Spaces:
Running
Running
Update media.py
Browse files
media.py
CHANGED
@@ -1,111 +1,159 @@
|
|
|
|
|
|
1 |
# --- LIBRARIES ---
|
2 |
import torch
|
3 |
import gradio as gr
|
4 |
import random
|
5 |
import time
|
6 |
-
from diffusers import AutoPipelineForText2Image, TextToVideoSDPipeline
|
7 |
import gc
|
8 |
import os
|
9 |
import imageio
|
|
|
|
|
|
|
|
|
10 |
|
11 |
-
# ---
|
12 |
-
|
13 |
-
|
14 |
-
device = "cuda"
|
15 |
-
torch_dtype = torch.float16 # Use float16 for GPU
|
16 |
-
print("✅ GPU detected. Using CUDA.")
|
17 |
-
else:
|
18 |
-
device = "cpu"
|
19 |
-
torch_dtype = torch.float32 # Use float32 for CPU
|
20 |
-
print("⚠️ No GPU detected. Using CPU. Performance will be slower.")
|
21 |
|
|
|
|
|
|
|
|
|
22 |
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
HF_TOKEN = os.environ.get('HF_TOKEN')
|
27 |
-
if HF_TOKEN:
|
28 |
login(token=HF_TOKEN)
|
29 |
print("✅ Hugging Face Authentication successful.")
|
30 |
-
|
31 |
-
print("
|
32 |
-
|
33 |
-
print("
|
34 |
|
35 |
# --- CONFIGURATION & STATE ---
|
36 |
available_models = {
|
37 |
"Fast Image (SDXL Turbo)": "stabilityai/sdxl-turbo",
|
38 |
"Quality Image (SDXL)": "stabilityai/stable-diffusion-xl-base-1.0",
|
|
|
39 |
"Video (Damo-Vilab)": "damo-vilab/text-to-video-ms-1.7b"
|
40 |
}
|
41 |
model_state = { "current_pipe": None, "loaded_model_name": None }
|
42 |
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
if model_state.get("loaded_model_name") != model_key:
|
47 |
-
|
48 |
-
yield {status_textbox: f"Unloading previous model..."}
|
49 |
if model_state.get("current_pipe"):
|
50 |
-
del model_state["current_pipe"]
|
51 |
-
gc.collect()
|
52 |
-
if device == "cuda":
|
53 |
-
torch.cuda.empty_cache()
|
54 |
-
|
55 |
model_id = available_models[model_key]
|
56 |
-
|
57 |
-
yield {status_textbox: f"Loading {model_id}... This can take a minute."}
|
58 |
-
|
59 |
-
# Adapt model loading based on hardware
|
60 |
-
if "Image" in model_key:
|
61 |
-
pipe = AutoPipelineForText2Image.from_pretrained(model_id, torch_dtype=torch_dtype, variant="fp16" if device == "cuda" else "fp32")
|
62 |
-
elif "Video" in model_key:
|
63 |
pipe = TextToVideoSDPipeline.from_pretrained(model_id, torch_dtype=torch_dtype)
|
64 |
-
|
65 |
-
|
|
|
|
|
66 |
pipe.to(device)
|
67 |
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
model_state["current_pipe"] = pipe
|
73 |
model_state["loaded_model_name"] = model_key
|
74 |
-
print(f"✅ Model loaded
|
75 |
|
76 |
pipe = model_state["current_pipe"]
|
77 |
generator = torch.Generator(device).manual_seed(seed)
|
78 |
-
|
79 |
-
|
80 |
-
if "
|
81 |
-
|
82 |
-
|
83 |
-
num_steps, guidance_scale = 1, 0.0
|
84 |
-
else:
|
85 |
-
num_steps, guidance_scale = int(steps), float(cfg_scale)
|
86 |
-
|
87 |
-
image = pipe(
|
88 |
-
prompt=prompt, negative_prompt=negative_prompt, num_inference_steps=num_steps,
|
89 |
-
guidance_scale=guidance_scale, width=int(width), height=int(height), generator=generator
|
90 |
-
).images[0]
|
91 |
-
print("✅ Image generation complete.")
|
92 |
-
yield {output_image: image, output_video: None, status_textbox: f"Seed used: {seed}"}
|
93 |
-
|
94 |
-
elif "Video" in model_key:
|
95 |
-
print("Generating video...")
|
96 |
video_frames = pipe(prompt=prompt, num_inference_steps=int(steps), height=320, width=576, num_frames=int(num_frames), generator=generator).frames
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
97 |
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
103 |
|
104 |
-
# --- GRADIO
|
105 |
with gr.Blocks(theme='gradio/soft') as demo:
|
|
|
106 |
gr.Markdown("# The Generative Media Suite")
|
107 |
-
|
108 |
-
gr.Markdown("Create fast images, high-quality images, or short videos. Created by cheeseman182.")
|
109 |
seed_state = gr.State(-1)
|
110 |
with gr.Row():
|
111 |
with gr.Column(scale=2):
|
@@ -125,6 +173,7 @@ with gr.Blocks(theme='gradio/soft') as demo:
|
|
125 |
output_image = gr.Image(label="Image Result", interactive=False, height="60vh", visible=True)
|
126 |
output_video = gr.Video(label="Video Result", interactive=False, height="60vh", visible=False)
|
127 |
status_textbox = gr.Textbox(label="Status", interactive=False)
|
|
|
128 |
def update_ui_on_model_change(model_key):
|
129 |
is_video = "Video" in model_key
|
130 |
is_turbo = "Turbo" in model_key
|
@@ -138,13 +187,14 @@ with gr.Blocks(theme='gradio/soft') as demo:
|
|
138 |
output_video: gr.update(visible=is_video)
|
139 |
}
|
140 |
model_selector.change(update_ui_on_model_change, model_selector, [steps_slider, cfg_slider, width_slider, height_slider, num_frames_slider, output_image, output_video])
|
|
|
141 |
click_event = generate_button.click(
|
142 |
fn=lambda s: (s if s != -1 else random.randint(0, 2**32 - 1)),
|
143 |
inputs=seed_input,
|
144 |
outputs=seed_state,
|
145 |
queue=False
|
146 |
).then(
|
147 |
-
fn=
|
148 |
inputs=[model_selector, prompt_input, negative_prompt_input, steps_slider, cfg_slider, width_slider, height_slider, seed_state, num_frames_slider],
|
149 |
outputs=[output_image, output_video, status_textbox]
|
150 |
)
|
|
|
1 |
+
# --- START OF FILE media.py (FINAL WITH LIVE PROGRESS) ---
|
2 |
+
|
3 |
# --- LIBRARIES ---
|
4 |
import torch
|
5 |
import gradio as gr
|
6 |
import random
|
7 |
import time
|
8 |
+
from diffusers import AutoPipelineForText2Image, TextToVideoSDPipeline, EulerAncestralDiscreteScheduler
|
9 |
import gc
|
10 |
import os
|
11 |
import imageio
|
12 |
+
import numpy as np
|
13 |
+
import threading
|
14 |
+
from queue import Queue, Empty as QueueEmpty
|
15 |
+
from PIL import Image
|
16 |
|
17 |
+
# --- SECURE AUTHENTICATION FOR HUGGING FACE SPACES ---
|
18 |
+
import os
|
19 |
+
from huggingface_hub import login
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
+
# This code will attempt to read the HF_TOKEN from the Space's secrets.
|
22 |
+
# On your local machine, this will do nothing unless you set it up, which isn't necessary.
|
23 |
+
# On the Hugging Face server, it will find the secret you just saved.
|
24 |
+
HF_TOKEN = os.environ.get('HF_TOKEN')
|
25 |
|
26 |
+
if HF_TOKEN:
|
27 |
+
print("✅ Found HF_TOKEN secret. Logging in...")
|
28 |
+
try:
|
|
|
|
|
29 |
login(token=HF_TOKEN)
|
30 |
print("✅ Hugging Face Authentication successful.")
|
31 |
+
except Exception as e:
|
32 |
+
print(f"❌ Hugging Face login failed: {e}")
|
33 |
+
else:
|
34 |
+
print("⚠️ No HF_TOKEN secret found. Gated models may not be available on the deployed app.")
|
35 |
|
36 |
# --- CONFIGURATION & STATE ---
|
37 |
available_models = {
|
38 |
"Fast Image (SDXL Turbo)": "stabilityai/sdxl-turbo",
|
39 |
"Quality Image (SDXL)": "stabilityai/stable-diffusion-xl-base-1.0",
|
40 |
+
"Photorealism (Juggernaut)": "RunDiffusion/Juggernaut-XL-v9",
|
41 |
"Video (Damo-Vilab)": "damo-vilab/text-to-video-ms-1.7b"
|
42 |
}
|
43 |
model_state = { "current_pipe": None, "loaded_model_name": None }
|
44 |
|
45 |
+
# --- THE FINAL GENERATION FUNCTION WITH LIVE PROGRESS ---
|
46 |
+
def generate_media_live_progress(model_key, prompt, negative_prompt, steps, cfg_scale, width, height, seed, num_frames):
|
47 |
+
# --- Model Loading (Unchanged) ---
|
48 |
if model_state.get("loaded_model_name") != model_key:
|
49 |
+
yield {output_image: None, output_video: None, status_textbox: f"Loading {model_key}..."}
|
|
|
50 |
if model_state.get("current_pipe"):
|
51 |
+
del model_state["current_pipe"]; gc.collect(); torch.cuda.empty_cache()
|
|
|
|
|
|
|
|
|
52 |
model_id = available_models[model_key]
|
53 |
+
if "Video" in model_key:
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
pipe = TextToVideoSDPipeline.from_pretrained(model_id, torch_dtype=torch_dtype)
|
55 |
+
else:
|
56 |
+
pipe = AutoPipelineForText2Image.from_pretrained(model_id, torch_dtype=torch_dtype, variant="fp16")
|
57 |
+
|
58 |
+
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
|
59 |
pipe.to(device)
|
60 |
|
61 |
+
if device == "cuda":
|
62 |
+
if "Video" not in model_key: pipe.enable_model_cpu_offload()
|
63 |
+
pipe.enable_vae_slicing()
|
|
|
64 |
model_state["current_pipe"] = pipe
|
65 |
model_state["loaded_model_name"] = model_key
|
66 |
+
print(f"✅ Model loaded on {device.upper()}.")
|
67 |
|
68 |
pipe = model_state["current_pipe"]
|
69 |
generator = torch.Generator(device).manual_seed(seed)
|
70 |
+
|
71 |
+
# --- Generation Logic ---
|
72 |
+
if "Video" in model_key:
|
73 |
+
# For video, we'll keep the simple status updates for now
|
74 |
+
yield {output_image: None, output_video: None, status_textbox: "Generating video..."}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
video_frames = pipe(prompt=prompt, num_inference_steps=int(steps), height=320, width=576, num_frames=int(num_frames), generator=generator).frames
|
76 |
+
video_frames_5d = np.array(video_frames)
|
77 |
+
video_frames_4d = np.squeeze(video_frames_5d)
|
78 |
+
video_uint8 = (video_frames_4d * 255).astype(np.uint8)
|
79 |
+
list_of_frames = [frame for frame in video_uint8]
|
80 |
+
video_path = f"video_{seed}.mp4"
|
81 |
+
imageio.mimsave(video_path, list_of_frames, fps=12)
|
82 |
+
yield {output_image: None, output_video: video_path, status_textbox: f"Video saved! Seed: {seed}"}
|
83 |
+
|
84 |
+
else: # Image Generation with Live Progress
|
85 |
+
progress_queue = Queue()
|
86 |
|
87 |
+
def run_pipe():
|
88 |
+
# This function runs in a separate thread
|
89 |
+
start_time = time.time()
|
90 |
+
|
91 |
+
def progress_callback(pipe, step, timestep, callback_kwargs):
|
92 |
+
# This is called by the pipeline at each step
|
93 |
+
elapsed_time = time.time() - start_time
|
94 |
+
# Avoid division by zero on the first step
|
95 |
+
if elapsed_time > 0:
|
96 |
+
its_per_sec = (step + 1) / elapsed_time
|
97 |
+
progress_queue.put((step + 1, its_per_sec))
|
98 |
+
return callback_kwargs
|
99 |
+
|
100 |
+
try:
|
101 |
+
# The final image is still generated using the pipeline's high-quality VAE
|
102 |
+
final_image = pipe(
|
103 |
+
prompt=prompt, negative_prompt=negative_prompt, num_inference_steps=int(steps),
|
104 |
+
guidance_scale=float(cfg_scale), width=int(width), height=int(height),
|
105 |
+
generator=generator,
|
106 |
+
callback_on_step_end=progress_callback
|
107 |
+
).images[0]
|
108 |
+
progress_queue.put(final_image) # Put the final result on the queue
|
109 |
+
except Exception as e:
|
110 |
+
print(f"An error occurred in the generation thread: {e}")
|
111 |
+
progress_queue.put(None) # Signal an error
|
112 |
+
|
113 |
+
# Start the generation in the background
|
114 |
+
thread = threading.Thread(target=run_pipe)
|
115 |
+
thread.start()
|
116 |
+
|
117 |
+
# In the main thread, listen for updates from the queue and yield to Gradio
|
118 |
+
total_steps = int(steps)
|
119 |
+
yield {status_textbox: "Generating..."} # Initial status
|
120 |
+
|
121 |
+
while True:
|
122 |
+
try:
|
123 |
+
update = progress_queue.get(timeout=1.0) # Wait for an update
|
124 |
+
|
125 |
+
if isinstance(update, Image.Image): # It's the final image
|
126 |
+
yield {output_image: update, status_textbox: f"Generation complete! Seed: {seed}"}
|
127 |
+
break
|
128 |
+
elif isinstance(update, tuple): # It's a progress update (step, speed)
|
129 |
+
current_step, its_per_sec = update
|
130 |
+
progress_percent = (current_step / total_steps) * 100
|
131 |
+
steps_remaining = total_steps - current_step
|
132 |
+
eta_seconds = steps_remaining / its_per_sec if its_per_sec > 0 else 0
|
133 |
+
eta_minutes, eta_seconds_rem = divmod(int(eta_seconds), 60)
|
134 |
+
|
135 |
+
status_text = (
|
136 |
+
f"Generating... {progress_percent:.0f}% ({current_step}/{total_steps}) | "
|
137 |
+
f"{its_per_sec:.2f}it/s | "
|
138 |
+
f"ETA: {eta_minutes:02d}:{eta_seconds_rem:02d}"
|
139 |
+
)
|
140 |
+
yield {status_textbox: status_text}
|
141 |
+
elif update is None: # An error occurred
|
142 |
+
yield {status_textbox: "Error during generation. Check console."}
|
143 |
+
break
|
144 |
+
except QueueEmpty:
|
145 |
+
if not thread.is_alive():
|
146 |
+
print("⚠️ Generation thread finished unexpectedly.")
|
147 |
+
yield {status_textbox: "Generation failed. Check console for details."}
|
148 |
+
break
|
149 |
+
|
150 |
+
thread.join()
|
151 |
|
152 |
+
# --- GRADIO UI ---
|
153 |
with gr.Blocks(theme='gradio/soft') as demo:
|
154 |
+
# (UI layout is the same, just point to the new function)
|
155 |
gr.Markdown("# The Generative Media Suite")
|
156 |
+
gr.Markdown("Create fast images, high-quality images, or short videos. Created by cheeseman182. (note: the speed on the status bar is wrong)")
|
|
|
157 |
seed_state = gr.State(-1)
|
158 |
with gr.Row():
|
159 |
with gr.Column(scale=2):
|
|
|
173 |
output_image = gr.Image(label="Image Result", interactive=False, height="60vh", visible=True)
|
174 |
output_video = gr.Video(label="Video Result", interactive=False, height="60vh", visible=False)
|
175 |
status_textbox = gr.Textbox(label="Status", interactive=False)
|
176 |
+
|
177 |
def update_ui_on_model_change(model_key):
|
178 |
is_video = "Video" in model_key
|
179 |
is_turbo = "Turbo" in model_key
|
|
|
187 |
output_video: gr.update(visible=is_video)
|
188 |
}
|
189 |
model_selector.change(update_ui_on_model_change, model_selector, [steps_slider, cfg_slider, width_slider, height_slider, num_frames_slider, output_image, output_video])
|
190 |
+
|
191 |
click_event = generate_button.click(
|
192 |
fn=lambda s: (s if s != -1 else random.randint(0, 2**32 - 1)),
|
193 |
inputs=seed_input,
|
194 |
outputs=seed_state,
|
195 |
queue=False
|
196 |
).then(
|
197 |
+
fn=generate_media_live_progress, # Use the new function with progress
|
198 |
inputs=[model_selector, prompt_input, negative_prompt_input, steps_slider, cfg_slider, width_slider, height_slider, seed_state, num_frames_slider],
|
199 |
outputs=[output_image, output_video, status_textbox]
|
200 |
)
|