import gradio as gr import os import sys import argparse import random import time from omegaconf import OmegaConf import torch import torchvision 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 ) SECRET_TOKEN = os.getenv('SECRET_TOKEN', 'default_secret') def download_model(): REPO_ID = 'Doubiiu/DynamiCrafter_1024' filename_list = ['model.ckpt'] if not os.path.exists('./checkpoints/dynamicrafter_1024_v1/'): os.makedirs('./checkpoints/dynamicrafter_1024_v1/') for filename in filename_list: local_file = os.path.join('./checkpoints/dynamicrafter_1024_v1/', filename) if not os.path.exists(local_file): hf_hub_download(repo_id=REPO_ID, filename=filename, local_dir='./checkpoints/dynamicrafter_1024_v1/', force_download=True) def infer(secret_token, image, prompt, steps=50, cfg_scale=7.5, eta=1.0, fs=3, seed=123): if secret_token != SECRET_TOKEN: raise gr.Error( f'Invalid secret token. Please fork the original space if you want to use it for yourself.') resolution = (576, 1024) download_model() ckpt_path='checkpoints/dynamicrafter_1024_v1/model.ckpt' config_file='configs/inference_1024_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() 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 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 img_tensor_repeat = repeat(z, 'b c t h w -> b c (repeat t) h w', repeat=frames) 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 video_path = './output.mp4' save_videos(batch_samples, './', filenames=['output'], fps=save_fps) model = model.cpu() # Read the content of the video file and encode it to base64 with open(video_path, "rb") as video_file: video_base64 = base64.b64encode(video_file.read()).decode('utf-8') # Prepend the appropriate data URI header with MIME type video_data_uri = 'data:video/mp4;base64,' + video_base64 # clean-up (otherwise there is a risk of "ghosting", eg. someone seeing the previous generated video", # of one of the steps go wrong) os.remove(video_path) return video_data_uri with gr.Blocks() as app: gr.HTML("""

This space is a REST API to programmatically generate MP4 videos.

Interested in using it? Look no further than the original space!

""") secret_token = gr.Text(label='Secret Token', max_lines=1) i2v_input_image = gr.Image(label="Input Image",elem_id="input_img") i2v_input_text = gr.Text(label='Prompts') i2v_seed = gr.Slider(label='Random Seed', minimum=0, maximum=10000, 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") i2v_steps = gr.Slider(minimum=1, maximum=60, step=1, elem_id="i2v_steps", label="Sampling steps", value=50) i2v_motion = gr.Slider(minimum=5, maximum=20, step=1, elem_id="i2v_motion", label="FPS", value=10) i2v_end_btn = gr.Button("Generate") i2v_output_video_base64 = gr.Text() i2v_end_btn.click(inputs=[secret_token, i2v_input_image, i2v_input_text, i2v_steps, i2v_cfg_scale, i2v_eta, i2v_motion, i2v_seed], outputs=[i2v_output_video_base64], fn = infer ) app.queue(max_size=4).launch(show_api=True)