Spaces:
Sleeping
Sleeping
File size: 9,956 Bytes
40e68f7 6ce1706 40e68f7 6ce1706 40e68f7 6ce1706 40e68f7 168c086 40e68f7 168c086 40e68f7 168c086 40e68f7 168c086 40e68f7 168c086 40e68f7 168c086 40e68f7 168c086 40e68f7 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 |
# Adapted from https://github.com/guoyww/AnimateDiff/blob/main/app.py
import spaces
import os
import json
import torch
import random
import gradio as gr
from glob import glob
from omegaconf import OmegaConf
from datetime import datetime
from safetensors import safe_open
from PIL import Image
from unet2d_custom import UNet2DConditionModel
import torch
from pipeline_stable_diffusion_custom import StableDiffusionPipeline
from diffusers import DDIMScheduler
from pnp_utils import *
import torchvision.transforms as T
from preprocess import get_timesteps
from preprocess import Preprocess
from pnp import PNP
sample_idx = 0
css = """
.toolbutton {
margin-buttom: 0em 0em 0em 0em;
max-width: 1.5em;
min-width: 1.5em !important;
height: 1.5em;
}
"""
class AnimateController:
def __init__(self):
self.sr = 44100
self.save_steps = 50
self.device = 'cuda'
self.seed = 42
self.extract_reverse = False
self.save_dir = 'latents'
self.steps = 50
self.inversion_prompt = ''
self.seed = 42
seed_everything(self.seed)
self.pnp = PNP(sd_version="1.4")
self.pnp.unet.to(self.device)
self.pnp.audio_projector.to(self.device)
# audio_projector_path = "ckpts/audio_projector_landscape.pth"
# gate_dict_path = "ckpts/landscape.pt"
# self.pnp.set_audio_projector(gate_dict_path, audio_projector_path)
@spaces.GPU
def preprocess(self, image=None):
model_key = "CompVis/stable-diffusion-v1-4"
toy_scheduler = DDIMScheduler.from_pretrained(model_key, subfolder="scheduler")
toy_scheduler.set_timesteps(self.save_steps)
timesteps_to_save, num_inference_steps = get_timesteps(toy_scheduler, num_inference_steps=self.save_steps,
strength=1.0,
device=self.device)
save_path = os.path.join(self.save_dir + "_forward")
os.makedirs(save_path, exist_ok=True)
model = Preprocess(self.device, sd_version='1.4', hf_key=None)
recon_image = model.extract_latents(data_path=image,
num_steps=self.steps,
save_path=save_path,
timesteps_to_save=timesteps_to_save,
inversion_prompt=self.inversion_prompt,
extract_reverse=False)
T.ToPILImage()(recon_image[0]).save(os.path.join(save_path, f'recon.jpg'))
@spaces.GPU
def generate(self, file=None, audio=None, prompt=None,
cfg_scale=5, image_path=None,
pnp_f_t=0.8, pnp_attn_t=0.8,):
image = self.pnp.run_pnp(
n_timesteps=50,
pnp_f_t=pnp_f_t, pnp_attn_t=pnp_attn_t,
prompt=prompt,
negative_prompt="",
audio_path=audio,
image_path=image_path,
cfg_scale=cfg_scale,
)
return image
# @spaces.GPU
# def update_audio_model(self, audio_model_update):
# print(f"changing ckpts audio model {audio_model_update}")
# if audio_model_update == "Landscape Model":
# audio_projector_path = "ckpts/audio_projector_landscape.pth"
# gate_dict_path = "ckpts/landscape.pt"
# else:
# audio_projector_path = "ckpts/audio_projector_gh.pth"
# gate_dict_path = "ckpts/greatest_hits.pt"
# self.pnp.set_audio_projector(gate_dict_path, audio_projector_path)
# self.pnp.changed_model = True
# # gate_dict = torch.load(gate_dict_path)
# # for name, param in self.pnp.unet.named_parameters():
# # if "adapter" in name:
# # param.data = gate_dict[name]
# # self.pnp.audio_projector.load_state_dict(torch.load(audio_projector_path))
# # self.pnp.unet.to(self.device)
# # self.pnp.audio_projector.to(self.device)
# return gr.Dropdown()
controller = AnimateController()
def ui():
with gr.Blocks(css=css) as demo:
gr.Markdown(
"""
# [SonicDiffusion: Audio-Driven Image Generation and Editing with Pretrained Diffusion Models]
"""
)
with gr.Row():
audio_input = gr.Audio(sources="upload", type="filepath")
prompt_textbox = gr.Textbox(label="Prompt", lines=2)
with gr.Row():
with gr.Column():
pnp_f_t = gr.Slider(label="PNP Residual Injection", step=0.1, value=0.8, minimum=0.0, maximum=1.0)
pnp_attn_t = gr.Slider(label="PNP Attention Injection", step=0.1, value=0.8, minimum=0.0, maximum=1.0)
with gr.Column():
audio_model_dropdown = gr.Dropdown(
label="Select SonicDiffusion model",
value="Landscape Model",
choices=["Landscape Model", "Greatest Hits Model"],
interactive=True,
)
# audio_model_dropdown.change(fn=controller.update_audio_model, inputs=[audio_model_dropdown], outputs=[audio_model_dropdown])
cfg_scale_slider = gr.Slider(label="CFG Scale", step=0.5, value=7.5, minimum=0, maximum=20)
with gr.Row():
preprocess_button = gr.Button(value="Preprocess", variant='primary')
generate_button = gr.Button(value="Generate", variant='primary')
with gr.Row():
with gr.Column():
image_input = gr.Image(label="Input Image Component", sources="upload", type="filepath")
with gr.Column():
output = gr.Image(label="Output Image Component",
height=512, width=512)
# with gr.Row():
# examples_img_1 = [
# [Image.open("assets/corridor.png")],
# [Image.open("assets/desert.png")],
# [Image.open("assets/forest.png")],
# [Image.open("assets/forest_painting.png")],
# [Image.open("assets/golf_field.png")],
# [Image.open("assets/human.png")],
# [Image.open("assets/wood.png")],
# [Image.open("assets/house.png")],
# [Image.open("assets/apple.png")],
# [Image.open("assets/chair.png")],
# [Image.open("assets/hands.png")],
# [Image.open("assets/pineapple.png")],
# [Image.open("assets/table.png")],
# ]
# gr.Examples(examples=examples_img_1,inputs=[image_input], label="Images")
# # examples_img_2 = [
# # [Image.open("assets/apple.png")],
# # [Image.open("assets/chair.png")],
# # [Image.open("assets/hands.png")],
# # [Image.open("assets/pineapple.png")],
# # [Image.open("assets/table.png")],
# # ]
# # gr.Examples(examples=examples,inputs=[image_input], label="Greatest Hits Images")
# examples2 = [
# ['./assets/fire_crackling.wav'],
# ['./assets/forest_birds.wav'],
# ['./assets/forest_stepping_on_branches.wav'],
# ['./assets/howling_wind.wav'],
# ['./assets/rain.wav'],
# ['./assets/splashing_water.wav'],
# ['./assets/splashing_water_soft.wav'],
# ['./assets/steps_on_snow.wav'],
# ['./assets/thunder.wav'],
# ['./assets/underwater.wav'],
# ['./assets/waterfall_burble.wav'],
# ['./assets/wind_noise_birds.wav'],
# ]
# gr.Examples(examples=examples2,inputs=[audio_input], label="Landscape Audios")
# examples3 = [
# ['./assets/cardboard.wav'],
# ['./assets/carpet.wav'],
# ['./assets/ceramic.wav'],
# ['./assets/cloth.wav'],
# ['./assets/gravel.wav'],
# ['./assets/leaf.wav'],
# ['./assets/metal.wav'],
# ['./assets/plastic_bag.wav'],
# ['./assets/plastic.wav'],
# ['./assets/rock.wav'],
# ['./assets/wood.wav'],
# ]
# gr.Examples(examples=examples3,inputs=[audio_input], label="Greatest Hits Audios")
preprocess_button.click(
fn=controller.preprocess,
inputs=[
image_input
],
outputs=output
)
generate_button.click(
fn=controller.generate,
inputs=[
audio_model_dropdown,
audio_input,
prompt_textbox,
cfg_scale_slider,
image_input,
pnp_f_t,
pnp_attn_t,
],
outputs=output
)
return demo
if __name__ == "__main__":
demo = ui()
demo.launch(share=True)
|