| |
| |
| |
| |
| @@ -2,6 +2,8 @@ |
| # Copyright (c) Alibaba, Inc. and its affiliates. |
| |
| import argparse |
| +import gc |
| +import json |
| import time |
| from datetime import datetime |
| import logging |
| @@ -178,9 +180,48 @@ def get_parser(): |
| type=float, |
| default=5.0, |
| help="Classifier free guidance scale.") |
| + parser.add_argument( |
| + "--batch_manifest", |
| + type=str, |
| + default=None, |
| + help=( |
| + "JSON or JSONL file with many inference jobs. The WanVace model is " |
| + "loaded once, then each job is processed sequentially. Each job can " |
| + "override src_video, src_mask, src_ref_images, save_dir, save_file, " |
| + "prompt, base_seed, frame_num, sample_steps, sample_shift, " |
| + "sample_guide_scale, sample_solver, size, and use_prompt_extend." |
| + )) |
| + parser.add_argument( |
| + "--skip_existing", |
| + type=str2bool, |
| + default=False, |
| + help="In batch mode, skip jobs whose output video already exists.") |
| + parser.add_argument( |
| + "--batch_continue_on_error", |
| + type=str2bool, |
| + default=True, |
| + help="In batch mode, continue after a failed job instead of aborting.") |
| return parser |
| |
| |
| +BATCH_JOB_FIELDS = { |
| + "size", |
| + "frame_num", |
| + "save_dir", |
| + "save_file", |
| + "src_video", |
| + "src_mask", |
| + "src_ref_images", |
| + "prompt", |
| + "use_prompt_extend", |
| + "base_seed", |
| + "sample_solver", |
| + "sample_steps", |
| + "sample_shift", |
| + "sample_guide_scale", |
| +} |
| + |
| + |
| def _init_logging(rank): |
| # logging |
| if rank == 0: |
| @@ -193,6 +234,175 @@ def _init_logging(rank): |
| logging.basicConfig(level=logging.ERROR) |
| |
| |
| +def load_batch_jobs(path): |
| + with open(path, "r", encoding="utf-8") as f: |
| + text = f.read().strip() |
| + if not text: |
| + return [] |
| + |
| + if path.endswith(".jsonl"): |
| + jobs = [] |
| + for line_no, line in enumerate(text.splitlines(), start=1): |
| + line = line.strip() |
| + if not line: |
| + continue |
| + job = json.loads(line) |
| + if not isinstance(job, dict): |
| + raise ValueError(f"{path}:{line_no} is not a JSON object") |
| + jobs.append(job) |
| + return jobs |
| + |
| + data = json.loads(text) |
| + if isinstance(data, dict) and "jobs" in data: |
| + data = data["jobs"] |
| + if not isinstance(data, list): |
| + raise ValueError("--batch_manifest JSON must be a list or an object with a 'jobs' list") |
| + if not all(isinstance(job, dict) for job in data): |
| + raise ValueError("--batch_manifest jobs must be JSON objects") |
| + return data |
| + |
| + |
| +def make_job_args(base_args, job): |
| + values = vars(base_args).copy() |
| + for static_key in ("model_name", "ckpt_dir"): |
| + if static_key in job and str(job[static_key]) != str(values[static_key]): |
| + raise ValueError( |
| + f"Batch job cannot override {static_key}; the model is loaded once " |
| + f"from the top-level {static_key}." |
| + ) |
| + for key, value in job.items(): |
| + normalized = key.replace("-", "_") |
| + if normalized in BATCH_JOB_FIELDS and value is not None: |
| + values[normalized] = value |
| + return argparse.Namespace(**values) |
| + |
| + |
| +def maybe_extend_prompt(args, prompt_expander, rank): |
| + if args.prompt is None: |
| + args.prompt = "" |
| + if args.use_prompt_extend and args.use_prompt_extend != 'plain': |
| + logging.info("Extending prompt ...") |
| + if rank == 0: |
| + prompt = prompt_expander.forward(args.prompt) |
| + logging.info(f"Prompt extended from '{args.prompt}' to '{prompt}'") |
| + input_prompt = [prompt] |
| + else: |
| + input_prompt = [None] |
| + if dist.is_initialized(): |
| + dist.broadcast_object_list(input_prompt, src=0) |
| + args.prompt = input_prompt[0] |
| + logging.info(f"Extended prompt: {args.prompt}") |
| + return args.prompt |
| + |
| + |
| +def output_path_for_args(args, model_name): |
| + if args.save_file is not None: |
| + return args.save_file |
| + if args.save_dir is None: |
| + save_dir = os.path.join('results', model_name, time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime(time.time()))) |
| + else: |
| + save_dir = args.save_dir |
| + return os.path.join(save_dir, 'out_video.mp4') |
| + |
| + |
| +def save_outputs(video, src_video, src_mask, src_ref_images, cfg, args): |
| + ret_data = {} |
| + if args.save_dir is None: |
| + save_dir = os.path.join('results', args.model_name, time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime(time.time()))) |
| + else: |
| + save_dir = args.save_dir |
| + if not os.path.exists(save_dir): |
| + os.makedirs(save_dir) |
| + |
| + if args.save_file is not None: |
| + save_file = args.save_file |
| + else: |
| + save_file = os.path.join(save_dir, 'out_video.mp4') |
| + cache_video( |
| + tensor=video[None], |
| + save_file=save_file, |
| + fps=cfg.sample_fps, |
| + nrow=1, |
| + normalize=True, |
| + value_range=(-1, 1)) |
| + logging.info(f"Saving generated video to {save_file}") |
| + ret_data['out_video'] = save_file |
| + |
| + save_file = os.path.join(save_dir, 'src_video.mp4') |
| + cache_video( |
| + tensor=src_video[0][None], |
| + save_file=save_file, |
| + fps=cfg.sample_fps, |
| + nrow=1, |
| + normalize=True, |
| + value_range=(-1, 1)) |
| + logging.info(f"Saving src_video to {save_file}") |
| + ret_data['src_video'] = save_file |
| + |
| + save_file = os.path.join(save_dir, 'src_mask.mp4') |
| + cache_video( |
| + tensor=src_mask[0][None], |
| + save_file=save_file, |
| + fps=cfg.sample_fps, |
| + nrow=1, |
| + normalize=True, |
| + value_range=(0, 1)) |
| + logging.info(f"Saving src_mask to {save_file}") |
| + ret_data['src_mask'] = save_file |
| + |
| + if src_ref_images[0] is not None: |
| + for i, ref_img in enumerate(src_ref_images[0]): |
| + save_file = os.path.join(save_dir, f'src_ref_image_{i}.png') |
| + cache_image( |
| + tensor=ref_img[:, 0, ...], |
| + save_file=save_file, |
| + nrow=1, |
| + normalize=True, |
| + value_range=(-1, 1)) |
| + logging.info(f"Saving src_ref_image_{i} to {save_file}") |
| + ret_data[f'src_ref_image_{i}'] = save_file |
| + return ret_data |
| + |
| + |
| +def run_one_job(wan_vace, cfg, args, device, rank, prompt_expander=None): |
| + maybe_extend_prompt(args, prompt_expander, rank) |
| + logging.info(f"Input prompt: {args.prompt}") |
| + |
| + src_video, src_mask, src_ref_images = wan_vace.prepare_source( |
| + [args.src_video], |
| + [args.src_mask], |
| + [None if args.src_ref_images is None else args.src_ref_images.split(',')], |
| + args.frame_num, |
| + SIZE_CONFIGS[args.size], |
| + device, |
| + ) |
| + |
| + logging.info(f"Generating video...") |
| + video = wan_vace.generate( |
| + args.prompt, |
| + src_video, |
| + src_mask, |
| + src_ref_images, |
| + size=SIZE_CONFIGS[args.size], |
| + frame_num=args.frame_num, |
| + shift=args.sample_shift, |
| + sample_solver=args.sample_solver, |
| + sampling_steps=args.sample_steps, |
| + guide_scale=args.sample_guide_scale, |
| + seed=args.base_seed, |
| + offload_model=args.offload_model) |
| + |
| + ret_data = {} |
| + if rank == 0: |
| + ret_data = save_outputs(video, src_video, src_mask, src_ref_images, cfg, args) |
| + |
| + del video, src_video, src_mask, src_ref_images |
| + gc.collect() |
| + if torch.cuda.is_available(): |
| + torch.cuda.empty_cache() |
| + return ret_data |
| + |
| + |
| def main(args): |
| args = argparse.Namespace(**args) if isinstance(args, dict) else args |
| args = validate_args(args) |
| @@ -235,6 +445,7 @@ def main(args): |
| ulysses_degree=args.ulysses_size, |
| ) |
| |
| + prompt_expander = None |
| if args.use_prompt_extend and args.use_prompt_extend != 'plain': |
| prompt_expander = get_annotator(config_type='prompt', config_task=args.use_prompt_extend, return_dict=False) |
| |
| @@ -250,26 +461,12 @@ def main(args): |
| dist.broadcast_object_list(base_seed, src=0) |
| args.base_seed = base_seed[0] |
| |
| - if args.prompt is None: |
| + if args.batch_manifest is None and args.prompt is None: |
| args.prompt = EXAMPLE_PROMPT[args.model_name]["prompt"] |
| args.src_video = EXAMPLE_PROMPT[args.model_name].get("src_video", None) |
| args.src_mask = EXAMPLE_PROMPT[args.model_name].get("src_mask", None) |
| args.src_ref_images = EXAMPLE_PROMPT[args.model_name].get("src_ref_images", None) |
| |
| - logging.info(f"Input prompt: {args.prompt}") |
| - if args.use_prompt_extend and args.use_prompt_extend != 'plain': |
| - logging.info("Extending prompt ...") |
| - if rank == 0: |
| - prompt = prompt_expander.forward(args.prompt) |
| - logging.info(f"Prompt extended from '{args.prompt}' to '{prompt}'") |
| - input_prompt = [prompt] |
| - else: |
| - input_prompt = [None] |
| - if dist.is_initialized(): |
| - dist.broadcast_object_list(input_prompt, src=0) |
| - args.prompt = input_prompt[0] |
| - logging.info(f"Extended prompt: {args.prompt}") |
| - |
| logging.info("Creating WanT2V pipeline.") |
| wan_vace = WanVace( |
| config=cfg, |
| @@ -282,82 +479,41 @@ def main(args): |
| t5_cpu=args.t5_cpu, |
| ) |
| |
| - src_video, src_mask, src_ref_images = wan_vace.prepare_source([args.src_video], |
| - [args.src_mask], |
| - [None if args.src_ref_images is None else args.src_ref_images.split(',')], |
| - args.frame_num, SIZE_CONFIGS[args.size], device) |
| - |
| - logging.info(f"Generating video...") |
| - video = wan_vace.generate( |
| - args.prompt, |
| - src_video, |
| - src_mask, |
| - src_ref_images, |
| - size=SIZE_CONFIGS[args.size], |
| - frame_num=args.frame_num, |
| - shift=args.sample_shift, |
| - sample_solver=args.sample_solver, |
| - sampling_steps=args.sample_steps, |
| - guide_scale=args.sample_guide_scale, |
| - seed=args.base_seed, |
| - offload_model=args.offload_model) |
| - |
| - ret_data = {} |
| - if rank == 0: |
| - if args.save_dir is None: |
| - save_dir = os.path.join('results', args.model_name, time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime(time.time()))) |
| - else: |
| - save_dir = args.save_dir |
| - if not os.path.exists(save_dir): |
| - os.makedirs(save_dir) |
| + if args.batch_manifest is not None: |
| + jobs = load_batch_jobs(args.batch_manifest) |
| + logging.info(f"Loaded {len(jobs)} batch jobs from {args.batch_manifest}") |
| + results = [] |
| + for idx, job in enumerate(jobs): |
| + job_args = validate_args(make_job_args(args, job)) |
| + job_name = job.get("name") or job.get("id") or idx |
| + out_video = output_path_for_args(job_args, job_args.model_name) |
| + if args.skip_existing and rank == 0 and os.path.exists(out_video): |
| + logging.info(f"[batch {idx + 1}/{len(jobs)}] skip existing {job_name}: {out_video}") |
| + results.append({"index": idx, "name": job_name, "status": "skipped", "out_video": out_video}) |
| + continue |
| + |
| + logging.info(f"[batch {idx + 1}/{len(jobs)}] start {job_name}") |
| + try: |
| + ret = run_one_job(wan_vace, cfg, job_args, device, rank, prompt_expander) |
| + if rank == 0: |
| + results.append({"index": idx, "name": job_name, "status": "ok", **ret}) |
| + logging.info(f"[batch {idx + 1}/{len(jobs)}] ok {job_name}") |
| + except Exception as exc: |
| + logging.exception(f"[batch {idx + 1}/{len(jobs)}] failed {job_name}: {exc}") |
| + if rank == 0: |
| + results.append({"index": idx, "name": job_name, "status": "failed", "error": str(exc)}) |
| + if not args.batch_continue_on_error: |
| + raise |
| |
| - if args.save_file is not None: |
| - save_file = args.save_file |
| - else: |
| - save_file = os.path.join(save_dir, 'out_video.mp4') |
| - cache_video( |
| - tensor=video[None], |
| - save_file=save_file, |
| - fps=cfg.sample_fps, |
| - nrow=1, |
| - normalize=True, |
| - value_range=(-1, 1)) |
| - logging.info(f"Saving generated video to {save_file}") |
| - ret_data['out_video'] = save_file |
| - |
| - save_file = os.path.join(save_dir, 'src_video.mp4') |
| - cache_video( |
| - tensor=src_video[0][None], |
| - save_file=save_file, |
| - fps=cfg.sample_fps, |
| - nrow=1, |
| - normalize=True, |
| - value_range=(-1, 1)) |
| - logging.info(f"Saving src_video to {save_file}") |
| - ret_data['src_video'] = save_file |
| - |
| - save_file = os.path.join(save_dir, 'src_mask.mp4') |
| - cache_video( |
| - tensor=src_mask[0][None], |
| - save_file=save_file, |
| - fps=cfg.sample_fps, |
| - nrow=1, |
| - normalize=True, |
| - value_range=(0, 1)) |
| - logging.info(f"Saving src_mask to {save_file}") |
| - ret_data['src_mask'] = save_file |
| - |
| - if src_ref_images[0] is not None: |
| - for i, ref_img in enumerate(src_ref_images[0]): |
| - save_file = os.path.join(save_dir, f'src_ref_image_{i}.png') |
| - cache_image( |
| - tensor=ref_img[:, 0, ...], |
| - save_file=save_file, |
| - nrow=1, |
| - normalize=True, |
| - value_range=(-1, 1)) |
| - logging.info(f"Saving src_ref_image_{i} to {save_file}") |
| - ret_data[f'src_ref_image_{i}'] = save_file |
| + if rank == 0: |
| + result_path = args.batch_manifest + ".results.jsonl" |
| + with open(result_path, "w", encoding="utf-8") as f: |
| + for row in results: |
| + f.write(json.dumps(row, ensure_ascii=False) + "\n") |
| + logging.info(f"Wrote batch results to {result_path}") |
| + ret_data = {"batch_results": results if rank == 0 else []} |
| + else: |
| + ret_data = run_one_job(wan_vace, cfg, args, device, rank, prompt_expander) |
| logging.info("Finished.") |
| return ret_data |
| |
|
|
