Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -1,21 +1,20 @@
|
|
1 |
-
import
|
2 |
import random
|
3 |
import warnings
|
4 |
-
import
|
5 |
import gradio as gr
|
6 |
import numpy as np
|
7 |
import spaces
|
8 |
import torch
|
9 |
from diffusers import FluxImg2ImgPipeline
|
10 |
-
from transformers import AutoProcessor, AutoModelForCausalLM
|
11 |
from gradio_imageslider import ImageSlider
|
12 |
from PIL import Image
|
13 |
from huggingface_hub import snapshot_download
|
14 |
import requests
|
15 |
-
import gc
|
16 |
|
17 |
-
#
|
18 |
-
|
|
|
19 |
|
20 |
css = """
|
21 |
#col-container {
|
@@ -28,14 +27,10 @@ css = """
|
|
28 |
}
|
29 |
"""
|
30 |
|
31 |
-
# Device setup
|
32 |
-
power_device = "ZeroGPU"
|
33 |
-
device = "cpu" # Start on CPU
|
34 |
-
|
35 |
# Get HuggingFace token
|
36 |
huggingface_token = os.getenv("HF_TOKEN")
|
37 |
|
38 |
-
# Download FLUX model
|
39 |
print("π₯ Downloading FLUX model...")
|
40 |
model_path = snapshot_download(
|
41 |
repo_id="black-forest-labs/FLUX.1-dev",
|
@@ -45,408 +40,318 @@ model_path = snapshot_download(
|
|
45 |
token=huggingface_token,
|
46 |
)
|
47 |
|
48 |
-
# Load
|
49 |
-
print("π₯ Loading
|
50 |
-
florence_model = AutoModelForCausalLM.from_pretrained(
|
51 |
-
"microsoft/Florence-2-large",
|
52 |
-
torch_dtype=torch.float32,
|
53 |
-
trust_remote_code=True,
|
54 |
-
attn_implementation="eager"
|
55 |
-
).to(device).eval()
|
56 |
-
|
57 |
-
florence_processor = AutoProcessor.from_pretrained(
|
58 |
-
"microsoft/Florence-2-large",
|
59 |
-
trust_remote_code=True
|
60 |
-
)
|
61 |
-
|
62 |
-
# Load FLUX pipeline
|
63 |
-
print("π₯ Loading FLUX Img2Img...")
|
64 |
pipe = FluxImg2ImgPipeline.from_pretrained(
|
65 |
model_path,
|
66 |
-
torch_dtype=torch.
|
|
|
67 |
)
|
68 |
|
69 |
# Enable memory optimizations
|
70 |
-
pipe.enable_model_cpu_offload()
|
71 |
pipe.enable_vae_tiling()
|
72 |
pipe.enable_vae_slicing()
|
73 |
pipe.vae.enable_tiling()
|
74 |
pipe.vae.enable_slicing()
|
75 |
|
76 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
|
78 |
-
|
79 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
|
|
|
81 |
|
82 |
-
|
83 |
-
|
84 |
-
words = caption.split()
|
85 |
-
truncated = []
|
86 |
-
current_length = 0
|
87 |
-
|
88 |
-
for word in words:
|
89 |
-
# Rough estimate: 1 word β 1.3 tokens
|
90 |
-
if current_length + len(word) * 1.3 > max_tokens:
|
91 |
-
break
|
92 |
-
truncated.append(word)
|
93 |
-
current_length += len(word) * 1.3
|
94 |
-
|
95 |
-
result = ' '.join(truncated)
|
96 |
-
if len(truncated) < len(words):
|
97 |
-
result += "..."
|
98 |
-
return result
|
99 |
|
100 |
|
101 |
def make_multiple_16(n):
|
102 |
-
"""Round to nearest multiple of 16"""
|
103 |
return ((n + 15) // 16) * 16
|
104 |
|
105 |
|
106 |
-
def
|
107 |
-
"""
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
inputs = florence_processor(
|
117 |
-
text=task_prompt,
|
118 |
-
images=image,
|
119 |
-
return_tensors="pt"
|
120 |
-
).to(device)
|
121 |
-
|
122 |
-
with torch.no_grad():
|
123 |
-
generated_ids = florence_model.generate(
|
124 |
-
input_ids=inputs["input_ids"],
|
125 |
-
pixel_values=inputs["pixel_values"],
|
126 |
-
max_new_tokens=256, # Reduced from 1024
|
127 |
-
num_beams=1, # Reduced from 3
|
128 |
-
do_sample=False, # Faster without sampling
|
129 |
-
)
|
130 |
-
|
131 |
-
generated_text = florence_processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
|
132 |
-
parsed_answer = florence_processor.post_process_generation(
|
133 |
-
generated_text,
|
134 |
-
task=task_prompt,
|
135 |
-
image_size=(image.width, image.height)
|
136 |
-
)
|
137 |
-
|
138 |
-
caption = parsed_answer[task_prompt]
|
139 |
-
# Truncate to avoid CLIP token limit
|
140 |
-
caption = truncate_caption(caption, max_tokens=70)
|
141 |
-
return caption
|
142 |
-
|
143 |
-
except Exception as e:
|
144 |
-
print(f"Caption generation failed: {e}")
|
145 |
-
return "high quality detailed image"
|
146 |
|
147 |
|
148 |
-
def
|
149 |
-
"""
|
150 |
-
w, h =
|
151 |
-
w_original, h_original = w, h
|
152 |
|
153 |
-
|
|
|
|
|
154 |
|
155 |
-
|
156 |
-
if w * h * upscale_factor**2 > MAX_PIXEL_BUDGET:
|
157 |
-
gr.Info("Resizing input to fit within processing limits...")
|
158 |
-
|
159 |
-
target_pixels = MAX_PIXEL_BUDGET / (upscale_factor ** 2)
|
160 |
-
scale = (target_pixels / (w * h)) ** 0.5
|
161 |
-
|
162 |
-
new_w = make_multiple_16(int(w * scale))
|
163 |
-
new_h = make_multiple_16(int(h * scale))
|
164 |
-
|
165 |
-
input_image = input_image.resize((new_w, new_h), Image.LANCZOS)
|
166 |
-
was_resized = True
|
167 |
-
|
168 |
-
# Ensure dimensions are multiples of 16
|
169 |
-
w, h = input_image.size
|
170 |
-
new_w = make_multiple_16(w)
|
171 |
-
new_h = make_multiple_16(h)
|
172 |
-
|
173 |
-
if new_w != w or new_h != h:
|
174 |
-
padded = Image.new('RGB', (new_w, new_h))
|
175 |
-
padded.paste(input_image, (0, 0))
|
176 |
-
input_image = padded
|
177 |
-
|
178 |
-
return input_image, w_original, h_original, was_resized
|
179 |
|
180 |
|
181 |
-
def
|
182 |
-
"""
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
187 |
|
188 |
|
189 |
-
@spaces.GPU(duration=
|
190 |
def enhance_image(
|
191 |
-
|
192 |
-
|
193 |
seed,
|
194 |
randomize_seed,
|
195 |
num_inference_steps,
|
196 |
-
upscale_factor,
|
197 |
denoising_strength,
|
198 |
-
use_generated_caption,
|
199 |
-
custom_prompt,
|
200 |
progress=gr.Progress(track_tqdm=True),
|
201 |
):
|
202 |
-
"""Main enhancement function
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
203 |
try:
|
204 |
-
#
|
205 |
-
torch.cuda.empty_cache()
|
206 |
-
gc.collect()
|
207 |
-
|
208 |
-
# Handle image input
|
209 |
-
if image_input is not None:
|
210 |
-
input_image = image_input
|
211 |
-
elif image_url:
|
212 |
-
response = requests.get(image_url, stream=True)
|
213 |
-
response.raise_for_status()
|
214 |
-
input_image = Image.open(response.raw)
|
215 |
-
else:
|
216 |
-
raise gr.Error("Please provide an image")
|
217 |
-
|
218 |
if randomize_seed:
|
219 |
seed = random.randint(0, MAX_SEED)
|
220 |
|
221 |
-
|
|
|
222 |
|
223 |
-
#
|
224 |
-
input_image
|
225 |
-
|
226 |
-
)
|
227 |
|
228 |
-
#
|
229 |
-
|
230 |
-
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
236 |
|
237 |
-
#
|
238 |
-
|
239 |
-
|
|
|
240 |
|
241 |
-
|
242 |
-
|
|
|
|
|
|
|
243 |
|
244 |
-
#
|
245 |
-
|
246 |
|
247 |
-
#
|
248 |
-
|
249 |
|
250 |
-
|
251 |
-
|
252 |
-
# Process center crop for now (to avoid timeout)
|
253 |
-
crop_size = min(1024, w, h)
|
254 |
-
left = (w - crop_size) // 2
|
255 |
-
top = (h - crop_size) // 2
|
256 |
-
|
257 |
-
cropped = upscaled.crop((left, top, left + crop_size, top + crop_size))
|
258 |
-
|
259 |
-
# Ensure dimensions are multiples of 16
|
260 |
-
crop_w = make_multiple_16(cropped.width)
|
261 |
-
crop_h = make_multiple_16(cropped.height)
|
262 |
-
|
263 |
-
if crop_w != cropped.width or crop_h != cropped.height:
|
264 |
-
padded_crop = Image.new('RGB', (crop_w, crop_h))
|
265 |
-
padded_crop.paste(cropped, (0, 0))
|
266 |
-
cropped = padded_crop
|
267 |
-
|
268 |
-
# Process with FLUX
|
269 |
-
with torch.inference_mode():
|
270 |
-
generator = torch.Generator(device="cuda").manual_seed(seed)
|
271 |
-
|
272 |
-
result_crop = pipe(
|
273 |
-
prompt=prompt,
|
274 |
-
image=cropped,
|
275 |
-
strength=denoising_strength,
|
276 |
-
num_inference_steps=num_inference_steps,
|
277 |
-
guidance_scale=1.0,
|
278 |
-
height=crop_h,
|
279 |
-
width=crop_w,
|
280 |
-
generator=generator,
|
281 |
-
).images[0]
|
282 |
-
|
283 |
-
# Crop back if padded
|
284 |
-
if crop_w != cropped.width or crop_h != cropped.height:
|
285 |
-
result_crop = result_crop.crop((0, 0, cropped.width, cropped.height))
|
286 |
-
|
287 |
-
# Paste enhanced crop back
|
288 |
-
result = upscaled.copy()
|
289 |
-
result.paste(result_crop, (left, top))
|
290 |
-
|
291 |
-
else:
|
292 |
-
# Process entire image if small enough
|
293 |
-
# Ensure dimensions are multiples of 16
|
294 |
-
proc_w = make_multiple_16(w)
|
295 |
-
proc_h = make_multiple_16(h)
|
296 |
-
|
297 |
-
if proc_w != w or proc_h != h:
|
298 |
-
padded = Image.new('RGB', (proc_w, proc_h))
|
299 |
-
padded.paste(upscaled, (0, 0))
|
300 |
-
upscaled = padded
|
301 |
-
|
302 |
-
with torch.inference_mode():
|
303 |
-
generator = torch.Generator(device="cuda").manual_seed(seed)
|
304 |
-
|
305 |
-
result = pipe(
|
306 |
-
prompt=prompt,
|
307 |
-
image=upscaled,
|
308 |
-
strength=denoising_strength,
|
309 |
-
num_inference_steps=num_inference_steps,
|
310 |
-
guidance_scale=1.0,
|
311 |
-
height=proc_h,
|
312 |
-
width=proc_w,
|
313 |
-
generator=generator,
|
314 |
-
).images[0]
|
315 |
-
|
316 |
-
# Crop back if padded
|
317 |
-
if proc_w != w or proc_h != h:
|
318 |
-
result = result.crop((0, 0, w, h))
|
319 |
|
320 |
-
|
321 |
-
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
|
|
|
|
|
|
|
|
|
|
326 |
|
327 |
-
#
|
328 |
-
|
|
|
329 |
|
330 |
-
#
|
331 |
pipe.to("cpu")
|
332 |
torch.cuda.empty_cache()
|
333 |
gc.collect()
|
334 |
|
335 |
-
|
|
|
|
|
|
|
|
|
336 |
|
337 |
except Exception as e:
|
338 |
-
#
|
339 |
pipe.to("cpu")
|
|
|
340 |
torch.cuda.empty_cache()
|
341 |
gc.collect()
|
342 |
-
raise gr.Error(f"
|
343 |
|
344 |
|
345 |
-
# Gradio
|
346 |
with gr.Blocks(css=css) as demo:
|
347 |
-
gr.HTML(
|
348 |
<div class="main-header">
|
349 |
-
<h1
|
350 |
-
<p>
|
351 |
-
<p>
|
352 |
</div>
|
353 |
""")
|
354 |
|
355 |
with gr.Row():
|
356 |
with gr.Column(scale=1):
|
357 |
-
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
-
|
362 |
-
label="Upload Image",
|
363 |
-
type="pil",
|
364 |
-
height=200
|
365 |
-
)
|
366 |
-
|
367 |
-
with gr.TabItem("URL"):
|
368 |
-
image_url = gr.Textbox(
|
369 |
-
label="Image URL",
|
370 |
-
placeholder="https://example.com/image.jpg"
|
371 |
-
)
|
372 |
-
|
373 |
-
use_generated_caption = gr.Checkbox(
|
374 |
-
label="Auto-generate caption",
|
375 |
-
value=True
|
376 |
)
|
377 |
|
378 |
-
|
379 |
-
label="
|
380 |
-
placeholder="
|
|
|
381 |
lines=2
|
382 |
)
|
383 |
|
384 |
-
|
385 |
-
|
386 |
-
|
387 |
-
|
388 |
-
|
389 |
-
value=2
|
390 |
-
)
|
391 |
-
|
392 |
-
num_inference_steps = gr.Slider(
|
393 |
-
label="Quality (Steps)",
|
394 |
-
minimum=15,
|
395 |
-
maximum=30,
|
396 |
-
step=1,
|
397 |
-
value=20,
|
398 |
-
info="Higher = better but slower"
|
399 |
-
)
|
400 |
-
|
401 |
-
denoising_strength = gr.Slider(
|
402 |
-
label="Enhancement Strength",
|
403 |
-
minimum=0.1,
|
404 |
-
maximum=0.5,
|
405 |
-
step=0.05,
|
406 |
-
value=0.3,
|
407 |
-
info="Higher = more changes"
|
408 |
-
)
|
409 |
-
|
410 |
-
with gr.Row():
|
411 |
-
randomize_seed = gr.Checkbox(label="Random seed", value=True)
|
412 |
-
seed = gr.Slider(
|
413 |
-
label="Seed",
|
414 |
-
minimum=0,
|
415 |
-
maximum=MAX_SEED,
|
416 |
step=1,
|
417 |
-
value=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
418 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
419 |
|
420 |
-
enhance_btn = gr.Button(
|
|
|
|
|
|
|
|
|
421 |
|
422 |
with gr.Column(scale=2):
|
423 |
-
|
424 |
result_slider = ImageSlider(
|
425 |
type="pil",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
426 |
interactive=False,
|
427 |
-
|
428 |
-
label=None
|
429 |
)
|
430 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
431 |
enhance_btn.click(
|
432 |
fn=enhance_image,
|
433 |
inputs=[
|
434 |
-
input_image,
|
435 |
-
|
436 |
-
|
|
|
|
|
|
|
437 |
],
|
438 |
-
outputs=[result_slider]
|
439 |
)
|
440 |
|
441 |
gr.HTML("""
|
442 |
-
<div style="margin-top:
|
443 |
-
<
|
|
|
444 |
</div>
|
445 |
""")
|
446 |
|
447 |
if __name__ == "__main__":
|
448 |
demo.queue(max_size=3).launch(
|
449 |
-
share=False,
|
450 |
server_name="0.0.0.0",
|
451 |
server_port=7860
|
452 |
)
|
|
|
1 |
+
import os
|
2 |
import random
|
3 |
import warnings
|
4 |
+
import gc
|
5 |
import gradio as gr
|
6 |
import numpy as np
|
7 |
import spaces
|
8 |
import torch
|
9 |
from diffusers import FluxImg2ImgPipeline
|
|
|
10 |
from gradio_imageslider import ImageSlider
|
11 |
from PIL import Image
|
12 |
from huggingface_hub import snapshot_download
|
13 |
import requests
|
|
|
14 |
|
15 |
+
# ESRGAN imports
|
16 |
+
from basicsr.archs.rrdbnet_arch import RRDBNet
|
17 |
+
from basicsr.utils import img2tensor, tensor2img
|
18 |
|
19 |
css = """
|
20 |
#col-container {
|
|
|
27 |
}
|
28 |
"""
|
29 |
|
|
|
|
|
|
|
|
|
30 |
# Get HuggingFace token
|
31 |
huggingface_token = os.getenv("HF_TOKEN")
|
32 |
|
33 |
+
# Download FLUX model if not already cached
|
34 |
print("π₯ Downloading FLUX model...")
|
35 |
model_path = snapshot_download(
|
36 |
repo_id="black-forest-labs/FLUX.1-dev",
|
|
|
40 |
token=huggingface_token,
|
41 |
)
|
42 |
|
43 |
+
# Load FLUX pipeline on CPU initially
|
44 |
+
print("π₯ Loading FLUX Img2Img pipeline...")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
pipe = FluxImg2ImgPipeline.from_pretrained(
|
46 |
model_path,
|
47 |
+
torch_dtype=torch.bfloat16,
|
48 |
+
use_safetensors=True
|
49 |
)
|
50 |
|
51 |
# Enable memory optimizations
|
|
|
52 |
pipe.enable_vae_tiling()
|
53 |
pipe.enable_vae_slicing()
|
54 |
pipe.vae.enable_tiling()
|
55 |
pipe.vae.enable_slicing()
|
56 |
|
57 |
+
# Download and load ESRGAN 4x-UltraSharp model
|
58 |
+
print("π₯ Loading ESRGAN 4x-UltraSharp...")
|
59 |
+
esrgan_path = "4x-UltraSharp.pth"
|
60 |
+
if not os.path.exists(esrgan_path):
|
61 |
+
print("Downloading ESRGAN model...")
|
62 |
+
url = "https://huggingface.co/uwg/upscaler/resolve/main/ESRGAN/4x-UltraSharp.pth"
|
63 |
+
response = requests.get(url)
|
64 |
+
with open(esrgan_path, "wb") as f:
|
65 |
+
f.write(response.content)
|
66 |
|
67 |
+
# Initialize ESRGAN model
|
68 |
+
esrgan_model = RRDBNet(
|
69 |
+
num_in_ch=3,
|
70 |
+
num_out_ch=3,
|
71 |
+
num_feat=64,
|
72 |
+
num_block=23,
|
73 |
+
num_grow_ch=32,
|
74 |
+
scale=4
|
75 |
+
)
|
76 |
+
state_dict = torch.load(esrgan_path, map_location='cpu')
|
77 |
+
if 'params_ema' in state_dict:
|
78 |
+
state_dict = state_dict['params_ema']
|
79 |
+
elif 'params' in state_dict:
|
80 |
+
state_dict = state_dict['params']
|
81 |
+
esrgan_model.load_state_dict(state_dict)
|
82 |
+
esrgan_model.eval()
|
83 |
|
84 |
+
print("β
All models loaded successfully!")
|
85 |
|
86 |
+
MAX_SEED = 1000000
|
87 |
+
MAX_INPUT_SIZE = 512 # Max input size before upscaling
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
88 |
|
89 |
|
90 |
def make_multiple_16(n):
|
91 |
+
"""Round to nearest multiple of 16 for FLUX requirements"""
|
92 |
return ((n + 15) // 16) * 16
|
93 |
|
94 |
|
95 |
+
def truncate_prompt(prompt, max_tokens=75):
|
96 |
+
"""Truncate prompt to avoid CLIP token limit (77 tokens)"""
|
97 |
+
if not prompt:
|
98 |
+
return ""
|
99 |
+
|
100 |
+
# Simple truncation by character count (rough approximation)
|
101 |
+
if len(prompt) > 250: # ~75 tokens
|
102 |
+
return prompt[:250] + "..."
|
103 |
+
return prompt
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
|
105 |
|
106 |
+
def prepare_image(image, max_size=MAX_INPUT_SIZE):
|
107 |
+
"""Prepare image for processing"""
|
108 |
+
w, h = image.size
|
|
|
109 |
|
110 |
+
# Limit input size
|
111 |
+
if w > max_size or h > max_size:
|
112 |
+
image.thumbnail((max_size, max_size), Image.LANCZOS)
|
113 |
|
114 |
+
return image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
115 |
|
116 |
|
117 |
+
def esrgan_upscale(image):
|
118 |
+
"""Upscale image 4x using ESRGAN"""
|
119 |
+
# Convert PIL to tensor
|
120 |
+
img_np = np.array(image).astype(np.float32) / 255.
|
121 |
+
img_tensor = img2tensor(img_np, bgr2rgb=False, float32=True)
|
122 |
+
|
123 |
+
# Upscale
|
124 |
+
with torch.no_grad():
|
125 |
+
output = esrgan_model(img_tensor.unsqueeze(0).cpu())
|
126 |
+
|
127 |
+
# Convert back to PIL
|
128 |
+
output_np = tensor2img(output.squeeze(0), rgb2bgr=False, min_max=(0, 1))
|
129 |
+
return Image.fromarray(output_np)
|
130 |
|
131 |
|
132 |
+
@spaces.GPU(duration=60) # 60 seconds should be enough
|
133 |
def enhance_image(
|
134 |
+
input_image,
|
135 |
+
prompt,
|
136 |
seed,
|
137 |
randomize_seed,
|
138 |
num_inference_steps,
|
|
|
139 |
denoising_strength,
|
|
|
|
|
140 |
progress=gr.Progress(track_tqdm=True),
|
141 |
):
|
142 |
+
"""Main enhancement function"""
|
143 |
+
if input_image is None:
|
144 |
+
raise gr.Error("Please upload an image")
|
145 |
+
|
146 |
+
# Clear memory
|
147 |
+
torch.cuda.empty_cache()
|
148 |
+
gc.collect()
|
149 |
+
|
150 |
try:
|
151 |
+
# Randomize seed if needed
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
if randomize_seed:
|
153 |
seed = random.randint(0, MAX_SEED)
|
154 |
|
155 |
+
# Prepare and validate prompt
|
156 |
+
prompt = truncate_prompt(prompt.strip() if prompt else "high quality, detailed")
|
157 |
|
158 |
+
# Prepare input image
|
159 |
+
input_image = prepare_image(input_image)
|
160 |
+
original_size = input_image.size
|
|
|
161 |
|
162 |
+
# Step 1: ESRGAN upscale (4x) on CPU
|
163 |
+
gr.Info("π Upscaling with ESRGAN 4x...")
|
164 |
+
with torch.no_grad():
|
165 |
+
# Move ESRGAN to GPU for faster processing
|
166 |
+
esrgan_model.to("cuda")
|
167 |
+
|
168 |
+
# Convert image for ESRGAN
|
169 |
+
img_np = np.array(input_image).astype(np.float32) / 255.
|
170 |
+
img_tensor = img2tensor(img_np, bgr2rgb=False, float32=True)
|
171 |
+
img_tensor = img_tensor.unsqueeze(0).to("cuda")
|
172 |
+
|
173 |
+
# Upscale
|
174 |
+
output_tensor = esrgan_model(img_tensor)
|
175 |
+
|
176 |
+
# Convert back to PIL
|
177 |
+
output_np = tensor2img(output_tensor.squeeze(0).cpu(), rgb2bgr=False, min_max=(0, 1))
|
178 |
+
upscaled_image = Image.fromarray(output_np)
|
179 |
+
|
180 |
+
# Move ESRGAN back to CPU to free memory
|
181 |
+
esrgan_model.to("cpu")
|
182 |
+
torch.cuda.empty_cache()
|
183 |
|
184 |
+
# Ensure dimensions are multiples of 16 for FLUX
|
185 |
+
w, h = upscaled_image.size
|
186 |
+
new_w = make_multiple_16(w)
|
187 |
+
new_h = make_multiple_16(h)
|
188 |
|
189 |
+
if new_w != w or new_h != h:
|
190 |
+
# Pad image to meet requirements
|
191 |
+
padded = Image.new('RGB', (new_w, new_h))
|
192 |
+
padded.paste(upscaled_image, (0, 0))
|
193 |
+
upscaled_image = padded
|
194 |
|
195 |
+
# Step 2: FLUX enhancement
|
196 |
+
gr.Info("π¨ Enhancing with FLUX...")
|
197 |
|
198 |
+
# Move pipeline to GPU
|
199 |
+
pipe.to("cuda")
|
200 |
|
201 |
+
# Generate with FLUX
|
202 |
+
generator = torch.Generator(device="cuda").manual_seed(seed)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
203 |
|
204 |
+
with torch.inference_mode():
|
205 |
+
result = pipe(
|
206 |
+
prompt=prompt,
|
207 |
+
image=upscaled_image,
|
208 |
+
strength=denoising_strength,
|
209 |
+
num_inference_steps=num_inference_steps,
|
210 |
+
guidance_scale=1.0, # Fixed at 1.0 for FLUX dev
|
211 |
+
height=new_h,
|
212 |
+
width=new_w,
|
213 |
+
generator=generator,
|
214 |
+
).images[0]
|
215 |
|
216 |
+
# Crop back if we padded
|
217 |
+
if new_w != w or new_h != h:
|
218 |
+
result = result.crop((0, 0, w, h))
|
219 |
|
220 |
+
# Move pipeline back to CPU
|
221 |
pipe.to("cpu")
|
222 |
torch.cuda.empty_cache()
|
223 |
gc.collect()
|
224 |
|
225 |
+
# Prepare images for slider (before/after)
|
226 |
+
input_resized = input_image.resize(result.size, Image.LANCZOS)
|
227 |
+
|
228 |
+
gr.Info("β
Enhancement complete!")
|
229 |
+
return [input_resized, result], seed
|
230 |
|
231 |
except Exception as e:
|
232 |
+
# Cleanup on error
|
233 |
pipe.to("cpu")
|
234 |
+
esrgan_model.to("cpu")
|
235 |
torch.cuda.empty_cache()
|
236 |
gc.collect()
|
237 |
+
raise gr.Error(f"Enhancement failed: {str(e)}")
|
238 |
|
239 |
|
240 |
+
# Create Gradio interface
|
241 |
with gr.Blocks(css=css) as demo:
|
242 |
+
gr.HTML("""
|
243 |
<div class="main-header">
|
244 |
+
<h1>π ESRGAN 4x + FLUX Enhancement</h1>
|
245 |
+
<p>Upload an image to upscale 4x with ESRGAN and enhance with FLUX</p>
|
246 |
+
<p>Optimized for <strong>ZeroGPU</strong> | Max input: 512x512 β Output: 2048x2048</p>
|
247 |
</div>
|
248 |
""")
|
249 |
|
250 |
with gr.Row():
|
251 |
with gr.Column(scale=1):
|
252 |
+
# Input section
|
253 |
+
input_image = gr.Image(
|
254 |
+
label="Input Image",
|
255 |
+
type="pil",
|
256 |
+
height=256
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
257 |
)
|
258 |
|
259 |
+
prompt = gr.Textbox(
|
260 |
+
label="Enhancement Prompt",
|
261 |
+
placeholder="Describe the desired enhancement (e.g., 'high quality, sharp details, vibrant colors')",
|
262 |
+
value="high quality, ultra detailed, sharp",
|
263 |
lines=2
|
264 |
)
|
265 |
|
266 |
+
with gr.Accordion("Advanced Settings", open=False):
|
267 |
+
num_inference_steps = gr.Slider(
|
268 |
+
label="Enhancement Steps",
|
269 |
+
minimum=10,
|
270 |
+
maximum=25,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
271 |
step=1,
|
272 |
+
value=18,
|
273 |
+
info="More steps = better quality but slower"
|
274 |
+
)
|
275 |
+
|
276 |
+
denoising_strength = gr.Slider(
|
277 |
+
label="Enhancement Strength",
|
278 |
+
minimum=0.1,
|
279 |
+
maximum=0.6,
|
280 |
+
step=0.05,
|
281 |
+
value=0.35,
|
282 |
+
info="Higher = more changes to the image"
|
283 |
)
|
284 |
+
|
285 |
+
with gr.Row():
|
286 |
+
randomize_seed = gr.Checkbox(
|
287 |
+
label="Randomize seed",
|
288 |
+
value=True
|
289 |
+
)
|
290 |
+
seed = gr.Slider(
|
291 |
+
label="Seed",
|
292 |
+
minimum=0,
|
293 |
+
maximum=MAX_SEED,
|
294 |
+
step=1,
|
295 |
+
value=42
|
296 |
+
)
|
297 |
|
298 |
+
enhance_btn = gr.Button(
|
299 |
+
"π¨ Enhance Image (4x Upscale)",
|
300 |
+
variant="primary",
|
301 |
+
size="lg"
|
302 |
+
)
|
303 |
|
304 |
with gr.Column(scale=2):
|
305 |
+
# Output section
|
306 |
result_slider = ImageSlider(
|
307 |
type="pil",
|
308 |
+
label="Before / After",
|
309 |
+
interactive=False,
|
310 |
+
height=512
|
311 |
+
)
|
312 |
+
|
313 |
+
used_seed = gr.Number(
|
314 |
+
label="Seed Used",
|
315 |
interactive=False,
|
316 |
+
visible=False
|
|
|
317 |
)
|
318 |
|
319 |
+
# Examples
|
320 |
+
gr.Examples(
|
321 |
+
examples=[
|
322 |
+
["high quality, ultra detailed, sharp"],
|
323 |
+
["cinematic, professional photography, enhanced details"],
|
324 |
+
["vibrant colors, high contrast, sharp focus"],
|
325 |
+
["photorealistic, 8k quality, fine details"],
|
326 |
+
],
|
327 |
+
inputs=[prompt],
|
328 |
+
label="Example Prompts"
|
329 |
+
)
|
330 |
+
|
331 |
+
# Event handler
|
332 |
enhance_btn.click(
|
333 |
fn=enhance_image,
|
334 |
inputs=[
|
335 |
+
input_image,
|
336 |
+
prompt,
|
337 |
+
seed,
|
338 |
+
randomize_seed,
|
339 |
+
num_inference_steps,
|
340 |
+
denoising_strength,
|
341 |
],
|
342 |
+
outputs=[result_slider, used_seed]
|
343 |
)
|
344 |
|
345 |
gr.HTML("""
|
346 |
+
<div style="margin-top: 2rem; text-align: center; color: #666;">
|
347 |
+
<p>π Pipeline: ESRGAN 4x-UltraSharp β FLUX Dev Enhancement</p>
|
348 |
+
<p>β‘ Optimized for ZeroGPU with automatic memory management</p>
|
349 |
</div>
|
350 |
""")
|
351 |
|
352 |
if __name__ == "__main__":
|
353 |
demo.queue(max_size=3).launch(
|
354 |
+
share=False,
|
355 |
server_name="0.0.0.0",
|
356 |
server_port=7860
|
357 |
)
|