Spaces:
Running
on
Zero
Running
on
Zero
Commit
•
4f5b1e9
1
Parent(s):
16490f6
Update app.py
Browse files
app.py
CHANGED
@@ -25,17 +25,18 @@ base_model = "black-forest-labs/FLUX.1-dev"
|
|
25 |
taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to(device)
|
26 |
good_vae = AutoencoderKL.from_pretrained(base_model, subfolder="vae", torch_dtype=dtype).to(device)
|
27 |
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=dtype, vae=taef1).to(device)
|
28 |
-
pipe_i2i = AutoPipelineForImage2Image.from_pretrained(
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
|
|
39 |
|
40 |
pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)
|
41 |
|
@@ -46,7 +47,7 @@ class calculateDuration:
|
|
46 |
def __enter__(self):
|
47 |
self.start_time = time.time()
|
48 |
return self
|
49 |
-
|
50 |
def __exit__(self, exc_type, exc_value, traceback):
|
51 |
self.end_time = time.time()
|
52 |
self.elapsed_time = self.end_time - self.start_time
|
@@ -66,7 +67,7 @@ def update_selection(evt: gr.SelectData, selected_indices, width, height):
|
|
66 |
selected_indices.append(selected_index)
|
67 |
else:
|
68 |
gr.Warning("You can select up to 2 LoRAs, remove one to select a new one.")
|
69 |
-
return(
|
70 |
gr.update(),
|
71 |
gr.update(),
|
72 |
gr.update(),
|
@@ -80,19 +81,19 @@ def update_selection(evt: gr.SelectData, selected_indices, width, height):
|
|
80 |
)
|
81 |
|
82 |
# Initialize outputs
|
83 |
-
selected_info_1 = ""
|
84 |
-
selected_info_2 = ""
|
85 |
lora_scale_1 = 0.95
|
86 |
lora_scale_2 = 0.95
|
87 |
lora_image_1 = None
|
88 |
lora_image_2 = None
|
89 |
if len(selected_indices) >= 1:
|
90 |
lora1 = loras[selected_indices[0]]
|
91 |
-
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}](
|
92 |
lora_image_1 = lora1['image']
|
93 |
if len(selected_indices) >= 2:
|
94 |
lora2 = loras[selected_indices[1]]
|
95 |
-
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}](
|
96 |
lora_image_2 = lora2['image']
|
97 |
|
98 |
# Update prompt placeholder based on last selected LoRA
|
@@ -128,11 +129,11 @@ def remove_lora_1(selected_indices):
|
|
128 |
lora_image_2 = None
|
129 |
if len(selected_indices) >= 1:
|
130 |
lora1 = loras[selected_indices[0]]
|
131 |
-
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}](
|
132 |
lora_image_1 = lora1['image']
|
133 |
if len(selected_indices) >= 2:
|
134 |
lora2 = loras[selected_indices[1]]
|
135 |
-
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}](
|
136 |
lora_image_2 = lora2['image']
|
137 |
return selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2
|
138 |
|
@@ -149,11 +150,11 @@ def remove_lora_2(selected_indices):
|
|
149 |
lora_image_2 = None
|
150 |
if len(selected_indices) >= 1:
|
151 |
lora1 = loras[selected_indices[0]]
|
152 |
-
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}](
|
153 |
lora_image_1 = lora1['image']
|
154 |
if len(selected_indices) >= 2:
|
155 |
lora2 = loras[selected_indices[1]]
|
156 |
-
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}](
|
157 |
lora_image_2 = lora2['image']
|
158 |
return selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2
|
159 |
|
@@ -163,8 +164,8 @@ def randomize_loras(selected_indices):
|
|
163 |
selected_indices = random.sample(range(len(loras)), 2)
|
164 |
lora1 = loras[selected_indices[0]]
|
165 |
lora2 = loras[selected_indices[1]]
|
166 |
-
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}](
|
167 |
-
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}](
|
168 |
lora_scale_1 = 0.95
|
169 |
lora_scale_2 = 0.95
|
170 |
lora_image_1 = lora1['image']
|
@@ -173,7 +174,7 @@ def randomize_loras(selected_indices):
|
|
173 |
|
174 |
@spaces.GPU(duration=70)
|
175 |
def generate_image(prompt_mash, steps, seed, cfg_scale, width, height, progress):
|
176 |
-
print("
|
177 |
pipe.to("cuda")
|
178 |
generator = torch.Generator(device="cuda").manual_seed(seed)
|
179 |
with calculateDuration("Generating image"):
|
@@ -208,8 +209,8 @@ def generate_image_to_image(prompt_mash, image_input_path, image_strength, steps
|
|
208 |
joint_attention_kwargs={"scale": 1.0},
|
209 |
output_type="pil",
|
210 |
).images[0]
|
211 |
-
return final_image
|
212 |
-
|
213 |
def run_lora(prompt, image_input, image_strength, cfg_scale, steps, selected_indices, lora_scale_1, lora_scale_2, randomize_seed, seed, width, height, progress=gr.Progress(track_tqdm=True)):
|
214 |
if not selected_indices:
|
215 |
raise gr.Error("You must select at least one LoRA before proceeding.")
|
@@ -235,27 +236,29 @@ def run_lora(prompt, image_input, image_strength, cfg_scale, steps, selected_ind
|
|
235 |
|
236 |
# Load LoRA weights with respective scales
|
237 |
lora_names = []
|
|
|
238 |
with calculateDuration("Loading LoRA weights"):
|
239 |
for idx, lora in enumerate(selected_loras):
|
240 |
lora_name = f"lora_{idx}"
|
241 |
lora_names.append(lora_name)
|
|
|
242 |
lora_path = lora['repo']
|
243 |
-
|
244 |
if image_input is not None:
|
245 |
-
if
|
246 |
-
pipe_i2i.load_lora_weights(lora_path, weight_name=
|
247 |
else:
|
248 |
pipe_i2i.load_lora_weights(lora_path, low_cpu_mem_usage=True, adapter_name=lora_name)
|
249 |
else:
|
250 |
-
if
|
251 |
-
pipe.load_lora_weights(lora_path, weight_name=
|
252 |
else:
|
253 |
pipe.load_lora_weights(lora_path, low_cpu_mem_usage=True, adapter_name=lora_name)
|
254 |
-
print(lora_names)
|
255 |
if image_input is not None:
|
256 |
-
pipe_i2i.set_adapters(lora_names, adapter_weights=
|
257 |
else:
|
258 |
-
pipe.set_adapters(lora_names, adapter_weights=
|
259 |
# Set random seed for reproducibility
|
260 |
with calculateDuration("Randomizing seed"):
|
261 |
if randomize_seed:
|
@@ -271,7 +274,7 @@ def run_lora(prompt, image_input, image_strength, cfg_scale, steps, selected_ind
|
|
271 |
final_image = None
|
272 |
step_counter = 0
|
273 |
for image in image_generator:
|
274 |
-
step_counter+=1
|
275 |
final_image = image
|
276 |
progress_bar = f'<div class="progress-container"><div class="progress-bar" style="--current: {step_counter}; --total: {steps};"></div></div>'
|
277 |
yield image, seed, gr.update(value=progress_bar, visible=True)
|
@@ -282,7 +285,7 @@ def get_huggingface_safetensors(link):
|
|
282 |
if len(split_link) == 2:
|
283 |
model_card = ModelCard.load(link)
|
284 |
base_model = model_card.data.get("base_model")
|
285 |
-
print(base_model)
|
286 |
if base_model not in ["black-forest-labs/FLUX.1-dev", "black-forest-labs/FLUX.1-schnell"]:
|
287 |
raise Exception("Not a FLUX LoRA!")
|
288 |
image_path = model_card.data.get("widget", [{}])[0].get("output", {}).get("url", None)
|
@@ -304,13 +307,26 @@ def get_huggingface_safetensors(link):
|
|
304 |
if not safetensors_name:
|
305 |
raise Exception("No *.safetensors file found in the repository")
|
306 |
return split_link[1], link, safetensors_name, trigger_word, image_url
|
|
|
|
|
307 |
|
308 |
def check_custom_model(link):
|
309 |
-
if link.
|
310 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
311 |
link_split = link.split("huggingface.co/")
|
312 |
return get_huggingface_safetensors(link_split[1])
|
313 |
-
|
|
|
|
|
|
|
314 |
return get_huggingface_safetensors(link)
|
315 |
|
316 |
def add_custom_lora(custom_lora, selected_indices):
|
@@ -319,18 +335,6 @@ def add_custom_lora(custom_lora, selected_indices):
|
|
319 |
try:
|
320 |
title, repo, path, trigger_word, image = check_custom_model(custom_lora)
|
321 |
print(f"Loaded custom LoRA: {repo}")
|
322 |
-
card = f'''
|
323 |
-
<div class="custom_lora_card">
|
324 |
-
<span>Loaded custom LoRA:</span>
|
325 |
-
<div class="card_internal">
|
326 |
-
<img src="{image}" />
|
327 |
-
<div>
|
328 |
-
<h3>{title}</h3>
|
329 |
-
<small>{"Using: <code><b>"+trigger_word+"</code></b> as the trigger word" if trigger_word else "No trigger word found. If there's a trigger word, include it in your prompt"}<br></small>
|
330 |
-
</div>
|
331 |
-
</div>
|
332 |
-
</div>
|
333 |
-
'''
|
334 |
existing_item_index = next((index for (index, item) in enumerate(loras) if item['repo'] == repo), None)
|
335 |
if existing_item_index is None:
|
336 |
new_item = {
|
@@ -340,7 +344,7 @@ def add_custom_lora(custom_lora, selected_indices):
|
|
340 |
"weights": path,
|
341 |
"trigger_word": trigger_word
|
342 |
}
|
343 |
-
print(new_item)
|
344 |
existing_item_index = len(loras)
|
345 |
loras.append(new_item)
|
346 |
|
@@ -349,32 +353,42 @@ def add_custom_lora(custom_lora, selected_indices):
|
|
349 |
# Update selected_indices if there's room
|
350 |
if len(selected_indices) < 2:
|
351 |
selected_indices.append(existing_item_index)
|
352 |
-
selected_info_1 = ""
|
353 |
-
selected_info_2 = ""
|
354 |
-
lora_scale_1 = 0.95
|
355 |
-
lora_scale_2 = 0.95
|
356 |
-
lora_image_1 = None
|
357 |
-
lora_image_2 = None
|
358 |
-
if len(selected_indices) >= 1:
|
359 |
-
lora1 = loras[selected_indices[0]]
|
360 |
-
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}](https://huggingface.co/{lora1['repo']}) ✨"
|
361 |
-
lora_image_1 = lora1['image']
|
362 |
-
if len(selected_indices) >= 2:
|
363 |
-
lora2 = loras[selected_indices[1]]
|
364 |
-
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}](https://huggingface.co/{lora2['repo']}) ✨"
|
365 |
-
lora_image_2 = lora2['image']
|
366 |
-
return (gr.update(visible=True, value=card), gr.update(visible=True), gr.update(value=gallery_items),
|
367 |
-
selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2)
|
368 |
else:
|
369 |
-
|
370 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
371 |
except Exception as e:
|
372 |
print(e)
|
373 |
-
|
|
|
374 |
else:
|
375 |
-
return gr.
|
376 |
|
377 |
-
def remove_custom_lora(
|
378 |
global loras
|
379 |
if loras:
|
380 |
custom_lora_repo = loras[-1]['repo']
|
@@ -387,21 +401,30 @@ def remove_custom_lora(custom_lora_info, custom_lora_button, selected_indices):
|
|
387 |
# Update gallery
|
388 |
gallery_items = [(item["image"], item["title"]) for item in loras]
|
389 |
# Update selected_info and images
|
390 |
-
selected_info_1 = ""
|
391 |
-
selected_info_2 = ""
|
392 |
lora_scale_1 = 0.95
|
393 |
lora_scale_2 = 0.95
|
394 |
lora_image_1 = None
|
395 |
lora_image_2 = None
|
396 |
if len(selected_indices) >= 1:
|
397 |
lora1 = loras[selected_indices[0]]
|
398 |
-
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}](
|
399 |
lora_image_1 = lora1['image']
|
400 |
if len(selected_indices) >= 2:
|
401 |
lora2 = loras[selected_indices[1]]
|
402 |
-
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}](
|
403 |
lora_image_2 = lora2['image']
|
404 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
405 |
|
406 |
run_lora.zerogpu = True
|
407 |
|
@@ -410,6 +433,7 @@ css = '''
|
|
410 |
#title{text-align: center}
|
411 |
#title h1{font-size: 3em; display:inline-flex; align-items:center}
|
412 |
#title img{width: 100px; margin-right: 0.5em}
|
|
|
413 |
#gallery .grid-wrap{height: 5vh}
|
414 |
#lora_list{background: var(--block-background-fill);padding: 0 1em .3em; font-size: 90%}
|
415 |
.custom_lora_card{margin-bottom: 1em}
|
@@ -460,6 +484,11 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css, delete_cache=(60, 3600)) as app:
|
|
460 |
remove_button_2 = gr.Button("Remove", size="sm")
|
461 |
with gr.Row():
|
462 |
with gr.Column():
|
|
|
|
|
|
|
|
|
|
|
463 |
gallery = gr.Gallery(
|
464 |
[(item["image"], item["title"]) for item in loras],
|
465 |
label="LoRA Gallery",
|
@@ -467,11 +496,6 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css, delete_cache=(60, 3600)) as app:
|
|
467 |
columns=5,
|
468 |
elem_id="gallery"
|
469 |
)
|
470 |
-
with gr.Group():
|
471 |
-
custom_lora = gr.Textbox(label="Custom LoRA", info="LoRA Hugging Face path", placeholder="multimodalart/vintage-ads-flux")
|
472 |
-
gr.Markdown("[Check the list of FLUX LoRAs](https://huggingface.co/models?other=base_model:adapter:black-forest-labs/FLUX.1-dev)", elem_id="lora_list")
|
473 |
-
custom_lora_info = gr.HTML(visible=False)
|
474 |
-
custom_lora_button = gr.Button("Remove custom LoRA", visible=False)
|
475 |
with gr.Column():
|
476 |
progress_bar = gr.Markdown(elem_id="progress", visible=False)
|
477 |
result = gr.Image(label="Generated Image")
|
@@ -484,15 +508,15 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css, delete_cache=(60, 3600)) as app:
|
|
484 |
with gr.Row():
|
485 |
cfg_scale = gr.Slider(label="CFG Scale", minimum=1, maximum=20, step=0.5, value=3.5)
|
486 |
steps = gr.Slider(label="Steps", minimum=1, maximum=50, step=1, value=28)
|
487 |
-
|
488 |
with gr.Row():
|
489 |
width = gr.Slider(label="Width", minimum=256, maximum=1536, step=64, value=1024)
|
490 |
height = gr.Slider(label="Height", minimum=256, maximum=1536, step=64, value=1024)
|
491 |
-
|
492 |
with gr.Row():
|
493 |
randomize_seed = gr.Checkbox(True, label="Randomize seed")
|
494 |
seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0, randomize=True)
|
495 |
-
|
496 |
gallery.select(
|
497 |
update_selection,
|
498 |
inputs=[selected_indices, width, height],
|
@@ -513,15 +537,15 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css, delete_cache=(60, 3600)) as app:
|
|
513 |
inputs=[selected_indices],
|
514 |
outputs=[selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2]
|
515 |
)
|
516 |
-
|
517 |
add_custom_lora,
|
518 |
inputs=[custom_lora, selected_indices],
|
519 |
-
outputs=[
|
520 |
)
|
521 |
-
|
522 |
remove_custom_lora,
|
523 |
-
inputs=[
|
524 |
-
outputs=[
|
525 |
)
|
526 |
gr.on(
|
527 |
triggers=[generate_button.click, prompt.submit],
|
|
|
25 |
taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to(device)
|
26 |
good_vae = AutoencoderKL.from_pretrained(base_model, subfolder="vae", torch_dtype=dtype).to(device)
|
27 |
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=dtype, vae=taef1).to(device)
|
28 |
+
pipe_i2i = AutoPipelineForImage2Image.from_pretrained(
|
29 |
+
base_model,
|
30 |
+
vae=good_vae,
|
31 |
+
transformer=pipe.transformer,
|
32 |
+
text_encoder=pipe.text_encoder,
|
33 |
+
tokenizer=pipe.tokenizer,
|
34 |
+
text_encoder_2=pipe.text_encoder_2,
|
35 |
+
tokenizer_2=pipe.tokenizer_2,
|
36 |
+
torch_dtype=dtype
|
37 |
+
)
|
38 |
+
|
39 |
+
MAX_SEED = 2**32 - 1
|
40 |
|
41 |
pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)
|
42 |
|
|
|
47 |
def __enter__(self):
|
48 |
self.start_time = time.time()
|
49 |
return self
|
50 |
+
|
51 |
def __exit__(self, exc_type, exc_value, traceback):
|
52 |
self.end_time = time.time()
|
53 |
self.elapsed_time = self.end_time - self.start_time
|
|
|
67 |
selected_indices.append(selected_index)
|
68 |
else:
|
69 |
gr.Warning("You can select up to 2 LoRAs, remove one to select a new one.")
|
70 |
+
return (
|
71 |
gr.update(),
|
72 |
gr.update(),
|
73 |
gr.update(),
|
|
|
81 |
)
|
82 |
|
83 |
# Initialize outputs
|
84 |
+
selected_info_1 = "Select a LoRA 1"
|
85 |
+
selected_info_2 = "Select a LoRA 2"
|
86 |
lora_scale_1 = 0.95
|
87 |
lora_scale_2 = 0.95
|
88 |
lora_image_1 = None
|
89 |
lora_image_2 = None
|
90 |
if len(selected_indices) >= 1:
|
91 |
lora1 = loras[selected_indices[0]]
|
92 |
+
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}]({lora1['repo']}) ✨"
|
93 |
lora_image_1 = lora1['image']
|
94 |
if len(selected_indices) >= 2:
|
95 |
lora2 = loras[selected_indices[1]]
|
96 |
+
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}]({lora2['repo']}) ✨"
|
97 |
lora_image_2 = lora2['image']
|
98 |
|
99 |
# Update prompt placeholder based on last selected LoRA
|
|
|
129 |
lora_image_2 = None
|
130 |
if len(selected_indices) >= 1:
|
131 |
lora1 = loras[selected_indices[0]]
|
132 |
+
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}]({lora1['repo']}) ✨"
|
133 |
lora_image_1 = lora1['image']
|
134 |
if len(selected_indices) >= 2:
|
135 |
lora2 = loras[selected_indices[1]]
|
136 |
+
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}]({lora2['repo']}) ✨"
|
137 |
lora_image_2 = lora2['image']
|
138 |
return selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2
|
139 |
|
|
|
150 |
lora_image_2 = None
|
151 |
if len(selected_indices) >= 1:
|
152 |
lora1 = loras[selected_indices[0]]
|
153 |
+
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}]({lora1['repo']}) ✨"
|
154 |
lora_image_1 = lora1['image']
|
155 |
if len(selected_indices) >= 2:
|
156 |
lora2 = loras[selected_indices[1]]
|
157 |
+
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}]({lora2['repo']}) ✨"
|
158 |
lora_image_2 = lora2['image']
|
159 |
return selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2
|
160 |
|
|
|
164 |
selected_indices = random.sample(range(len(loras)), 2)
|
165 |
lora1 = loras[selected_indices[0]]
|
166 |
lora2 = loras[selected_indices[1]]
|
167 |
+
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}]({lora1['repo']}) ✨"
|
168 |
+
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}]({lora2['repo']}) ✨"
|
169 |
lora_scale_1 = 0.95
|
170 |
lora_scale_2 = 0.95
|
171 |
lora_image_1 = lora1['image']
|
|
|
174 |
|
175 |
@spaces.GPU(duration=70)
|
176 |
def generate_image(prompt_mash, steps, seed, cfg_scale, width, height, progress):
|
177 |
+
print("Generating image...")
|
178 |
pipe.to("cuda")
|
179 |
generator = torch.Generator(device="cuda").manual_seed(seed)
|
180 |
with calculateDuration("Generating image"):
|
|
|
209 |
joint_attention_kwargs={"scale": 1.0},
|
210 |
output_type="pil",
|
211 |
).images[0]
|
212 |
+
return final_image
|
213 |
+
|
214 |
def run_lora(prompt, image_input, image_strength, cfg_scale, steps, selected_indices, lora_scale_1, lora_scale_2, randomize_seed, seed, width, height, progress=gr.Progress(track_tqdm=True)):
|
215 |
if not selected_indices:
|
216 |
raise gr.Error("You must select at least one LoRA before proceeding.")
|
|
|
236 |
|
237 |
# Load LoRA weights with respective scales
|
238 |
lora_names = []
|
239 |
+
lora_weights = []
|
240 |
with calculateDuration("Loading LoRA weights"):
|
241 |
for idx, lora in enumerate(selected_loras):
|
242 |
lora_name = f"lora_{idx}"
|
243 |
lora_names.append(lora_name)
|
244 |
+
lora_weights.append(lora_scale_1 if idx == 0 else lora_scale_2)
|
245 |
lora_path = lora['repo']
|
246 |
+
weight_name = lora.get("weights")
|
247 |
if image_input is not None:
|
248 |
+
if weight_name:
|
249 |
+
pipe_i2i.load_lora_weights(lora_path, weight_name=weight_name, low_cpu_mem_usage=True, adapter_name=lora_name)
|
250 |
else:
|
251 |
pipe_i2i.load_lora_weights(lora_path, low_cpu_mem_usage=True, adapter_name=lora_name)
|
252 |
else:
|
253 |
+
if weight_name:
|
254 |
+
pipe.load_lora_weights(lora_path, weight_name=weight_name, low_cpu_mem_usage=True, adapter_name=lora_name)
|
255 |
else:
|
256 |
pipe.load_lora_weights(lora_path, low_cpu_mem_usage=True, adapter_name=lora_name)
|
257 |
+
print("Loaded LoRAs:", lora_names)
|
258 |
if image_input is not None:
|
259 |
+
pipe_i2i.set_adapters(lora_names, adapter_weights=lora_weights)
|
260 |
else:
|
261 |
+
pipe.set_adapters(lora_names, adapter_weights=lora_weights)
|
262 |
# Set random seed for reproducibility
|
263 |
with calculateDuration("Randomizing seed"):
|
264 |
if randomize_seed:
|
|
|
274 |
final_image = None
|
275 |
step_counter = 0
|
276 |
for image in image_generator:
|
277 |
+
step_counter += 1
|
278 |
final_image = image
|
279 |
progress_bar = f'<div class="progress-container"><div class="progress-bar" style="--current: {step_counter}; --total: {steps};"></div></div>'
|
280 |
yield image, seed, gr.update(value=progress_bar, visible=True)
|
|
|
285 |
if len(split_link) == 2:
|
286 |
model_card = ModelCard.load(link)
|
287 |
base_model = model_card.data.get("base_model")
|
288 |
+
print(f"Base model: {base_model}")
|
289 |
if base_model not in ["black-forest-labs/FLUX.1-dev", "black-forest-labs/FLUX.1-schnell"]:
|
290 |
raise Exception("Not a FLUX LoRA!")
|
291 |
image_path = model_card.data.get("widget", [{}])[0].get("output", {}).get("url", None)
|
|
|
307 |
if not safetensors_name:
|
308 |
raise Exception("No *.safetensors file found in the repository")
|
309 |
return split_link[1], link, safetensors_name, trigger_word, image_url
|
310 |
+
else:
|
311 |
+
raise Exception("Invalid Hugging Face repository link")
|
312 |
|
313 |
def check_custom_model(link):
|
314 |
+
if link.endswith(".safetensors"):
|
315 |
+
# Treat as direct link to the LoRA weights
|
316 |
+
title = os.path.basename(link)
|
317 |
+
repo = link
|
318 |
+
path = None # No specific weight name
|
319 |
+
trigger_word = ""
|
320 |
+
image_url = None
|
321 |
+
return title, repo, path, trigger_word, image_url
|
322 |
+
elif link.startswith("https://"):
|
323 |
+
if "huggingface.co" in link:
|
324 |
link_split = link.split("huggingface.co/")
|
325 |
return get_huggingface_safetensors(link_split[1])
|
326 |
+
else:
|
327 |
+
raise Exception("Unsupported URL")
|
328 |
+
else:
|
329 |
+
# Assume it's a Hugging Face model path
|
330 |
return get_huggingface_safetensors(link)
|
331 |
|
332 |
def add_custom_lora(custom_lora, selected_indices):
|
|
|
335 |
try:
|
336 |
title, repo, path, trigger_word, image = check_custom_model(custom_lora)
|
337 |
print(f"Loaded custom LoRA: {repo}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
338 |
existing_item_index = next((index for (index, item) in enumerate(loras) if item['repo'] == repo), None)
|
339 |
if existing_item_index is None:
|
340 |
new_item = {
|
|
|
344 |
"weights": path,
|
345 |
"trigger_word": trigger_word
|
346 |
}
|
347 |
+
print(f"New LoRA: {new_item}")
|
348 |
existing_item_index = len(loras)
|
349 |
loras.append(new_item)
|
350 |
|
|
|
353 |
# Update selected_indices if there's room
|
354 |
if len(selected_indices) < 2:
|
355 |
selected_indices.append(existing_item_index)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
356 |
else:
|
357 |
+
gr.Warning("You can select up to 2 LoRAs, remove one to select a new one.")
|
358 |
+
|
359 |
+
# Update selected_info and images
|
360 |
+
selected_info_1 = "Select a LoRA 1"
|
361 |
+
selected_info_2 = "Select a LoRA 2"
|
362 |
+
lora_scale_1 = 0.95
|
363 |
+
lora_scale_2 = 0.95
|
364 |
+
lora_image_1 = None
|
365 |
+
lora_image_2 = None
|
366 |
+
if len(selected_indices) >= 1:
|
367 |
+
lora1 = loras[selected_indices[0]]
|
368 |
+
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}]({lora1['repo']}) ✨"
|
369 |
+
lora_image_1 = lora1['image']
|
370 |
+
if len(selected_indices) >= 2:
|
371 |
+
lora2 = loras[selected_indices[1]]
|
372 |
+
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}]({lora2['repo']}) ✨"
|
373 |
+
lora_image_2 = lora2['image']
|
374 |
+
return (
|
375 |
+
gr.update(value=gallery_items),
|
376 |
+
selected_info_1,
|
377 |
+
selected_info_2,
|
378 |
+
selected_indices,
|
379 |
+
lora_scale_1,
|
380 |
+
lora_scale_2,
|
381 |
+
lora_image_1,
|
382 |
+
lora_image_2
|
383 |
+
)
|
384 |
except Exception as e:
|
385 |
print(e)
|
386 |
+
gr.Error(str(e))
|
387 |
+
return gr.NoChange(), gr.NoChange(), gr.NoChange(), selected_indices, gr.NoChange(), gr.NoChange(), gr.NoChange(), gr.NoChange()
|
388 |
else:
|
389 |
+
return gr.NoChange(), gr.NoChange(), gr.NoChange(), selected_indices, gr.NoChange(), gr.NoChange(), gr.NoChange(), gr.NoChange()
|
390 |
|
391 |
+
def remove_custom_lora(selected_indices):
|
392 |
global loras
|
393 |
if loras:
|
394 |
custom_lora_repo = loras[-1]['repo']
|
|
|
401 |
# Update gallery
|
402 |
gallery_items = [(item["image"], item["title"]) for item in loras]
|
403 |
# Update selected_info and images
|
404 |
+
selected_info_1 = "Select a LoRA 1"
|
405 |
+
selected_info_2 = "Select a LoRA 2"
|
406 |
lora_scale_1 = 0.95
|
407 |
lora_scale_2 = 0.95
|
408 |
lora_image_1 = None
|
409 |
lora_image_2 = None
|
410 |
if len(selected_indices) >= 1:
|
411 |
lora1 = loras[selected_indices[0]]
|
412 |
+
selected_info_1 = f"### LoRA 1 Selected: [{lora1['title']}]({lora1['repo']}) ✨"
|
413 |
lora_image_1 = lora1['image']
|
414 |
if len(selected_indices) >= 2:
|
415 |
lora2 = loras[selected_indices[1]]
|
416 |
+
selected_info_2 = f"### LoRA 2 Selected: [{lora2['title']}]({lora2['repo']}) ✨"
|
417 |
lora_image_2 = lora2['image']
|
418 |
+
return (
|
419 |
+
gr.update(value=gallery_items),
|
420 |
+
selected_info_1,
|
421 |
+
selected_info_2,
|
422 |
+
selected_indices,
|
423 |
+
lora_scale_1,
|
424 |
+
lora_scale_2,
|
425 |
+
lora_image_1,
|
426 |
+
lora_image_2
|
427 |
+
)
|
428 |
|
429 |
run_lora.zerogpu = True
|
430 |
|
|
|
433 |
#title{text-align: center}
|
434 |
#title h1{font-size: 3em; display:inline-flex; align-items:center}
|
435 |
#title img{width: 100px; margin-right: 0.5em}
|
436 |
+
#gallery{height: 260px}
|
437 |
#gallery .grid-wrap{height: 5vh}
|
438 |
#lora_list{background: var(--block-background-fill);padding: 0 1em .3em; font-size: 90%}
|
439 |
.custom_lora_card{margin-bottom: 1em}
|
|
|
484 |
remove_button_2 = gr.Button("Remove", size="sm")
|
485 |
with gr.Row():
|
486 |
with gr.Column():
|
487 |
+
with gr.Group():
|
488 |
+
custom_lora = gr.Textbox(label="Custom LoRA", info="LoRA Hugging Face path or *.safetensors public URL", placeholder="multimodalart/vintage-ads-flux")
|
489 |
+
add_custom_lora_button = gr.Button("Add Custom LoRA")
|
490 |
+
remove_custom_lora_button = gr.Button("Remove Custom LoRA")
|
491 |
+
gr.Markdown("[Check the list of FLUX LoRAs](https://huggingface.co/models?other=base_model:adapter:black-forest-labs/FLUX.1-dev)", elem_id="lora_list")
|
492 |
gallery = gr.Gallery(
|
493 |
[(item["image"], item["title"]) for item in loras],
|
494 |
label="LoRA Gallery",
|
|
|
496 |
columns=5,
|
497 |
elem_id="gallery"
|
498 |
)
|
|
|
|
|
|
|
|
|
|
|
499 |
with gr.Column():
|
500 |
progress_bar = gr.Markdown(elem_id="progress", visible=False)
|
501 |
result = gr.Image(label="Generated Image")
|
|
|
508 |
with gr.Row():
|
509 |
cfg_scale = gr.Slider(label="CFG Scale", minimum=1, maximum=20, step=0.5, value=3.5)
|
510 |
steps = gr.Slider(label="Steps", minimum=1, maximum=50, step=1, value=28)
|
511 |
+
|
512 |
with gr.Row():
|
513 |
width = gr.Slider(label="Width", minimum=256, maximum=1536, step=64, value=1024)
|
514 |
height = gr.Slider(label="Height", minimum=256, maximum=1536, step=64, value=1024)
|
515 |
+
|
516 |
with gr.Row():
|
517 |
randomize_seed = gr.Checkbox(True, label="Randomize seed")
|
518 |
seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0, randomize=True)
|
519 |
+
|
520 |
gallery.select(
|
521 |
update_selection,
|
522 |
inputs=[selected_indices, width, height],
|
|
|
537 |
inputs=[selected_indices],
|
538 |
outputs=[selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2]
|
539 |
)
|
540 |
+
add_custom_lora_button.click(
|
541 |
add_custom_lora,
|
542 |
inputs=[custom_lora, selected_indices],
|
543 |
+
outputs=[gallery, selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2]
|
544 |
)
|
545 |
+
remove_custom_lora_button.click(
|
546 |
remove_custom_lora,
|
547 |
+
inputs=[selected_indices],
|
548 |
+
outputs=[gallery, selected_info_1, selected_info_2, selected_indices, lora_scale_1, lora_scale_2, lora_image_1, lora_image_2]
|
549 |
)
|
550 |
gr.on(
|
551 |
triggers=[generate_button.click, prompt.submit],
|