#!/usr/bin/env python3 """Composite DeLight background and character with Qwen Edit 2511 references. Picture 1 is the character-free coastal plate. Picture 2 is the DeLight character. This intentionally does not provide a mask: Qwen receives the two image roles directly and creates one integrated scene. """ from __future__ import annotations import argparse import hashlib import importlib.metadata import json import time from pathlib import Path from PIL import Image ASSETS = Path(__file__).resolve().parent PROJECT_ROOT = ASSETS.parents[3] CACHE_DIR = PROJECT_ROOT / ".tmp" / "download" / "huggingface" / "hub" MODEL_ID = "Qwen/Qwen-Image-Edit-2511" SCENE_CONFIG = { "a": {"background": "p7-5-4-qwen-2509-studio-delight-camera-a-background-v1-size-1280x1280-seed-62294-steps-10.png", "character": "p7-5-4-qwen-2511-bfs-head-v5-delight-character-cutout-a-quarter-left-v1-size-1280x1280-seed-62294-steps-10.png", "place": "coastal"}, "b": {"background": "p7-5-4-qwen-2509-studio-delight-background-b-size-1280x1280-seed-62294-steps-10.png", "character": "p7-5-4-qwen-2511-bfs-head-v5-delight-character-cutout-b-quarter-left-v1-size-1280x1280-seed-62294-steps-10.png", "place": "wildflower meadow"}, "c": {"background": "p7-5-4-qwen-2509-studio-delight-background-c-size-1280x1280-seed-62294-steps-10.png", "character": "p7-5-4-qwen-2511-bfs-head-v5-delight-character-cutout-c-quarter-left-v1-size-1280x1280-seed-62294-steps-10.png", "place": "city park"}, } def sha256(path: Path) -> str: return hashlib.sha256(path.read_bytes()).hexdigest() def square(path: Path, size: int) -> Image.Image: """Use equal RGB canvases without distorting the two references.""" with Image.open(path) as source: source = source.convert("RGB") if source.size != (size, size): source = source.resize((size, size), Image.Resampling.LANCZOS) return source.copy() def main() -> None: parser = argparse.ArgumentParser(description=__doc__) parser.add_argument("--scene-id", choices=tuple(SCENE_CONFIG), default="c") parser.add_argument("--background", type=Path, help="Override the Scene's character-free DeLight background.") parser.add_argument("--character", type=Path, help="Override the Scene's BFS-refined DeLight character.") parser.add_argument("--prompt", help="Override the scene-aware positive integration prompt.") parser.add_argument("--seed", type=int, default=62294) parser.add_argument("--steps", type=int, default=10) parser.add_argument("--size", type=int, default=1280) parser.add_argument("--run-label", help="Defaults to the selected Scene's BFS 45-degree integration label.") parser.add_argument("--output-dir", type=Path, default=ASSETS) parser.add_argument("--dry-run", action="store_true") args = parser.parse_args() if args.steps < 1 or args.size < 32 or args.size % 32: parser.error("--steps must be positive and --size must be a positive multiple of 32") config = SCENE_CONFIG[args.scene_id] background = (args.background or ASSETS / config["background"]).resolve() character = (args.character or ASSETS / config["character"]).resolve() prompt = args.prompt or f"Place the woman in Picture 2 into Picture 1. Preserve the {config['place']} background and composition of Picture 1. Preserve the split-leap pose, identity, and outfit of the woman in Picture 2." run_label = args.run_label or f"scene-{args.scene_id}-bfs-quarter-left-v1" for path in (background, character): if not path.is_file(): raise FileNotFoundError(path) output_dir = args.output_dir.resolve() stem = f"p7-5-4-qwen-2511-delight-multireference-composite-{run_label}-size-{args.size}x{args.size}-seed-{args.seed}-steps-{args.steps}" output, result = output_dir / f"{stem}.png", output_dir / f"{stem}-result.json" if args.dry_run: print(json.dumps({"scene_id": args.scene_id, "background": str(background), "character": str(character), "prompt": prompt, "output": str(output), "result": str(result)}, ensure_ascii=False)) return import torch from diffusers import QwenImageEditPlusPipeline started = time.monotonic() pipeline = QwenImageEditPlusPipeline.from_pretrained( MODEL_ID, torch_dtype=torch.bfloat16, cache_dir=CACHE_DIR, local_files_only=True ) pipeline.enable_attention_slicing("max") pipeline.enable_sequential_cpu_offload() try: image = pipeline( image=[square(background, args.size), square(character, args.size)], prompt=prompt, width=args.size, height=args.size, generator=torch.Generator("cpu").manual_seed(args.seed), true_cfg_scale=4.0, guidance_scale=1.0, negative_prompt=" ", num_inference_steps=args.steps, ).images[0].convert("RGB") finally: del pipeline torch.cuda.empty_cache() output_dir.mkdir(parents=True, exist_ok=True) image.save(output) result.write_text( json.dumps( { "status": "generated", "stage": "delight_multireference_composite", "execution_mode": "direct Diffusers; Qwen Image Edit 2511; no mask; no ComfyUI", "model": MODEL_ID, "inputs": [ {"role": "Picture 1: DeLight character-free background", "path": str(background), "sha256": sha256(background)}, {"role": "Picture 2: DeLight character identity, pose, and outfit", "path": str(character), "sha256": sha256(character)}, ], "scene_id": args.scene_id, "prompt": prompt, "seed": args.seed, "steps": args.steps, "true_cfg_scale": 4.0, "output": {"path": str(output), "sha256": sha256(output), "width": image.width, "height": image.height}, "runtime": {name: importlib.metadata.version(name) for name in ("diffusers", "torch", "transformers", "accelerate")}, "elapsed_seconds": round(time.monotonic() - started, 2), }, ensure_ascii=False, indent=2, ) + "\n", encoding="utf-8", ) print(json.dumps({"output": str(output), "result": str(result)}, ensure_ascii=False)) if __name__ == "__main__": main()