Update app.py

app.py

@@ -1,69 +1,394 @@
+import copy
+import os
+import random
+
+os.system('pip install dashscope')
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs:
-https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("damo-vilab/modelscope-text-to-video-synthesis")
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-    
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-    
-    messages.append({"role": "user", "content": message})
-
-    # NOTE: Video models don't usually use "streaming" generation, so we'll just call once
-    payload = {
-        "inputs": message,
-        "parameters": {
-            "max_new_tokens": max_tokens,
-            "temperature": temperature,
-            "top_p": top_p,
-        }
+import dashscope
+from dashscope import VideoSynthesis
+from examples import t2v_examples, i2v_examples
+import time
+
+DASHSCOPE_API_KEY = os.getenv('DASHSCOPE_API_KEY')
+dashscope.api_key = DASHSCOPE_API_KEY
+
+KEEP_SUCCESS_TASK = 3600 * 10
+KEEP_RUNING_TASK = 3600 * 1
+# the total running task number in 1800 seconds
+LIMIT_RUNING_TASK = 10
+LIMIT_HISTORY_RUNING_TASK = 20
+FRESH_TIME = None
+
+task_status = {}
+
+total_task_info = {
+    "total_process_cost": 0,
+    "total_complete_task": 0,
+    "total_submit": 0,
+    "latest_1h_submit_status": {
+
     }
+}
+
+def get_submit_code():
+    submit_code = random.randint(0, 2147483647)
+    #
+    for sub_c, sub_info in copy.deepcopy(total_task_info["latest_1h_submit_status"]).items():
+        if time.time() - sub_info > 3600:
+            total_task_info["latest_1h_submit_status"].pop(sub_c)
+    return submit_code
+def t2v_generation(prompt, resolution, watermark_wan, seed=-1):
+    seed = seed if seed >= 0 else random.randint(0, 2147483647)
+    total_task_info["latest_1h_submit_status"][get_submit_code()] = time.time()
+    total_task_info["total_submit"] += 1
+    if not allow_task_num():
+        gr.Info(f"Warning: The number of running tasks is too large, the estimate waiting time is {get_waiting_time('-1')} s.")
+        return None, gr.Button(visible=True)
+    try:
+        rsp = VideoSynthesis.call(model="wanx2.1-t2v-plus", prompt=prompt, seed=seed,
+                                  watermark_wanx=watermark_wan, size=resolution)
+        video_url = rsp.output.video_url
+        return video_url, gr.Button(visible=True)
+    except Exception as e:
+        gr.Warning(f"Warning: {e}")
+        return None, gr.Button(visible=True)
+
+
+def t2v_generation_async(prompt, size, watermark_wan, seed=-1):
+    print(seed)
+    seed = seed if seed >= 0 else random.randint(0, 2147483647)
+    total_task_info["latest_1h_submit_status"][get_submit_code()] = time.time()
+    total_task_info["total_submit"] += 1
+    print(seed)
+    if not allow_task_num():
+        gr.Info(f"Warning: The number of running tasks is too large, the estimate waiting time is {get_waiting_time('-1')} s.")
+        return None, False, gr.Button(visible=True), gr.Button(visible=False), gr.Slider(), gr.Slider()
+    try:
+        rsp = VideoSynthesis.async_call(model="wanx2.1-t2v-plus",
+                                        prompt=prompt,
+                                        size=size,
+                                        seed=seed,
+                                        watermark_wanx=watermark_wan)
+        task_id = rsp.output.task_id
+        status = False
+        return task_id, status, gr.Button(visible=False), gr.Button(visible=True), get_cost_time(task_id), get_waiting_time(task_id)
+    except Exception as e:
+        gr.Warning(f"Warning: {e}")
+        return None, True, gr.Button(), gr.Button(),  gr.Slider(), gr.Slider()
+
+
+def i2v_generation(prompt, image, watermark_wan, seed=-1):
+    seed = seed if seed >= 0 else random.randint(0, 2147483647)
+    video_url = None
+    try:
+        rsp = VideoSynthesis.call(model="wanx2.1-i2v-plus", prompt=prompt, img_url=image,
+                                  seed=seed,
+                                  watermark_wanx=watermark_wan
+                                  )
+        video_url = rsp.output.video_url
+    except Exception as e:
+        gr.Warning(f"Warning: {e}")
+    return video_url
+
+
+def i2v_generation_async(prompt, image, watermark_wan, seed=-1):
+    seed = seed if seed >= 0 else random.randint(0, 2147483647)
+    total_task_info["latest_1h_submit_status"][get_submit_code()] = time.time()
+    total_task_info["total_submit"] += 1
+    if not allow_task_num():
+        gr.Info(f"Warning: The number of running tasks is too large, the estimate waiting time is {get_waiting_time('-1')} s.")
+        return "", None, gr.Button(visible=True), gr.Button(visible=False), gr.Slider(), gr.Slider()
+    try:
+        rsp = VideoSynthesis.async_call(model="wanx2.1-i2v-plus", prompt=prompt, seed=seed,
+                                        img_url=image, watermark_wanx=watermark_wan)
+        print(rsp)
+        task_id = rsp.output.task_id
+        status = False
+        return task_id, status, gr.Button(visible=False), gr.Button(visible=True), get_cost_time(task_id), get_waiting_time(task_id)
+    except Exception as e:
+        gr.Warning(f"Warning: {e}")
+        return "", None, gr.Button(), gr.Button(), gr.Slider(), gr.Slider()
+
+def get_result_with_task_id(task_id):
+    if task_id == "": return True, None
+    try:
+        rsp = VideoSynthesis.fetch(task=task_id)
+        print(rsp)
+        if rsp.output.task_status == "FAILED":
+            gr.Info(f"Warning: task running {rsp.output.task_status}")
+            status = True
+            video_url = None
+        else:
+            video_url = rsp.output.video_url
+            video_url = video_url if video_url != "" else None
+            status = video_url is not None
+        if status:
+            total_task_info["total_complete_task"] += 1
+            total_task_info["total_process_cost"] += time.time() - task_status[task_id]["time"]
+            print(total_task_info["total_complete_task"], total_task_info["total_process_cost"])
+    except:
+        video_url = None
+        status = False
+    return status, None if video_url == "" else video_url
+    # return True, "https://dashscope-result-wlcb.oss-cn-wulanchabu.aliyuncs.com/1d/f8/20250220/e7d3f375/ccc590a2-7e90-4d92-84bc-22668db42979.mp4?Expires=1740137152&OSSAccessKeyId=LTAI5tQZd8AEcZX6KZV4G8qL&Signature=i3S3jA5FY6XYfvzZNHnvQiPzZSw%3D"
+
+
+
+def allow_task_num():
+    num = 0
+    total_num = 0
+    for task_id in task_status:
+        if not task_status[task_id]["status"] and task_status[task_id]["time"] + 1800 > time.time():
+            num += 1
+        if not task_status[task_id]["status"]:
+            total_num += 1
+    return num < LIMIT_RUNING_TASK or total_num < LIMIT_HISTORY_RUNING_TASK
+
+def get_waiting_time(task_id):
+    # if the num of running task < Limit
+    # waiting time = num * 480s
+    # task_id not in task_status, return a large number
+    # prediction the waiting time
+    # avg_cost * latest submit time
+    num = 0
+    for task_id in task_status:
+        if not task_status[task_id]["status"]:
+            num += 1
+    latest_submit_tasks = len(total_task_info["latest_1h_submit_status"])
+    print("latest submit tasks", latest_submit_tasks)
+    if task_id in task_status:
+        return int(640 - (time.time() - task_status[task_id]["time"]))
+    else:
+        return int(latest_submit_tasks * (total_task_info["total_process_cost"]/(total_task_info["total_complete_task"]+1)))
+
+def online_get_waiting_time(task, t2v_task_id, i2v_task_id):
+    task_id = t2v_task_id if task == "t2v" else i2v_task_id
+    return get_waiting_time(task_id)
+
+def clean_task_status():
+    # clean the task over 1800 seconds
+    for task_id in copy.deepcopy(task_status):
+        if task_id == "": continue
+        # finished task, keep 3600 seconds
+        if task_status[task_id]["status"]:
+            if task_status[task_id]["time"] + KEEP_SUCCESS_TASK < time.time():
+                task_status.pop(task_id)
+        else:
+            # clean the task over 3600 * 2 seconds
+            if task_status[task_id]["time"] + KEEP_RUNING_TASK < time.time():
+                task_status.pop(task_id)
+
 
-    # Post directly to the model
-    response = client.post(json=payload)
+def get_cost_time(task_id):
+    if task_id in task_status and not task_status[task_id]["status"]:
+        et = int(time.time() - task_status[task_id]["time"])
+        return f"{et:.2f}"
+    else:
+        return gr.Textbox()
+
+def online_get_cost_time(task, t2v_task_id, i2v_task_id):
+    task_id = t2v_task_id if task == "t2v" else i2v_task_id
+    return get_cost_time(task_id)
 
-    video_url = response.get("video", None)
 
-    if video_url:
-        yield video_url
+def get_process_bar(task, t2v_task_id, i2v_task_id, status):
+    task_id = t2v_task_id if task == "t2v" else i2v_task_id
+    clean_task_status()
+    if task_id not in task_status:
+        task_status[task_id] = {
+            "value": 0 if not task_id == "" else 100,
+            "status": status if not task_id == "" else True,
+            "time": time.time(),
+            "url": None
+        }
+    if not task_status[task_id]["status"]:
+        # only when > 50% do check status
+        if task_status[task_id]["value"] >= 5 and task_status[task_id]["value"] % 5 == 0:
+            status, video_url = get_result_with_task_id(task_id)
+        else:
+            status, video_url = False, None
+        task_status[task_id]["status"] = status
+        task_status[task_id]["url"] = video_url
+    if task_status[task_id]["status"]:
+        task_status[task_id]["value"] = 100
     else:
-        yield "Failed to generate video."
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs:
-https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are generating a creative video.", label="System message"),
-        gr.Slider(minimum=1, maximum=1000, value=250, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=2.0, value=1.0, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.9,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
-
-if __name__ == "__main__":
-    demo.launch()
+        task_status[task_id]["value"] += 5
+    if task_status[task_id]["value"] >= 100 and not task_status[task_id]["status"]:
+        task_status[task_id]["value"] = 95
+    # print(task_id, task_status[task_id], task_status)
+    value = task_status[task_id]["value"]
+    return gr.Slider(label=f"({value}%)Generating" if value % 2 == 1 else f"({value}%)Generating.....", value=value)
+
+
+with gr.Blocks() as demo:
+    gr.HTML("""
+               <div style="text-align: center; font-size: 32px; font-weight: bold; margin-bottom: 20px;">
+                   Wan2.1: Open and Advanced Large-Scale Video Generative Models
+               </div>
+               <div style="text-align: center;">
+                   <a href="https://github.com/Wan-Video/Wan2.1">Code</a> |
+                   <a href="https://huggingface.co/Wan-AI">Huggingface</a> |
+                   <a href="https://modelscope.cn/organization/Wan-AI">Modelscope</a>
+               </div>
+                <div style="text-align: center; font-size: 16px; font-weight: bold; margin-bottom: 20px;">
+                   We are excited to announce that Wan's international experience page is officially live, supporting image and video generation, and it's completely free. We welcome you to try it out!
+                   <a href="https://wan.video/wanxiang/creation">Wan Web</a>
+               </div>
+               """)
+    t2v_task_id = gr.State(value="")
+    i2v_task_id = gr.State(value="")
+    status = gr.State(value=False)
+    task = gr.State(value="t2v")
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                with gr.Tabs():
+                    # Text to Video Tab
+                    with gr.TabItem("Text to Video") as t2v_tab:
+                        with gr.Row():
+                            txt2vid_prompt = gr.Textbox(
+                                label="Prompt",
+                                placeholder="Describe the video you want to generate",
+                                lines=19,
+                            )
+                        with gr.Row():
+                            resolution = gr.Dropdown(
+                                label="Resolution",
+                                choices=["1280*720", "960*960", "720*1280", "1088*832", "832*1088"],
+                                value="1280*720",
+                            )
+                        with gr.Row():
+                            run_t2v_button = gr.Button("Generate Video")
+                            t2v_refresh_status = gr.Button("Refresh Generating Status", visible=False)
+                    # Image to Video Tab
+                    with gr.TabItem("Image to Video") as i2v_tab:
+                        with gr.Row():
+                            with gr.Column():
+                                img2vid_image = gr.Image(
+                                    type="filepath",
+                                    label="Upload Input Image",
+                                    elem_id="image_upload",
+                                )
+                                img2vid_prompt = gr.Textbox(
+                                    label="Prompt",
+                                    placeholder="Describe the video you want to generate",
+                                    value="",
+                                    lines=5,
+                                )
+                        with gr.Row():
+                            run_i2v_button = gr.Button("Generate Video")
+                            i2v_refresh_status = gr.Button("Refresh Generating Status", visible=False)
+        with gr.Column():
+            with gr.Row():
+                result_gallery = gr.Video(label='Generated Video',
+                                          interactive=False,
+                                          height=500)
+            with gr.Row():
+                watermark_wan = gr.Checkbox(label="Watermark", value=True, visible=True, container=False)
+                seed = gr.Number(label="Seed", value=-1, container=True)
+                # cost_time = gr.Number(label="Cost Time(secs)", value=online_get_cost_time, interactive=False,
+                #                       every=FRESH_TIME, inputs=[task, t2v_task_id, i2v_task_id], container=True)
+                cost_time = gr.Number(label="Cost Time(secs)", value=0, interactive=False, container=True)
+                # waiting_time = gr.Number(label="Estimated Waiting Time(secs)", value=online_get_waiting_time, interactive=False,
+                #                       every=FRESH_TIME, inputs=[task, t2v_task_id, i2v_task_id], container=True)
+                waiting_time = gr.Number(label="Estimated Waiting Time(secs)", value=0, interactive=False, container=True)
+                # process_bar = gr.Slider(show_label=True, label="", value=get_process_bar, maximum=100,
+                #                         interactive=True, every=FRESH_TIME, inputs=[task, t2v_task_id, i2v_task_id, status], container=True)
+                process_bar = gr.Slider(show_label=True, label="", value=100, maximum=100,
+                                        interactive=True, container=True)
+    with gr.Row():
+        gr.Markdown('<span style="color: blue;">Due to automatic refresh of task status causing significant network congestion, please manually click the "Refresh Generating Status" button to check the task status.</span>')
+    fake_video = gr.Video(label='Examples', visible=False, interactive=False)
+    with gr.Row(visible=True) as t2v_eg:
+        gr.Examples(t2v_examples,
+                    inputs=[txt2vid_prompt, result_gallery],
+                    outputs=[result_gallery])
+
+    with gr.Row(visible=False) as i2v_eg:
+        gr.Examples(i2v_examples,
+                    inputs=[img2vid_prompt, img2vid_image, result_gallery],
+                    outputs=[result_gallery])
+
+
+    def process_change(task_id, task):
+        status = task_status.get(task_id, {"status":False})["status"]
+        if status:
+            video_url = task_status[task_id]["url"]
+            ret_t2v_btn = gr.Button(visible=True) if task == 't2v' else gr.Button()
+            ret_t2v_status_btn = gr.Button(visible=False) if task == 't2v' else gr.Button()
+            ret_i2v_btn = gr.Button(visible=True) if task == 'i2v' else gr.Button()
+            ret_i2v_status_btn = gr.Button(visible=False) if task == 'i2v' else gr.Button()
+            return gr.Video(value=video_url), ret_t2v_btn, ret_i2v_btn, ret_t2v_status_btn, ret_i2v_status_btn
+        return gr.Video(value=None), gr.Button(), gr.Button(), gr.Button(), gr.Button()
+
+    def online_process_change(task, t2v_task_id, i2v_task_id):
+        task_id = t2v_task_id if task == 't2v' else i2v_task_id
+        return process_change(task_id, task)
+
+    process_bar.change(online_process_change, inputs=[task, t2v_task_id, i2v_task_id],
+                       outputs=[result_gallery, run_t2v_button, run_i2v_button,
+                                t2v_refresh_status, i2v_refresh_status])
+
+
+    def switch_i2v_tab():
+        return gr.Row(visible=False), gr.Row(visible=True), "i2v"
+
+
+    def switch_t2v_tab():
+        return gr.Row(visible=True), gr.Row(visible=False), "t2v"
+
+
+    i2v_tab.select(switch_i2v_tab, outputs=[t2v_eg, i2v_eg, task])
+    t2v_tab.select(switch_t2v_tab, outputs=[t2v_eg, i2v_eg, task])
+
+    run_t2v_button.click(
+        fn=t2v_generation_async,
+        inputs=[txt2vid_prompt, resolution, watermark_wan, seed],
+        outputs=[t2v_task_id, status, run_t2v_button, t2v_refresh_status, cost_time, waiting_time],
+    )
+
+    run_i2v_button.click(
+        fn=i2v_generation_async,
+        inputs=[img2vid_prompt, img2vid_image, watermark_wan, seed],
+        outputs=[i2v_task_id, status, run_i2v_button, i2v_refresh_status, cost_time, waiting_time],
+    )
+
+    def status_refresh(task_id, task, status):
+        if task_id in task_status and not task_status[task_id]["status"]:
+            cost_time = int(time.time() - task_status[task_id]["time"])
+        else:
+            cost_time = 0
+        status, video_url = get_result_with_task_id(task_id)
+        if task_id not in task_status:
+            task_status[task_id] = {"status": status, "url": video_url, "time": time.time(), "value": 100 if status else 0}
+        else:
+            task_status[task_id]["status"] = status
+            task_status[task_id]["url"] = video_url
+        waiting_time = get_waiting_time(task_id)
+        value = task_status.get(task_id, {"value": 100})["value"]
+        value = max(value, int(cost_time*100/waiting_time))
+        task_status[task_id]["value"] = value if value < 100 else 100
+        if not video_url == "" and status: value = 100
+        process_bar = gr.Slider(label=f"({value}%)Generating" if value % 2 == 1 else f"({value}%)Generating.....", value=value)
+        process_change_ret = process_change(task_id, task)
+        return *process_change_ret, cost_time, waiting_time, process_bar
+
+
+    t2v_refresh_status.click(
+        fn=status_refresh,
+        inputs=[t2v_task_id, task, status],
+        outputs=[result_gallery, run_t2v_button, run_i2v_button,
+                t2v_refresh_status, i2v_refresh_status,
+                cost_time, waiting_time, process_bar]
+    )
+
+    i2v_refresh_status.click(
+        fn=status_refresh,
+        inputs=[i2v_task_id, task, status],
+        outputs=[result_gallery, run_t2v_button, run_i2v_button,
+                 t2v_refresh_status, i2v_refresh_status,
+                 cost_time, waiting_time, process_bar]
+    )
+
+#demo.queue(max_size=10)
+demo.launch(ssr_mode=False)