import spaces import gradio as gr import os import sys import time from omegaconf import OmegaConf import torch from pytorch_lightning import seed_everything from huggingface_hub import hf_hub_download from einops import repeat import torchvision.transforms as transforms from utils.utils import instantiate_from_config sys.path.insert(0, "scripts/evaluation") from funcs import ( batch_ddim_sampling, load_model_checkpoint, get_latent_z, save_videos ) def download_model(): REPO_ID = 'Doubiiu/DynamiCrafter_512_Interp' filename_list = ['model.ckpt'] if not os.path.exists('./checkpoints/dynamicrafter_512_interp_v1/'): os.makedirs('./checkpoints/dynamicrafter_512_interp_v1/') for filename in filename_list: local_file = os.path.join('./checkpoints/dynamicrafter_512_interp_v1/', filename) if not os.path.exists(local_file): hf_hub_download(repo_id=REPO_ID, filename=filename, local_dir='./checkpoints/dynamicrafter_512_interp_v1/', force_download=True) download_model() ckpt_path='checkpoints/dynamicrafter_512_interp_v1/model.ckpt' config_file='configs/inference_512_v1.0.yaml' config = OmegaConf.load(config_file) model_config = config.pop("model", OmegaConf.create()) model_config['params']['unet_config']['params']['use_checkpoint']=False model = instantiate_from_config(model_config) assert os.path.exists(ckpt_path), "Error: checkpoint Not Found!" model = load_model_checkpoint(model, ckpt_path) model.eval() model = model.cuda() @spaces.GPU(duration=300) def infer(image, prompt, steps=50, cfg_scale=7.5, eta=1.0, fs=3, seed=123, image2=None): resolution = (320, 512) save_fps = 8 seed_everything(seed) transform = transforms.Compose([ transforms.Resize(min(resolution)), transforms.CenterCrop(resolution), ]) torch.cuda.empty_cache() print('start:', prompt, time.strftime('%Y-%m-%d %H:%M:%S',time.localtime(time.time()))) start = time.time() if steps > 60: steps = 60 batch_size=1 channels = model.model.diffusion_model.out_channels frames = model.temporal_length h, w = resolution[0] // 8, resolution[1] // 8 noise_shape = [batch_size, channels, frames, h, w] # text cond with torch.no_grad(), torch.cuda.amp.autocast(): text_emb = model.get_learned_conditioning([prompt]) # img cond img_tensor = torch.from_numpy(image).permute(2, 0, 1).float().to(model.device) img_tensor = (img_tensor / 255. - 0.5) * 2 image_tensor_resized = transform(img_tensor) #3,256,256 videos = image_tensor_resized.unsqueeze(0) # bchw z = get_latent_z(model, videos.unsqueeze(2)) #bc,1,hw if image2 is not None: img_tensor2 = torch.from_numpy(image2).permute(2, 0, 1).float().to(model.device) img_tensor2 = (img_tensor2 / 255. - 0.5) * 2 image_tensor_resized2 = transform(img_tensor2) #3,h,w videos2 = image_tensor_resized2.unsqueeze(0) # bchw z2 = get_latent_z(model, videos2.unsqueeze(2)) #bc,1,hw img_tensor_repeat = repeat(z, 'b c t h w -> b c (repeat t) h w', repeat=frames) img_tensor_repeat = torch.zeros_like(img_tensor_repeat) ## old img_tensor_repeat[:,:,:1,:,:] = z if image2 is not None: img_tensor_repeat[:,:,-1:,:,:] = z2 else: img_tensor_repeat[:,:,-1:,:,:] = z cond_images = model.embedder(img_tensor.unsqueeze(0)) ## blc img_emb = model.image_proj_model(cond_images) imtext_cond = torch.cat([text_emb, img_emb], dim=1) fs = torch.tensor([fs], dtype=torch.long, device=model.device) cond = {"c_crossattn": [imtext_cond], "fs": fs, "c_concat": [img_tensor_repeat]} ## inference batch_samples = batch_ddim_sampling(model, cond, noise_shape, n_samples=1, ddim_steps=steps, ddim_eta=eta, cfg_scale=cfg_scale) ## b,samples,c,t,h,w ## remove the last frame for looping video if image2 is None: batch_samples = batch_samples[:,:,:,:-1,...] video_path = './output.mp4' save_videos(batch_samples, './', filenames=['output'], fps=save_fps) return video_path i2v_examples_interp_512 = [ ['prompts/512_interp/smile_01.png', 'a smiling girl', 50, 7.5, 1.0, 5, 12306, 'prompts/512_interp/smile_02.png'], ['prompts/512_interp/stone01_01.png', 'rotating view', 50, 7.5, 1.0, 5, 123, 'prompts/512_interp/stone01_02.png'], ['prompts/512_interp/walk_01.png', 'man walking', 50, 7.5, 1.0, 5, 345, 'prompts/512_interp/walk_02.png'], ] i2v_examples_loop_512 = [ ['prompts/512_loop/24.png', 'a beach with waves and clouds at sunset', 50, 7.5, 1.0, 5, 234], ['prompts/512_loop/36.png', 'clothes swaying in the wind', 50, 7.5, 1.0, 5, 123], ['prompts/512_loop/40.png', 'flowers swaying in the wind', 50, 7.5, 1.0, 5, 234], ] css = """#input_img {max-width: 512px !important} #input_img2 {max-width: 512px !important} #output_vid {max-width: 512px; max-height: 320px} """ with gr.Blocks(analytics_enabled=False, css=css) as dynamicrafter_iface: gr.Markdown("

DynamiCrafter: Animating Open-domain Images with Video Diffusion Priors

\

\ Jinbo Xing, \ Menghan Xia, Yong Zhang, \ Haoxin Chen, Wangbo Yu,\ Hanyuan Liu, Xintao Wang,\ Tien-Tsin Wong,\ Ying Shan\

\ If DynamiCrafter is useful, please help star the \ [Github Repo]\ , which is important to Open-Source projects. Thanks!\ [ArXiv] \ [Project Page]
") #######generative frame interpolation and looping video generation###### with gr.Tab(label='Generative Frame Interpolation_320x512'): with gr.Column(): with gr.Row(): with gr.Column(): with gr.Row(): i2v_input_image = gr.Image(label="Input Image1",elem_id="input_img") with gr.Row(): i2v_input_text = gr.Text(label='Prompts') with gr.Row(): i2v_seed = gr.Slider(label='Random Seed', minimum=0, maximum=50000, step=1, value=123) i2v_eta = gr.Slider(minimum=0.0, maximum=1.0, step=0.1, label='ETA', value=1.0, elem_id="i2v_eta") i2v_cfg_scale = gr.Slider(minimum=1.0, maximum=15.0, step=0.5, label='CFG Scale', value=7.5, elem_id="i2v_cfg_scale") with gr.Row(): i2v_steps = gr.Slider(minimum=1, maximum=50, step=1, elem_id="i2v_steps", label="Sampling steps", value=50) i2v_motion = gr.Slider(minimum=5, maximum=30, step=1, elem_id="i2v_motion", label="FPS", value=10) i2v_end_btn = gr.Button("Generate") with gr.Column(): with gr.Row(): i2v_input_image2 = gr.Image(label="Input Image2",elem_id="input_img2") with gr.Row(): i2v_output_video = gr.Video(label="Generated Video",elem_id="output_vid",autoplay=True,show_share_button=True) gr.Examples(examples=i2v_examples_interp_512, inputs=[i2v_input_image, i2v_input_text, i2v_steps, i2v_cfg_scale, i2v_eta, i2v_motion, i2v_seed, i2v_input_image2], outputs=[i2v_output_video], fn = infer, cache_examples=True, ) i2v_end_btn.click(inputs=[i2v_input_image, i2v_input_text, i2v_steps, i2v_cfg_scale, i2v_eta, i2v_motion, i2v_seed, i2v_input_image2], outputs=[i2v_output_video], fn = infer ) #######generative frame interpolation and looping video generation###### with gr.Tab(label='Looping Video Generation_320x512'): with gr.Column(): with gr.Row(): with gr.Column(): with gr.Row(): i2v_input_image = gr.Image(label="Input Image",elem_id="input_img") with gr.Row(): i2v_input_text = gr.Text(label='Prompts') with gr.Row(): i2v_seed = gr.Slider(label='Random Seed', minimum=0, maximum=50000, step=1, value=123) i2v_eta = gr.Slider(minimum=0.0, maximum=1.0, step=0.1, label='ETA', value=1.0, elem_id="i2v_eta") i2v_cfg_scale = gr.Slider(minimum=1.0, maximum=15.0, step=0.5, label='CFG Scale', value=7.5, elem_id="i2v_cfg_scale") with gr.Row(): i2v_steps = gr.Slider(minimum=1, maximum=50, step=1, elem_id="i2v_steps", label="Sampling steps", value=50) i2v_motion = gr.Slider(minimum=5, maximum=30, step=1, elem_id="i2v_motion", label="FPS", value=5) i2v_end_btn = gr.Button("Generate") # with gr.Tab(label='Result'): with gr.Row(): i2v_output_video = gr.Video(label="Generated Video",elem_id="output_vid",autoplay=True,show_share_button=True) gr.Examples(examples=i2v_examples_loop_512, inputs=[i2v_input_image, i2v_input_text, i2v_steps, i2v_cfg_scale, i2v_eta, i2v_motion, i2v_seed], outputs=[i2v_output_video], fn = infer, cache_examples=True, ) i2v_end_btn.click(inputs=[i2v_input_image, i2v_input_text, i2v_steps, i2v_cfg_scale, i2v_eta, i2v_motion, i2v_seed], outputs=[i2v_output_video], fn = infer ) dynamicrafter_iface.queue(max_size=12).launch(show_api=True)