File size: 4,622 Bytes
4ec41f2 | 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 | #!/usr/bin/env python3
"""Generate the eight bundled StatePlay examples with one model load."""
from __future__ import annotations
import argparse
import csv
import json
import sys
from pathlib import Path
import torch
from PIL import Image
REPO_ROOT = Path(__file__).resolve().parents[1]
if str(REPO_ROOT) not in sys.path:
sys.path.insert(0, str(REPO_ROOT))
from stateplay import NEG_PROMPT, StatePlayPipeline # noqa: E402
from stateplay.utils import save_mp4 # noqa: E402
def build_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(description="Generate all eight bundled StatePlay examples")
parser.add_argument("--base-model", default=str(REPO_ROOT / "base_model"), help="Base-model root containing Wan-AI/ (default: <repo>/base_model)")
parser.add_argument("--checkpoint", default=str(REPO_ROOT / "examples" / "checkpoint" / "StatePlay.safetensors"))
parser.add_argument("--manifest", default=str(REPO_ROOT / "examples" / "manifest.csv"))
parser.add_argument("--output-dir", default=str(REPO_ROOT / "examples" / "generated"))
parser.add_argument("--device", default="cuda")
parser.add_argument("--only", nargs="*", metavar="ID", help="Optional IDs, e.g. --only 01 03. Default: all eight.")
parser.add_argument("--skip-existing", action="store_true")
parser.add_argument("--steps", type=int, default=30)
parser.add_argument("--cfg", type=float, default=5.0)
parser.add_argument("--action-cfg", type=float, default=1.0)
parser.add_argument("--seed", type=int, default=2)
return parser
def main() -> None:
args = build_parser().parse_args()
manifest_path = Path(args.manifest).resolve()
examples_root = manifest_path.parent
output_dir = Path(args.output_dir).resolve()
output_dir.mkdir(parents=True, exist_ok=True)
with manifest_path.open(newline="", encoding="utf-8") as handle:
rows = list(csv.DictReader(handle))
if args.only:
wanted = {item.zfill(2) for item in args.only}
rows = [row for row in rows if row["id"].zfill(2) in wanted]
missing = wanted - {row["id"].zfill(2) for row in rows}
if missing:
raise ValueError(f"Unknown example IDs: {sorted(missing)}")
if not rows:
raise ValueError("No examples selected")
print(f"Loading StatePlay once for {len(rows)} example(s)...", flush=True)
pipeline = StatePlayPipeline.from_pretrained(base_model_dir=args.base_model, checkpoint_path=args.checkpoint, torch_dtype=torch.bfloat16).to(args.device)
run_records = []
for position, row in enumerate(rows, start=1):
input_dir = examples_root / row["input_dir"]
video_output = output_dir / row["output_name"]
state_output = video_output.with_name(f"{video_output.stem}_state.txt")
expected_outputs_exist = video_output.exists() and (pipeline.uses_latent_state or state_output.exists())
if args.skip_existing and expected_outputs_exist:
print(f"[{position}/{len(rows)}] skip existing {video_output.name}", flush=True)
continue
prompt = (input_dir / "prompt.txt").read_text(encoding="utf-8").strip()
actions = input_dir / "actions.parquet"
with Image.open(input_dir / "first_frame.png") as image:
result = pipeline(
image=image.convert("RGB"), actions_parquet=str(actions), state_parquet=str(actions),
state_txt=None if pipeline.uses_latent_state else str(state_output),
state_sampling="end", prompt=prompt, negative_prompt=NEG_PROMPT,
num_frames=101, num_inference_steps=args.steps, cfg_scale=args.cfg, state_cfg_scale=1.0,
action_cfg_scale=args.action_cfg, seed=args.seed,
)
save_mp4(result.frames[0], str(video_output), fps=20)
run_records.append({
"id": row["id"], "input_dir": row["input_dir"], "output_video": video_output.name,
"output_state": None if pipeline.uses_latent_state else state_output.name,
"checkpoint": Path(args.checkpoint).name,
"parameters": {"state_sampling": "end", "steps": args.steps, "cfg": args.cfg, "state_cfg": 1.0,
"action_cfg": args.action_cfg, "seed": args.seed, "num_frames": 101, "fps": 20},
})
print(f"[{position}/{len(rows)}] saved {video_output}", flush=True)
(output_dir / "run_manifest.json").write_text(json.dumps(run_records, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
print(f"Finished. Outputs: {output_dir}", flush=True)
if __name__ == "__main__":
main()
|