Result", show_label=False, interactive=False, elem_classes="input-image", elem_id="row")
|
| 135 |
-
|
| 136 |
-
|
| 137 |
with gr.Row(equal_height=True):
|
| 138 |
with gr.Column():
|
| 139 |
prompt = gr.Text(
|
|
@@ -145,14 +193,14 @@ Image editing and manipulation model guidance-distilled from FLUX.1 Kontext [pro
|
|
| 145 |
container=True,
|
| 146 |
scale=1
|
| 147 |
)
|
| 148 |
-
|
| 149 |
with gr.Column():
|
| 150 |
-
download_image = gr.File(label="Download Image", elem_id="row-height", scale=0)
|
| 151 |
run_button = gr.Button("Run", scale=1)
|
| 152 |
|
| 153 |
with gr.Row():
|
| 154 |
with gr.Accordion("Advanced Settings", open=False):
|
| 155 |
-
|
| 156 |
seed = gr.Slider(
|
| 157 |
label="Seed",
|
| 158 |
minimum=0,
|
|
@@ -168,39 +216,30 @@ Image editing and manipulation model guidance-distilled from FLUX.1 Kontext [pro
|
|
| 168 |
minimum=1,
|
| 169 |
maximum=10,
|
| 170 |
step=0.1,
|
| 171 |
-
value=
|
| 172 |
)
|
| 173 |
|
| 174 |
steps = gr.Slider(
|
| 175 |
label="Steps",
|
| 176 |
minimum=1,
|
| 177 |
maximum=40,
|
| 178 |
-
value=
|
| 179 |
step=1
|
| 180 |
-
)
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
["flowers.png", "turn the flowers into sunflowers"],
|
| 185 |
-
["monster.png", "make this monster ride a skateboard on the beach"],
|
| 186 |
-
["cat.png", "make this cat happy"]
|
| 187 |
-
],
|
| 188 |
-
inputs=[input_image, prompt],
|
| 189 |
-
outputs=[result, download_image, seed],
|
| 190 |
-
fn=infer_example,
|
| 191 |
-
cache_examples=False
|
| 192 |
-
)
|
| 193 |
-
|
| 194 |
gr.on(
|
| 195 |
triggers=[run_button.click, prompt.submit],
|
| 196 |
fn = infer,
|
| 197 |
-
inputs = [
|
| 198 |
-
outputs = [result,
|
| 199 |
)
|
|
|
|
| 200 |
reuse_button.click(
|
| 201 |
-
fn = lambda image: image,
|
| 202 |
inputs = [result],
|
| 203 |
-
outputs = [
|
| 204 |
)
|
| 205 |
|
| 206 |
demo.launch(mcp_server=True)
|
|
|
|
| 6 |
import devicetorch
|
| 7 |
import gradio as gr
|
| 8 |
import numpy as np
|
|
|
|
| 9 |
from PIL import Image
|
| 10 |
+
from dfloat11 import DFloat11Model
|
| 11 |
+
#from kontext_pipeline import FluxKontextPipeline
|
| 12 |
from diffusers import FluxKontextPipeline
|
| 13 |
from diffusers.utils import load_image
|
|
|
|
| 14 |
|
| 15 |
+
# Load Kontext model
|
| 16 |
MAX_SEED = np.iinfo(np.int32).max
|
| 17 |
|
| 18 |
+
pipe = FluxKontextPipeline.from_pretrained("fuliucansheng/FLUX.1-Kontext-dev-diffusers", torch_dtype=torch.bfloat16).to("cuda")
|
| 19 |
DFloat11Model.from_pretrained(
|
| 20 |
"DFloat11/FLUX.1-Kontext-dev-DF11",
|
| 21 |
device="cpu",
|
|
|
|
| 23 |
)
|
| 24 |
pipe.enable_model_cpu_offload()
|
| 25 |
|
| 26 |
+
def concatenate_images(images, direction="horizontal"):
|
|
|
|
| 27 |
"""
|
| 28 |
+
Concatenate multiple PIL images either horizontally or vertically.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
|
| 30 |
Args:
|
| 31 |
+
images: List of PIL Images
|
| 32 |
+
direction: "horizontal" or "vertical"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
|
| 34 |
Returns:
|
| 35 |
+
PIL Image: Concatenated image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
"""
|
| 37 |
+
if not images:
|
| 38 |
+
return None
|
| 39 |
+
|
| 40 |
+
# Filter out None images
|
| 41 |
+
valid_images = [img for img in images if img is not None]
|
| 42 |
+
|
| 43 |
+
if not valid_images:
|
| 44 |
+
return None
|
| 45 |
+
|
| 46 |
+
if len(valid_images) == 1:
|
| 47 |
+
return valid_images[0].convert("RGB")
|
| 48 |
+
|
| 49 |
+
# Convert all images to RGB
|
| 50 |
+
valid_images = [img.convert("RGB") for img in valid_images]
|
| 51 |
+
|
| 52 |
+
if direction == "horizontal":
|
| 53 |
+
# Calculate total width and max height
|
| 54 |
+
total_width = sum(img.width for img in valid_images)
|
| 55 |
+
max_height = max(img.height for img in valid_images)
|
| 56 |
+
|
| 57 |
+
# Create new image
|
| 58 |
+
concatenated = Image.new('RGB', (total_width, max_height), (255, 255, 255))
|
| 59 |
+
|
| 60 |
+
# Paste images
|
| 61 |
+
x_offset = 0
|
| 62 |
+
for img in valid_images:
|
| 63 |
+
# Center image vertically if heights differ
|
| 64 |
+
y_offset = (max_height - img.height) // 2
|
| 65 |
+
concatenated.paste(img, (x_offset, y_offset))
|
| 66 |
+
x_offset += img.width
|
| 67 |
+
|
| 68 |
+
else: # vertical
|
| 69 |
+
# Calculate max width and total height
|
| 70 |
+
max_width = max(img.width for img in valid_images)
|
| 71 |
+
total_height = sum(img.height for img in valid_images)
|
| 72 |
+
|
| 73 |
+
# Create new image
|
| 74 |
+
concatenated = Image.new('RGB', (max_width, total_height), (255, 255, 255))
|
| 75 |
+
|
| 76 |
+
# Paste images
|
| 77 |
+
y_offset = 0
|
| 78 |
+
for img in valid_images:
|
| 79 |
+
# Center image horizontally if widths differ
|
| 80 |
+
x_offset = (max_width - img.width) // 2
|
| 81 |
+
concatenated.paste(img, (x_offset, y_offset))
|
| 82 |
+
y_offset += img.height
|
| 83 |
+
|
| 84 |
+
return concatenated
|
| 85 |
+
|
| 86 |
+
def infer(input_images, prompt, seed=42, randomize_seed=False, guidance_scale=4.0, steps=25, progress=gr.Progress(track_tqdm=True)):
|
| 87 |
+
|
| 88 |
if randomize_seed:
|
| 89 |
seed = random.randint(0, MAX_SEED)
|
| 90 |
|
| 91 |
+
# Handle input_images - it could be a single image or a list of images
|
| 92 |
+
if input_images is None:
|
| 93 |
+
raise gr.Error("Please upload at least one image.")
|
| 94 |
+
|
| 95 |
+
# If it's a single image (not a list), convert to list
|
| 96 |
+
if not isinstance(input_images, list):
|
| 97 |
+
input_images = [input_images]
|
| 98 |
+
|
| 99 |
+
# Filter out None images
|
| 100 |
+
valid_images = [img[0] for img in input_images if img is not None]
|
| 101 |
+
|
| 102 |
+
if not valid_images:
|
| 103 |
+
raise gr.Error("Please upload at least one valid image.")
|
| 104 |
+
|
| 105 |
+
# Concatenate images horizontally
|
| 106 |
+
concatenated_image = concatenate_images(valid_images, "horizontal")
|
| 107 |
+
|
| 108 |
+
if concatenated_image is None:
|
| 109 |
+
raise gr.Error("Failed to process the input images.")
|
| 110 |
+
|
| 111 |
+
# original_width, original_height = concatenated_image.size
|
| 112 |
+
|
| 113 |
+
# if original_width >= original_height:
|
| 114 |
+
# new_width = 1024
|
| 115 |
+
# new_height = int(original_height * (new_width / original_width))
|
| 116 |
+
# new_height = round(new_height / 64) * 64
|
| 117 |
+
# else:
|
| 118 |
+
# new_height = 1024
|
| 119 |
+
# new_width = int(original_width * (new_height / original_height))
|
| 120 |
+
# new_width = round(new_width / 64) * 64
|
| 121 |
+
|
| 122 |
+
#concatenated_image_resized = concatenated_image.resize((new_width, new_height), Image.LANCZOS)
|
| 123 |
|
| 124 |
+
final_prompt = f"From the provided reference images, create a unified, cohesive image such that {prompt}. Maintain the identity and characteristics of each subject while adjusting their proportions, scale, and positioning to create a harmonious, naturally balanced composition. Blend and integrate all elements seamlessly with consistent lighting, perspective, and style.the final result should look like a single naturally captured scene where all subjects are properly sized and positioned relative to each other, not assembled from multiple sources."
|
| 125 |
+
|
| 126 |
+
image = pipe(
|
| 127 |
+
image=concatenated_image,
|
| 128 |
+
prompt=final_prompt,
|
| 129 |
+
guidance_scale=guidance_scale,
|
| 130 |
+
width=concatenated_image.size[0],
|
| 131 |
+
height=concatenated_image.size[1],
|
| 132 |
+
num_inference_steps=steps,
|
| 133 |
+
generator=torch.Generator().manual_seed(seed),
|
| 134 |
+
).images[0]
|
| 135 |
+
|
| 136 |
gradio_temp_dir = os.environ.get('GRADIO_TEMP_DIR', tempfile.gettempdir())
|
| 137 |
temp_file_path = os.path.join(gradio_temp_dir, "image.png")
|
| 138 |
image.save(temp_file_path, format="PNG")
|
|
|
|
| 141 |
gc.collect()
|
| 142 |
devicetorch.empty_cache(torch)
|
| 143 |
|
| 144 |
+
return image, seed, gr.update(visible=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 145 |
|
| 146 |
css="""
|
| 147 |
#col-container {
|
|
|
|
| 154 |
#row {
|
| 155 |
min-height: 40vh; !Important
|
| 156 |
}
|
|
|
|
| 157 |
#row-height {
|
| 158 |
height: 65px !important
|
| 159 |
}
|
|
|
|
| 162 |
with gr.Blocks(css=css) as demo:
|
| 163 |
|
| 164 |
with gr.Column(elem_id="col-container"):
|
| 165 |
+
gr.Markdown(f"""# FLUX.1 Kontext [dev] - Multi-Image
|
| 166 |
+
Flux Kontext with multiple image input support - compose a new image with elements from multiple images using Kontext [dev]
|
| 167 |
""")
|
| 168 |
with gr.Row(equal_height=True):
|
| 169 |
with gr.Column():
|
| 170 |
+
input_images = gr.Gallery(
|
| 171 |
+
label="Upload image(s) for editing",
|
| 172 |
+
show_label=True,
|
| 173 |
+
elem_id="gallery_input",
|
| 174 |
+
columns=3,
|
| 175 |
+
rows=2,
|
| 176 |
+
object_fit="contain",
|
| 177 |
+
height="auto",
|
| 178 |
+
file_types=['image'],
|
| 179 |
+
type='pil'
|
| 180 |
+
)
|
| 181 |
+
|
| 182 |
with gr.Column():
|
| 183 |
result = gr.Image(label="Result", show_label=False, interactive=False, elem_classes="input-image", elem_id="row")
|
| 184 |
+
|
|
|
|
| 185 |
with gr.Row(equal_height=True):
|
| 186 |
with gr.Column():
|
| 187 |
prompt = gr.Text(
|
|
|
|
| 193 |
container=True,
|
| 194 |
scale=1
|
| 195 |
)
|
| 196 |
+
|
| 197 |
with gr.Column():
|
| 198 |
+
download_image = gr.File(label="Download Image", elem_id="row-height", interactive=False, scale=0)
|
| 199 |
run_button = gr.Button("Run", scale=1)
|
| 200 |
|
| 201 |
with gr.Row():
|
| 202 |
with gr.Accordion("Advanced Settings", open=False):
|
| 203 |
+
|
| 204 |
seed = gr.Slider(
|
| 205 |
label="Seed",
|
| 206 |
minimum=0,
|
|
|
|
| 216 |
minimum=1,
|
| 217 |
maximum=10,
|
| 218 |
step=0.1,
|
| 219 |
+
value=4.0,
|
| 220 |
)
|
| 221 |
|
| 222 |
steps = gr.Slider(
|
| 223 |
label="Steps",
|
| 224 |
minimum=1,
|
| 225 |
maximum=40,
|
| 226 |
+
value=25,
|
| 227 |
step=1
|
| 228 |
+
)
|
| 229 |
+
|
| 230 |
+
reuse_button = gr.Button("Reuse this image", visible=False)
|
| 231 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 232 |
gr.on(
|
| 233 |
triggers=[run_button.click, prompt.submit],
|
| 234 |
fn = infer,
|
| 235 |
+
inputs = [input_images, prompt, seed, randomize_seed, guidance_scale, steps],
|
| 236 |
+
outputs = [result, seed, reuse_button]
|
| 237 |
)
|
| 238 |
+
|
| 239 |
reuse_button.click(
|
| 240 |
+
fn = lambda image: [image] if image is not None else [], # Convert single image to list for gallery
|
| 241 |
inputs = [result],
|
| 242 |
+
outputs = [input_images]
|
| 243 |
)
|
| 244 |
|
| 245 |
demo.launch(mcp_server=True)
|