Update app.py

app.py

@@ -1,158 +1,270 @@
+# app.py
 import os
+import oss2
 import sys
-import subprocess
+import uuid
 import shutil
-import spaces
+import time
 import gradio as gr
-import torch
-from huggingface_hub import snapshot_download
-
-# ── Patch wan/modules/t5.py before importing wan ─────────────────────────────
-def clone_and_patch_wan():
-    if not os.path.exists("./Wan2.2"):
-        subprocess.run(
-            ["git", "clone", "https://github.com/Wan-Video/Wan2.2.git", "./Wan2.2"],
-            check=True
+import requests
+from pathlib import Path
+from datetime import datetime, timedelta
+import dashscope
+# from dashscope.utils.oss_utils import check_and_upload_local
+
+DASHSCOPE_API_KEY = os.getenv("DASHSCOPE_API_KEY")
+dashscope.api_key = DASHSCOPE_API_KEY
+
+def get_upload_policy(api_key, model_name):
+    """获取文件上传凭证"""
+    url = "https://dashscope.aliyuncs.com/api/v1/uploads"
+    headers = {
+        "Authorization": f"Bearer {api_key}",
+        "Content-Type": "application/json"
+    }
+    params = {
+        "action": "getPolicy",
+        "model": model_name
+    }
+    response = requests.get(url, headers=headers, params=params)
+    if response.status_code != 200:
+        raise Exception(f"Failed to get upload policy: {response.text}")
+    return response.json()['data']
+
+def upload_file_to_oss(policy_data, file_path):
+    """将文件上传到临时存储OSS"""
+    file_name = Path(file_path).name
+    key = f"{policy_data['upload_dir']}/{file_name}"
+    with open(file_path, 'rb') as file:
+        files = {
+            'OSSAccessKeyId': (None, policy_data['oss_access_key_id']),
+            'Signature': (None, policy_data['signature']),
+            'policy': (None, policy_data['policy']),
+            'x-oss-object-acl': (None, policy_data['x_oss_object_acl']),
+            'x-oss-forbid-overwrite': (None, policy_data['x_oss_forbid_overwrite']),
+            'key': (None, key),
+            'success_action_status': (None, '200'),
+            'file': (file_name, file)
+        }
+        response = requests.post(policy_data['upload_host'], files=files)
+        if response.status_code != 200:
+            raise Exception(f"Failed to upload file: {response.text}")
+    return f"oss://{key}"
+
+def upload_file_and_get_url(api_key, model_name, file_path):
+    """上传文件并获取URL"""
+    # 1. 获取上传凭证，上传凭证接口有限流，超出限流将导致请求失败
+    policy_data = get_upload_policy(api_key, model_name)
+    # 2. 上传文件到OSS
+    oss_url = upload_file_to_oss(policy_data, file_path)
+    return oss_url
+
+
+class WanAnimateApp:
+    def __init__(self, url, get_url):
+        self.url = url
+        self.get_url = get_url
+
+    def predict(
+        self,
+        ref_img,
+        video,
+        model_id,
+        model,
+    ):
+        # Upload files to OSS if needed and get URLs
+        image_url = upload_file_and_get_url(DASHSCOPE_API_KEY, model_id, ref_img)
+        video_url = upload_file_and_get_url(DASHSCOPE_API_KEY, model_id, video)
+
+        # Prepare the request payload
+        payload = {
+            "model": model_id,
+            "input": {
+                "image_url": image_url,
+                "video_url": video_url
+            },
+            "parameters": {
+                "check_image": True,
+                "mode": model,
+            }
+        }
+
+        # Set up headers
+        headers = {
+            "X-DashScope-Async": "enable",
+            "X-DashScope-OssResourceResolve": "enable",
+            "Authorization": f"Bearer {DASHSCOPE_API_KEY}",
+            "Content-Type": "application/json"
+        }
+
+        # Make the initial API request
+        url = self.url
+        response = requests.post(url, json=payload, headers=headers, timeout=60)
+
+        # Check if request was successful
+        if response.status_code != 200:
+            raise Exception(f"Initial request failed with status code {response.status_code}: {response.text}")
+
+        # Get the task ID from response
+        result = response.json()
+        task_id = result.get("output", {}).get("task_id")
+        if not task_id:
+            raise Exception("Failed to get task ID from response")
+
+        # Poll for results
+        get_url = f"{self.get_url}/{task_id}"
+        headers = {
+            "Authorization": f"Bearer {DASHSCOPE_API_KEY}",
+            "Content-Type": "application/json"
+        }
+
+        while True:
+            response = requests.get(get_url, headers=headers, timeout=60)
+            if response.status_code != 200:
+                raise Exception(f"Failed to get task status: {response.status_code}: {response.text}")
+
+            result = response.json()
+            print(result)
+            task_status = result.get("output", {}).get("task_status")
+
+            if task_status == "SUCCEEDED":
+                # Task completed successfully, return video URL
+                video_url = result["output"]["results"]["video_url"]
+                return video_url, "SUCCEEDED"
+            elif task_status == "PENDING" or task_status == "RUNNING":
+                # Task is still running, wait and retry
+                time.sleep(10)  # Wait 10 seconds before polling again
+            else:
+                # Task failed or unknown, raise an exception with error message
+                error_msg = result.get("output", {}).get("message", "Unknown error")
+                code_msg = result.get("output", {}).get("code", "Unknown code")
+                print(f"\n\nTask failed: {error_msg} Code: {code_msg} TaskId: {task_id}\n\n")
+                return None, f"Task failed: {error_msg} Code: {code_msg} TaskId: {task_id}"
+
+
+def start_app():
+    import argparse
+    parser = argparse.ArgumentParser(description="Wan2.2-Animate 视频生成工具")
+    args = parser.parse_args()
+
+    url = "https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis/"
+    get_url = f"https://dashscope.aliyuncs.com/api/v1/tasks/"
+
+    app = WanAnimateApp(url=url, get_url=get_url)
+
+    with gr.Blocks(title="Wan2.2-Animate 视频生成") as demo:
+        gr.HTML("""
+<div style="padding: 2rem; text-align: center; max-width: 1200px; margin: 0 auto; font-family: Arial, sans-serif;">
+  <h1 style="font-size: 2.5rem; font-weight: bold; margin-bottom: 0.5rem; color: #333;">
+    Wan2.2-Animate: Unified Character Animation and Replacement with Holistic Replication
+  </h1>
+  <h3 style="font-size: 2.5rem; font-weight: bold; margin-bottom: 0.5rem; color: #333;">
+    Wan2.2-Animate: 统一的角色动画和视频人物替换模型
+  </h3>
+  <div style="font-size: 1.25rem; margin-bottom: 1.5rem; color: #555;">
+    Tongyi Lab, Alibaba
+  </div>
+  <div style="display: flex; flex-wrap: wrap; justify-content: center; gap: 1rem; margin-bottom: 1rem;">
+    <a href="https://arxiv.org/abs/2509.14055" target="_blank" style="display: inline-flex; align-items: center; padding: 0.5rem 1rem; background-color: #f0f0f0; color: #333; text-decoration: none; border-radius: 9999px; font-weight: 500;">
+      <span style="margin-right: 0.5rem;">📄</span><span>Paper</span>
+    </a>
+    <a href="https://github.com/Wan-Video/Wan2.2" target="_blank" style="display: inline-flex; align-items: center; padding: 0.5rem 1rem; background-color: #f0f0f0; color: #333; text-decoration: none; border-radius: 9999px; font-weight: 500;">
+      <span style="margin-right: 0.5rem;">💻</span><span>GitHub</span>
+    </a>
+    <a href="https://huggingface.co/Wan-AI/Wan2.2-Animate-14B" target="_blank" style="display: inline-flex; align-items: center; padding: 0.5rem 1rem; background-color: #f0f0f0; color: #333; text-decoration: none; border-radius: 9999px; font-weight: 500;">
+      <span style="margin-right: 0.5rem;">🤗</span><span>HF Model</span>
+    </a>
+    <a href="https://www.modelscope.cn/models/Wan-AI/Wan2.2-Animate-14B" target="_blank" style="display: inline-flex; align-items: center; padding: 0.5rem 1rem; background-color: #f0f0f0; color: #333; text-decoration: none; border-radius: 9999px; font-weight: 500;">
+      <span style="margin-right: 0.5rem;">🤖</span><span>MS Model</span>
+    </a>
+  </div>
+</div>
+""")
+        gr.HTML("""
+<details>
+<summary>‼️Usage (使用说明)</summary>
+Wan-Animate supports two mode:
+<ul>
+  <li>Move Mode: animate the character in input image with movements from the input video</li>
+  <li>Mix Mode: replace the character in input video with the character in input image</li>
+</ul>
+Currently, the following restrictions apply to inputs:
+<ul>
+  <li>Video file size: Less than 200MB</li>
+  <li>Video resolution: The shorter side must be greater than 200, and the longer side must be less than 2048</li>
+  <li>Video duration: 2s to 30s</li>
+  <li>Video aspect ratio: 1:3 to 3:1</li>
+  <li>Video formats: mp4, avi, mov</li>
+  <li>Image file size: Less than 5MB</li>
+  <li>Image resolution: The shorter side must be greater than 200, and the longer side must be less than 4096</li>
+  <li>Image formats: jpg, png, jpeg, webp, bmp</li>
+</ul>
+<ul>
+  <li> wan-pro: 25fps, 720p </li>
+  <li> wan-std: 15fps, 720p </li>
+</ul>
+</details>
+""")
+        with gr.Row():
+            with gr.Column():
+                ref_img = gr.Image(
+                    label="Reference Image(参考图像)",
+                    type="filepath",
+                    sources=["upload"],
+                )
+                video = gr.Video(
+                    label="Template Video(模版视频)",
+                    sources=["upload"],
+                )
+                with gr.Row():
+                    model_id = gr.Dropdown(
+                        label="Mode(模式)",
+                        choices=["wan2.2-animate-move", "wan2.2-animate-mix"],
+                        value="wan2.2-animate-move",
+                        info=""
+                    )
+                    model = gr.Dropdown(
+                        label="推理质量(Inference Quality)",
+                        choices=["wan-pro", "wan-std"],
+                        value="wan-pro",
+                    )
+                run_button = gr.Button("Generate Video(生成视频)")
+            with gr.Column():
+                output_video = gr.Video(label="Output Video(输出视频)")
+                output_status = gr.Textbox(label="Status(状态)")
+
+        run_button.click(
+            fn=app.predict,
+            inputs=[
+                ref_img,
+                video,
+                model_id,
+                model,
+            ],
+            outputs=[output_video, output_status],
         )
-    t5_path = "./Wan2.2/wan/modules/t5.py"
-    with open(t5_path, "r") as f:
-        src = f.read()
-    if "device=torch.cuda.current_device()," in src:
-        src = src.replace("device=torch.cuda.current_device(),", "device=0,")
-        with open(t5_path, "w") as f:
-            f.write(src)
-        print("[patch] t5.py patched: replaced current_device() with 0")
-
-clone_and_patch_wan()
-
-if "./Wan2.2" not in sys.path:
-    sys.path.insert(0, "./Wan2.2")
-
-# ── Download SAM2 CPU model ───────────────────────────────────────────────────
-if not os.path.exists("./process_checkpoint/sam2"):
-    snapshot_download(
-        repo_id="alexnasa/sam2_C_cpu",
-        local_dir="./process_checkpoint/sam2",
-    )
-    print("[init] SAM2 CPU model downloaded")
-
-# ── Download Wan2.2-Animate-14B (skip large unused files) ────────────────────
-if not os.path.exists("./Wan2.2-Animate-14B"):
-    snapshot_download(
-        repo_id="Wan-AI/Wan2.2-Animate-14B",
-        local_dir="./Wan2.2-Animate-14B",
-        ignore_patterns=[
-            "models_t5_*",
-            "google/*",
-            "tokenizer*",
-            "special_tokens_map.json",
-            "xlm-roberta-large/*",
-            "relighting_lora.ckpt",
-            "relighting_lora/*",
-            "process_checkpoint/sam2/*",
+
+        example_data = [
+            ['./examples/mov/1/1.jpeg', './examples/mov/1/1.mp4', 'wan2.2-animate-move', 'wan-pro'],
+            ['./examples/mov/2/2.jpeg', './examples/mov/2/2.mp4', 'wan2.2-animate-move', 'wan-pro'],
+            ['./examples/mix/1/1.jpeg', './examples/mix/1/1.mp4', 'wan2.2-animate-mix', 'wan-pro'],
+            ['./examples/mix/2/2.jpeg', './examples/mix/2/2.mp4', 'wan2.2-animate-mix', 'wan-pro']
         ]
-    )
-    print("[init] Wan2.2-Animate-14B downloaded")
-
-# ── Symlink SAM2 into model's expected path ───────────────────────────────────
-sam2_dst = "./Wan2.2-Animate-14B/process_checkpoint/sam2"
-sam2_src = "./process_checkpoint/sam2"
-if not os.path.exists(sam2_dst) and os.path.exists(sam2_src):
-    os.makedirs(os.path.dirname(sam2_dst), exist_ok=True)
-    os.symlink(os.path.abspath(sam2_src), sam2_dst)
-    print("[init] SAM2 symlink created")
-
-# ── Copy helper scripts ───────────────────────────────────────────────────────
-for fname in ["generate.py", "preprocess_data.py"]:
-    if os.path.exists(f"./{fname}") and not os.path.exists(f"./Wan2.2/{fname}"):
-        shutil.copy(f"./{fname}", f"./Wan2.2/{fname}")
-
-# ── Lazy model init ───────────────────────────────────────────────────────────
-_wan_animate = None
-
-def get_wan_animate():
-    global _wan_animate
-    if _wan_animate is None:
-        sys.path.insert(0, "./Wan2.2")
-        from generate import load_model
-        _wan_animate = load_model(False)
-    return _wan_animate
-
-# ── Inference ─────────────────────────────────────────────────────────────────
-@spaces.GPU(duration=300)
-def run_animate(ref_image, template_video, mode, quality, max_duration):
-    import uuid
-    from generate import generate
-
-    wan_animate = get_wan_animate()
-
-    uid = str(uuid.uuid4())[:8]
-    work_dir = f"/tmp/wan_{uid}"
-    os.makedirs(work_dir, exist_ok=True)
-
-    try:
-        ref_path  = os.path.join(work_dir, "ref.jpg")
-        tmpl_path = os.path.join(work_dir, "template.mp4")
-
-        import numpy as np
-        from PIL import Image
-        if isinstance(ref_image, np.ndarray):
-            Image.fromarray(ref_image).save(ref_path)
-        else:
-            shutil.copy(ref_image, ref_path)
-        shutil.copy(template_video, tmpl_path)
-
-        pose_path = os.path.join(work_dir, "pose.mp4")
-        face_path = os.path.join(work_dir, "face.png")
-        bg_path   = os.path.join(work_dir, "bg.png")
-        mask_path = os.path.join(work_dir, "mask.png")
-
-        from preprocess_data import preprocess
-        preprocess(
-            ref_image=ref_path,
-            template_video=tmpl_path,
-            output_pose=pose_path,
-            output_face=face_path,
-            output_bg=bg_path,
-            output_mask=mask_path,
-            mode=mode,
-        )
 
-        out_path = os.path.join(work_dir, "output.mp4")
-        generate(
-            wan_animate=wan_animate,
-            src_pose_path=pose_path,
-            src_face_path=face_path,
-            src_bg_path=bg_path,
-            src_mask_path=mask_path,
-            src_ref_path=ref_path,
-            save_file=out_path,
-        )
+        if example_data:
+            gr.Examples(
+                examples=example_data,
+                inputs=[ref_img, video, model_id, model],
+                outputs=[output_video, output_status],
+                fn=app.predict,
+                cache_examples="lazy",
+            )
 
-        return out_path, "Done!"
-    except Exception as e:
-        return None, f"Error: {e}"
-
-# ── UI ────────────────────────────────────────────────────────────────────────
-with gr.Blocks(title="Wan2.2 Animate") as demo:
-    gr.Markdown("## Wan2.2 Animate — ZeroGPU (Free A100)")
-    with gr.Row():
-        with gr.Column():
-            ref_image      = gr.Image(label="Reference Image", type="numpy")
-            template_video = gr.Video(label="Template Video")
-            mode           = gr.Dropdown(["normal", "tiktok"], value="normal", label="Mode")
-            quality        = gr.Dropdown(["standard", "high"], value="standard", label="Quality")
-            max_duration   = gr.Slider(1, 10, value=5, step=1, label="Max Duration (s)")
-            btn            = gr.Button("Generate", variant="primary")
-        with gr.Column():
-            out_video = gr.Video(label="Output Video")
-            status    = gr.Textbox(label="Status", interactive=False)
-
-    btn.click(
-        run_animate,
-        inputs=[ref_image, template_video, mode, quality, max_duration],
-        outputs=[out_video, status],
-    )
+        demo.queue(default_concurrency_limit=100)
+        demo.launch(
+            server_name="0.0.0.0",
+            server_port=7860
+        )
 
 if __name__ == "__main__":
-    demo.launch()
+    start_app()