File size: 10,676 Bytes
b1828f2 824b515 fe91ef5 824b515 5098655 824b515 b1828f2 fe91ef5 824b515 fe91ef5 f4723f0 fe91ef5 5098655 824b515 a48ca5b 5098655 824b515 5098655 824b515 5098655 824b515 5098655 824b515 5098655 824b515 fe91ef5 824b515 fe91ef5 824b515 fe91ef5 20a2d44 fe91ef5 824b515 fe91ef5 824b515 fe91ef5 20a2d44 fe91ef5 d4a2e0d fe91ef5 5098655 fe91ef5 20a2d44 fe91ef5 824b515 fe91ef5 20a2d44 5098655 824b515 2ad9d00 824b515 2ad9d00 824b515 2ad9d00 824b515 5098655 824b515 fe91ef5 5098655 fe91ef5 5098655 fe91ef5 2846498 fe91ef5 824b515 fe91ef5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 |
import gradio as gr
import os
import spaces
import sys
from copy import deepcopy
sys.path.append('./VADER-VideoCrafter/scripts/main')
sys.path.append('./VADER-VideoCrafter/scripts')
sys.path.append('./VADER-VideoCrafter')
from train_t2v_lora import main_fn, setup_model
examples = [
["Fairy and Magical Flowers: A fairy tends to enchanted, glowing flowers.", 'huggingface-hps-aesthetic',
8, 901, 384, 512, 12.0, 25, 1.0, 24, 10],
["A cat playing an electric guitar in a loft with industrial-style decor and soft, multicolored lights.",
'huggingface-hps-aesthetic', 8, 208, 384, 512, 12.0, 25, 1.0, 24, 10],
["A raccoon playing a guitar under a blossoming cherry tree.",
'huggingface-hps-aesthetic', 8, 180, 384, 512, 12.0, 25, 1.0, 24, 10],
["A raccoon playing an electric bass in a garage band setting.",
'huggingface-hps-aesthetic', 8, 400, 384, 512, 12.0, 25, 1.0, 24, 10],
["A talking bird with shimmering feathers and a melodious voice finds a legendary treasure, guiding through enchanted forests, ancient ruins, and mystical challenges.",
"huggingface-pickscore", 16, 200, 384, 512, 12.0, 25, 1.0, 24, 10],
["A snow princess stands on the balcony of her ice castle, her hair adorned with delicate snowflakes, overlooking her serene realm.",
"huggingface-pickscore", 16, 400, 384, 512, 12.0, 25, 1.0, 24, 10],
["A mermaid with flowing hair and a shimmering tail discovers a hidden underwater kingdom adorned with coral palaces, glowing pearls, and schools of colorful fish, encountering both wonders and dangers along the way.",
"huggingface-pickscore", 16, 800, 384, 512, 12.0, 25, 1.0, 24, 10],
]
model = setup_model()
@spaces.GPU(duration=180)
def gradio_main_fn(prompt, lora_model, lora_rank, seed, height, width, unconditional_guidance_scale, ddim_steps, ddim_eta,
frames, savefps):
global model
if model is None:
return "Model is not loaded. Please load the model first."
video_path = main_fn(prompt=prompt,
lora_model=lora_model,
lora_rank=int(lora_rank),
seed=int(seed),
height=int(height),
width=int(width),
unconditional_guidance_scale=float(unconditional_guidance_scale),
ddim_steps=int(ddim_steps),
ddim_eta=float(ddim_eta),
frames=int(frames),
savefps=int(savefps),
model=deepcopy(model))
return video_path
def reset_fn():
return ("A mermaid with flowing hair and a shimmering tail discovers a hidden underwater kingdom adorned with coral palaces, glowing pearls, and schools of colorful fish, encountering both wonders and dangers along the way.",
200, 384, 512, 12.0, 25, 1.0, 24, 16, 10, "huggingface-pickscore")
def update_lora_rank(lora_model):
if lora_model == "huggingface-pickscore":
return gr.update(value=16)
elif lora_model == "huggingface-hps-aesthetic":
return gr.update(value=8)
else: # "Base Model"
return gr.update(value=8)
def update_dropdown(lora_rank):
if lora_rank == 16:
return gr.update(value="huggingface-pickscore")
elif lora_rank == 8:
return gr.update(value="huggingface-hps-aesthetic")
else: # 0
return gr.update(value="Base Model")
custom_css = """
#centered {
display: flex;
justify-content: center;
width: 60%;
margin: 0 auto;
}
.column-centered {
display: flex;
flex-direction: column;
align-items: center;
width: 60%;
}
#image-upload {
flex-grow: 1;
}
#params .tabs {
display: flex;
flex-direction: column;
flex-grow: 1;
}
#params .tabitem[style="display: block;"] {
flex-grow: 1;
display: flex !important;
}
#params .gap {
flex-grow: 1;
}
#params .form {
flex-grow: 1 !important;
}
#params .form > :last-child{
flex-grow: 1;
}
"""
with gr.Blocks(css=custom_css) as demo:
with gr.Row():
with gr.Column():
gr.HTML(
"""
<h1 style='text-align: center; font-size: 3.2em; margin-bottom: 0.5em; font-family: Arial, sans-serif; margin: 20px;'>
Video Diffusion Alignment via Reward Gradient
</h1>
"""
)
gr.HTML(
"""
<style>
body {
font-family: Arial, sans-serif;
text-align: center;
margin: 50px;
}
a {
text-decoration: none !important;
color: black !important;
}
</style>
<body>
<div style="font-size: 1.4em; margin-bottom: 0.5em; ">
<a href="https://mihirp1998.github.io">Mihir Prabhudesai</a><sup>*</sup>
<a href="https://russellmendonca.github.io/">Russell Mendonca</a><sup>*</sup>
<a href="mailto: zheyangqin.qzy@gmail.com">Zheyang Qin</a><sup>*</sup>
<a href="https://www.cs.cmu.edu/~katef/">Katerina Fragkiadaki</a><sup></sup>
<a href="https://www.cs.cmu.edu/~dpathak/">Deepak Pathak</a><sup></sup>
</div>
<div style="font-size: 1.3em; font-style: italic;">
Carnegie Mellon University
</div>
</body>
"""
)
gr.HTML(
"""
<head>
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.0.0-beta3/css/all.min.css">
<style>
.button-container {
display: flex;
justify-content: center;
gap: 10px;
margin-top: 10px;
}
.button-container a {
display: inline-flex;
align-items: center;
padding: 10px 20px;
border-radius: 30px;
border: 1px solid #ccc;
text-decoration: none;
color: #333 !important;
font-size: 16px;
text-decoration: none !important;
}
.button-container a i {
margin-right: 8px;
}
</style>
</head>
<div class="button-container">
<a href="https://arxiv.org/abs/2407.08737" class="btn btn-outline-primary">
<i class="fa-solid fa-file-pdf"></i> Paper
</a>
<a href="https://vader-vid.github.io/" class="btn btn-outline-danger">
<i class="fa-solid fa-video"></i> Website
<a href="https://github.com/mihirp1998/VADER" class="btn btn-outline-secondary">
<i class="fa-brands fa-github"></i> Code
</a>
</div>
"""
)
with gr.Row(elem_id="centered"):
with gr.Column(elem_id="params"):
lora_model = gr.Dropdown(
label="VADER Model",
choices=["huggingface-pickscore", "huggingface-hps-aesthetic"],
value="huggingface-pickscore"
)
lora_rank = gr.Slider(minimum=8, maximum=16, label="LoRA Rank", step = 8, value=16)
prompt = gr.Textbox(placeholder="Enter prompt text here", lines=4, label="Text Prompt",
value="A mermaid with flowing hair and a shimmering tail discovers a hidden underwater kingdom adorned with coral palaces, glowing pearls, and schools of colorful fish, encountering both wonders and dangers along the way.")
run_btn = gr.Button("Run Inference")
with gr.Column():
output_video = gr.Video(elem_id="image-upload")
with gr.Row(elem_id="centered"):
with gr.Column():
seed = gr.Slider(minimum=0, maximum=65536, label="Seed", step = 1, value=200)
with gr.Row():
height = gr.Slider(minimum=0, maximum=512, label="Height", step = 16, value=384)
width = gr.Slider(minimum=0, maximum=512, label="Width", step = 16, value=512)
with gr.Row():
frames = gr.Slider(minimum=0, maximum=50, label="Frames", step = 1, value=24)
savefps = gr.Slider(minimum=0, maximum=30, label="Save FPS", step = 1, value=10)
with gr.Row():
DDIM_Steps = gr.Slider(minimum=0, maximum=50, label="DDIM Steps", step = 1, value=25)
unconditional_guidance_scale = gr.Slider(minimum=0, maximum=50, label="Guidance Scale", step = 0.1, value=12.0)
DDIM_Eta = gr.Slider(minimum=0, maximum=1, label="DDIM Eta", step = 0.01, value=1.0)
# reset button
reset_btn = gr.Button("Reset")
reset_btn.click(fn=reset_fn, outputs=[prompt, seed, height, width, unconditional_guidance_scale, DDIM_Steps, DDIM_Eta, frames, lora_rank, savefps, lora_model])
run_btn.click(fn=gradio_main_fn,
inputs=[prompt, lora_model, lora_rank,
seed, height, width, unconditional_guidance_scale,
DDIM_Steps, DDIM_Eta, frames, savefps],
outputs=output_video
)
lora_model.change(fn=update_lora_rank, inputs=lora_model, outputs=lora_rank)
lora_rank.change(fn=update_dropdown, inputs=lora_rank, outputs=lora_model)
gr.Examples(examples=examples,
inputs=[prompt, lora_model, lora_rank, seed,
height, width, unconditional_guidance_scale,
DDIM_Steps, DDIM_Eta, frames, savefps],
outputs=output_video,
fn=gradio_main_fn,
run_on_click=False,
cache_examples="lazy",
)
demo.launch(share=True) |