add video model

.idea/.gitignore

@@ -0,0 +1,8 @@
+# Default ignored files
+/shelf/
+/workspace.xml
+# Editor-based HTTP Client requests
+/httpRequests/
+# Datasource local storage ignored files
+/dataSources/
+/dataSources.local.xml

.idea/GenAI-Arena.iml

@@ -0,0 +1,15 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<module type="PYTHON_MODULE" version="4">
+  <component name="NewModuleRootManager">
+    <content url="file://$MODULE_DIR$" />
+    <orderEntry type="inheritedJdk" />
+    <orderEntry type="sourceFolder" forTests="false" />
+  </component>
+  <component name="PyDocumentationSettings">
+    <option name="format" value="GOOGLE" />
+    <option name="myDocStringFormat" value="Google" />
+  </component>
+  <component name="TemplatesService">
+    <option name="TEMPLATE_CONFIGURATION" value="Jinja2" />
+  </component>
+</module>

.idea/inspectionProfiles/profiles_settings.xml

@@ -0,0 +1,6 @@
+<component name="InspectionProjectProfileManager">
+  <settings>
+    <option name="USE_PROJECT_PROFILE" value="false" />
+    <version value="1.0" />
+  </settings>
+</component>

.idea/modules.xml

@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="ProjectModuleManager">
+    <modules>
+      <module fileurl="file://$PROJECT_DIR$/.idea/GenAI-Arena.iml" filepath="$PROJECT_DIR$/.idea/GenAI-Arena.iml" />
+    </modules>
+  </component>
+</project>

.idea/vcs.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="VcsDirectoryMappings">
+    <mapping directory="" vcs="Git" />
+  </component>
+</project>

app.py

@@ -2,6 +2,7 @@ import gradio as gr
 import os
 from serve.gradio_web import *
 from serve.gradio_web_image_editing import *
+from serve.gradio_web_video_generation import *
 from serve.leaderboard import build_leaderboard_tab
 from model.model_manager import ModelManager
 from pathlib import Path
@@ -47,6 +48,22 @@ def build_combine_demo(models, elo_results_file, leaderboard_table_file):
                     with gr.Tab("About Us", id=9):
                         build_about()
 
+            with gr.Tab("Video Generation", id=10):
+                with gr.Tabs() as tabs_vg:
+                    with gr.Tab("Video Generation Arena (battle)", id=10):
+                        build_side_by_side_ui_anony_vg(models)
+
+                    with gr.Tab("Video Generation Arena (side-by-side)", id=11):
+                        build_side_by_side_ui_named_vg(models)
+
+                    with gr.Tab("Video Generation Direct Chat", id=12):
+                        build_single_model_ui_vg(models, add_promotion_links=True)
+                    if elo_results_file and 'video_generation' in elo_results_file:
+                        with gr.Tab("Video Generation Leaderboard", id=13):
+                            build_leaderboard_tab(elo_results_file['video_generation'], leaderboard_table_file['video_generation'])
+                    with gr.Tab("About Us", id=14):
+                        build_about()
+
     return demo
 
 
@@ -63,6 +80,8 @@ def load_elo_results(elo_results_dir):
                 elo_results_file['t2i_generation'] = file
             elif 'image_editing' in file.name:
                 elo_results_file['image_editing'] = file
+            elif 'video_generation' in file.name:
+                elo_results_file['video_generation'] = file
             else:
                 raise ValueError(f"Unknown file name: {file.name}")
         for file in elo_results_dir.glob('*_leaderboard.csv'):
@@ -70,6 +89,8 @@ def load_elo_results(elo_results_dir):
                 leaderboard_table_file['t2i_generation'] = file
             elif 'image_editing' in file.name:
                 leaderboard_table_file['image_editing'] = file
+            elif 'video_generation' in file.name:
+                leaderboard_table_file['video_generation'] = file
             else:
                 raise ValueError(f"Unknown file name: {file.name}")
             

model/model_manager.py

@@ -5,12 +5,13 @@ import requests
 import io, base64, json
 import spaces
 from PIL import Image
-from .models import IMAGE_GENERATION_MODELS, IMAGE_EDITION_MODELS, load_pipeline
+from .models import IMAGE_GENERATION_MODELS, IMAGE_EDITION_MODELS, VIDEO_GENERATION_MODELS, load_pipeline
 
 class ModelManager:
     def __init__(self):
         self.model_ig_list = IMAGE_GENERATION_MODELS
         self.model_ie_list = IMAGE_EDITION_MODELS
+        self.model_vg_list = VIDEO_GENERATION_MODELS
         self.loaded_models = {}
 
     def load_model_pipe(self, model_name):
@@ -79,4 +80,34 @@ class ModelManager:
             for future in concurrent.futures.as_completed(future_to_result):
                 result = future.result()
                 results.append(result)
-        return results[0], results[1], model_names[0], model_names[1]
+        return results[0], results[1], model_names[0], model_names[1]
+
+    @spaces.GPU(duration=150)
+    def generate_video_vg(self, prompt, model_name):
+        pipe = self.load_model_pipe(model_name)
+        result = pipe(prompt=prompt)
+        return result
+
+    def generate_video_vg_parallel_anony(self, prompt, model_A, model_B):
+        if model_A == "" and model_B == "":
+            model_names = random.sample([model for model in self.model_vg_list], 2)
+        else:
+            model_names = [model_A, model_B]
+
+        results = []
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            future_to_result = {executor.submit(self.generate_video_vg, prompt, model): model for model in model_names}
+            for future in concurrent.futures.as_completed(future_to_result):
+                result = future.result()
+                results.append(result)
+        return results[0], results[1], model_names[0], model_names[1]
+
+    def generate_video_vg_parallel(self, prompt, model_A, model_B):
+        results = []
+        model_names = [model_A, model_B]
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            future_to_result = {executor.submit(self.generate_video_vg, prompt, model): model for model in model_names}
+            for future in concurrent.futures.as_completed(future_to_result):
+                result = future.result()
+                results.append(result)
+        return results[0], results[1]

model/model_registry.py

@@ -159,15 +159,30 @@ register_model_info(
 )
 
 register_model_info(
-    ["fal_stable_cascade"],
+    ["fal_stable-cascade_text2image"],
     "StableCascade",
     "https://fal.ai/models/stable-cascade/api",
     "StableCascade is a generative model that can generate high-quality images from text prompts.",
 )
+
+register_model_info(
+    ["fal_fast-animatediff/text-to-video_text2video"],
+    "AnimateDiff",
+    "https://fal.ai/models/fast-animatediff-t2v",
+    "AnimateDiff is a text-driven models that produce diverse and personalized animated images.",
+)
+
+register_model_info(
+    ["fal_fast-animatediff/turbo/text-to-video_text2video"],
+    "AnimateDiff Turbo",
+    "https://fal.ai/models/fast-animatediff-t2v-turbo",
+    "AnimateDiff Turbo is a lightning version of AnimateDiff.",
+)
     
 
 models = ['imagenhub_LCM_generation','imagenhub_SDXLTurbo_generation','imagenhub_SDXL_generation',
-              'imagenhub_OpenJourney_generation','imagenhub_PixArtAlpha_generation','imagenhub_SDXLLightning_generation',
-              'imagenhub_StableCascade_generation','imagenhub_PlaygroundV2_generation', 'fal_Playground-v25_generation', 'fal_stable_cascade',
-              'imagenhub_CycleDiffusion_edition', 'imagenhub_Pix2PixZero_edition', 'imagenhub_Prompt2prompt_edition',
-              'imagenhub_SDEdit_edition', 'imagenhub_InstructPix2Pix_edition', 'imagenhub_MagicBrush_edition', 'imagenhub_PNP_edition']
+          'imagenhub_OpenJourney_generation','imagenhub_PixArtAlpha_generation','imagenhub_SDXLLightning_generation',
+          'imagenhub_StableCascade_generation','imagenhub_PlaygroundV2_generation', 'fal_Playground-v25_generation', 'fal_stable-cascade_text2image',
+          'imagenhub_CycleDiffusion_edition', 'imagenhub_Pix2PixZero_edition', 'imagenhub_Prompt2prompt_edition',
+          'imagenhub_SDEdit_edition', 'imagenhub_InstructPix2Pix_edition', 'imagenhub_MagicBrush_edition', 'imagenhub_PNP_edition'
+          "fal_fast-animatediff/turbo/text-to-video_text2video", "fal_fast-animatediff/text-to-video_text2video"]

model/models/__init__.py

@@ -1,12 +1,14 @@
 from .imagenhub_models import load_imagenhub_model
 from .playground_api import load_playground_model
-# from .fal_api_models import load_fal_model
+from .fal_api_models import load_fal_model
 
 IMAGE_GENERATION_MODELS = ['imagenhub_LCM_generation','imagenhub_SDXLTurbo_generation','imagenhub_SDXL_generation', 'imagenhub_PixArtAlpha_generation',
                             'imagenhub_OpenJourney_generation','imagenhub_SDXLLightning_generation', 'imagenhub_StableCascade_generation',
                             'playground_PlayGroundV2_generation', 'playground_PlayGroundV2.5_generation']
 IMAGE_EDITION_MODELS = ['imagenhub_CycleDiffusion_edition', 'imagenhub_Pix2PixZero_edition', 'imagenhub_Prompt2prompt_edition',
                         'imagenhub_SDEdit_edition', 'imagenhub_InstructPix2Pix_edition', 'imagenhub_MagicBrush_edition', 'imagenhub_PNP_edition']
+VIDEO_GENERATION_MODELS = ['fal_fast-animatediff/text-to-video_text2video',
+                           'fal_fast-animatediff/turbo/text-to-video_text2video']
 
 
 def load_pipeline(model_name):
@@ -24,7 +26,7 @@ def load_pipeline(model_name):
     elif model_source == "playground":
         pipe = load_playground_model(model_name)
     elif model_source == "fal":
-        raise NotImplementedError("FAL models are not supported yet")
+        pipe = load_fal_model(model_name, model_type)
     else:
         raise ValueError(f"Model source {model_source} not supported")
     return pipe

model/models/fal_api_models.py

@@ -1,49 +1,73 @@
-import fal
+import fal_client
+from PIL import Image
+import requests
+import io
+import os
+
 
 class FalModel():
     def __init__(self, model_name, model_type):
         self.model_name = model_name
-        self.modle_type = model_type
+        self.model_type = model_type
+        os.environ['FAL_KEY'] = os.environ['FalAPI']
     
     def __call__(self, *args, **kwargs):
-        
         if self.model_type == "text2image":
             assert "prompt" in kwargs, "prompt is required for text2image model"
-            handler = fal.apps.submit(
+            handler = fal_client.submit(
                 f"fal-ai/{self.model_name}",
                 arguments={
                     "prompt": kwargs["prompt"]
                 },
             )
-            
-            for event in handler.iter_events():
-                if isinstance(event, fal.apps.InProgress):
+            for event in handler.iter_events(with_logs=True):
+                if isinstance(event, fal_client.InProgress):
                     print('Request in progress')
                     print(event.logs)
-
             result = handler.get()
+            result_url = result['images'][0]['url']
+            response = requests.get(result_url)
+            result = Image.open(io.BytesIO(response.content))
             return result
         elif self.model_type == "image2image":
-            assert "image" in kwargs or "image_url" in kwargs, "image or image_url is required for image2image model"
-            if "image" in kwargs:
-                image_url = None
-                pass
-            handler = fal.apps.submit(
+            raise NotImplementedError("image2image model is not implemented yet")
+            # assert "image" in kwargs or "image_url" in kwargs, "image or image_url is required for image2image model"
+            # if "image" in kwargs:
+            #     image_url = None
+            #     pass
+            # handler = fal_client.submit(
+            #     f"fal-ai/{self.model_name}",
+            #     arguments={
+            #         "image_url": image_url
+            #     },
+            # )
+            #
+            # for event in handler.iter_events():
+            #     if isinstance(event, fal_client.InProgress):
+            #         print('Request in progress')
+            #         print(event.logs)
+            #
+            # result = handler.get()
+            # return result
+        elif self.model_type == "text2video":
+            assert "prompt" in kwargs, "prompt is required for text2video model"
+            handler = fal_client.submit(
                 f"fal-ai/{self.model_name}",
                 arguments={
-                    "image_url": image_url
+                    "prompt": kwargs["prompt"]
                 },
             )
-            
-            for event in handler.iter_events():
-                if isinstance(event, fal.apps.InProgress):
+
+            for event in handler.iter_events(with_logs=True):
+                if isinstance(event, fal_client.InProgress):
                     print('Request in progress')
                     print(event.logs)
 
             result = handler.get()
-            return result
-        elif self.model_type == "text2video":
-            raise NotImplementedError("text2video model is not implemented yet")
+            print("result video: ====")
+            print(result)
+            result_url = result['video']['url']
+            return result_url
         else:
             raise ValueError("model_type must be text2image or image2image")
 

requirements.txt

@@ -48,3 +48,5 @@ krippendorff
 statsmodels
 plotly
 -e git+https://github.com/TIGER-AI-Lab/ImagenHub.git#egg=imagen-hub
+fal_client
+

serve/constants.py

@@ -2,6 +2,7 @@ import os
 
 LOGDIR = os.getenv("LOGDIR", "./GenAI-Arena-hf-logs/vote_log")
 IMAGE_DIR = os.getenv("IMAGE_DIR", f"{LOGDIR}/images")
+VIDEO_DIR = os.getenv("VIDEO_DIR", f"{LOGDIR}/videos")
 
 SERVER_PORT = os.getenv("SERVER_PORT", 7860)
 ROOT_PATH = os.getenv("ROOT_PATH", None)
@@ -13,5 +14,7 @@ LOG_SERVER_ADDR = os.getenv("LOG_SERVER_ADDR", f"{LOG_SERVER}{LOG_SERVER_SUBDOMA
 # LOG SERVER API ENDPOINTS
 APPEND_JSON = "append_json"
 SAVE_IMAGE = "save_image"
+SAVE_VIDEO = "save_video"
 SAVE_LOG = "save_log"
 
+

serve/gradio_web.py

@@ -82,7 +82,7 @@ Find out who is the 🥇conditional image generation models! More models are goi
     with gr.Row():
         clear_btn = gr.Button(value="🎲 New Round", interactive=False)
         regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
-        share_btn = gr.Button(value="📷  Share")
+        # share_btn = gr.Button(value="📷  Share")
 
     gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
 
@@ -162,12 +162,12 @@ Find out who is the 🥇conditional image generation models! More models are goi
         outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
     )
 
-    share_btn.click(
-        share_click,
-        inputs=[state0, state1, model_selector_left, model_selector_right],
-        outputs=[],
-        js=share_js
-    )
+    # share_btn.click(
+    #     share_click,
+    #     inputs=[state0, state1, model_selector_left, model_selector_right],
+    #     outputs=[],
+    #     js=share_js
+    # )
     
     
 def build_side_by_side_ui_named(models):
@@ -241,7 +241,7 @@ def build_side_by_side_ui_named(models):
     with gr.Row():
         clear_btn = gr.Button(value="🗑️  Clear history", interactive=False)
         regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
-        share_btn = gr.Button(value="📷  Share")
+        # share_btn = gr.Button(value="📷  Share")
 
     gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
 
@@ -322,12 +322,12 @@ def build_side_by_side_ui_named(models):
         outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
     )
 
-    share_btn.click(
-        share_click,
-        inputs=[state0, state1, model_selector_left, model_selector_right],
-        outputs=[],
-        js=share_js
-    )
+    # share_btn.click(
+    #     share_click,
+    #     inputs=[state0, state1, model_selector_left, model_selector_right],
+    #     outputs=[],
+    #     js=share_js
+    # )
 
 def build_single_model_ui(models, add_promotion_links=False):
     promotion = (

serve/gradio_web_image_editing.py

@@ -97,7 +97,7 @@ Find out who is the 🥇conditional image edition models!
     with gr.Row() as button_row:
         clear_btn = gr.Button(value="🎲 New Round", interactive=False)
         regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
-        share_btn = gr.Button(value="📷  Share")
+        # share_btn = gr.Button(value="📷  Share")
 
     gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
 
@@ -194,12 +194,7 @@ Find out who is the 🥇conditional image edition models!
         inputs=[state0, state1, dummy_left_model, dummy_right_model],
         outputs=[model_selector_left, model_selector_right] + input_list + vote_btns,
     )
-    share_btn.click(
-        share_click,
-        inputs=[state0, state1, model_selector_left, model_selector_right],
-        outputs=[],
-        js=share_js
-    )
+
 
 
     
@@ -292,7 +287,7 @@ def build_side_by_side_ui_named_ie(models):
     with gr.Row() as button_row:
         clear_btn = gr.Button(value="🗑️  Clear history", interactive=False)
         regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
-        share_btn = gr.Button(value="📷  Share")
+        # share_btn = gr.Button(value="📷  Share")
 
     gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
     gr.Examples(
@@ -393,12 +388,6 @@ def build_side_by_side_ui_named_ie(models):
         outputs=[model_selector_left, model_selector_right] + input_list + vote_btns,
     )
 
-    share_btn.click(
-        share_click,
-        inputs=[state0, state1, model_selector_left, model_selector_right],
-        outputs=[],
-        js=share_js
-    )
 
 
 def build_single_model_ui_ie(models, add_promotion_links=False):

serve/gradio_web_video_generation.py

@@ -0,0 +1,482 @@
+from .utils import *
+from .vote_utils import (
+    upvote_last_response_vg as upvote_last_response,
+    downvote_last_response_vg as downvote_last_response,
+    flag_last_response_vg as flag_last_response,
+    leftvote_last_response_vgm as leftvote_last_response,
+    rightvote_last_response_vgm as rightvote_last_response,
+    tievote_last_response_vgm as tievote_last_response,
+    bothbad_vote_last_response_vgm as bothbad_vote_last_response,
+    # share_click_vgm as share_click,
+    generate_vg,
+    generate_vgm,
+    generate_vgm_annoy,
+    share_js
+)
+from functools import partial
+
+
+def build_side_by_side_ui_anony_vg(models):
+    notice_markdown = """
+# ⚔️  GenAI-Arena ⚔️ : Benchmarking Visual Generative Models in the Wild
+| [GitHub](https://github.com/TIGER-AI-Lab/ImagenHub) | [Paper](https://arxiv.org/abs/2310.01596) | [Dataset](https://huggingface.co/ImagenHub) | 
+## 📜 Rules
+- Input prompt to two anonymous models in same area (e.g., AnimateDiff, AnimateDiff-Turbo) and vote for the better one!
+- When the results are ready, click the button below to vote.
+- Vote won't be counted if model identity is revealed during conversation.
+- Click "New Round" to start a new round.
+
+## 🏆 Arena Elo 
+Find out who is the 🥇conditional video generation models! More models are going to be supported.
+
+## 👇 Generating now!
+
+"""
+
+    model_list = models.model_vg_list
+
+    state0 = gr.State()
+    state1 = gr.State()
+    gen_func = partial(generate_vgm_annoy, models.generate_video_vg_parallel_anony)
+
+    gr.Markdown(notice_markdown, elem_id="notice_markdown")
+
+    with gr.Group(elem_id="share-region-anony"):
+        with gr.Accordion("🔍 Expand to see all Arena players", open=False):
+            model_description_md = get_model_description_md(model_list)
+            gr.Markdown(model_description_md, elem_id="model_description_markdown")
+        with gr.Row():
+            with gr.Column():
+                chatbot_left = gr.Video(width=512, label="Model A", autoplay=True)
+            with gr.Column():
+                chatbot_right = gr.Video(width=512, label="Model B", autoplay=True)
+
+        with gr.Row():
+            with gr.Column():
+                model_selector_left = gr.Markdown("", visible=False)
+            with gr.Column():
+                model_selector_right = gr.Markdown("", visible=False)
+        with gr.Row():
+            slow_warning = gr.Markdown("", elem_id="notice_markdown")
+
+        with gr.Row():
+            leftvote_btn = gr.Button(
+                value="👈  A is better", visible=False, interactive=False
+            )
+            rightvote_btn = gr.Button(
+                value="👉  B is better", visible=False, interactive=False
+            )
+            tie_btn = gr.Button(value="🤝  Tie", visible=False, interactive=False)
+            bothbad_btn = gr.Button(
+                value="👎  Both are bad", visible=False, interactive=False
+            )
+
+    with gr.Row():
+        textbox = gr.Textbox(
+            show_label=False,
+            placeholder="👉 Enter your prompt and press ENTER",
+            container=True,
+            elem_id="input_box",
+        )
+        send_btn = gr.Button(value="Send", variant="primary", scale=0)
+
+    with gr.Row():
+        clear_btn = gr.Button(value="🎲 New Round", interactive=False)
+        regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
+        # share_btn = gr.Button(value="📷  Share")
+
+    gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
+
+    dummy_video_output = gr.Video(width=512, visible=False)
+    gr.Examples(
+        examples=[["a cute dog is playing a ball", 'https://fal-cdn.batuhan-941.workers.dev/files/lion/BFcM30vpeyRHFfdhbfFoY.mp4'],
+                  ["Buildings on fire, old film still", 'https://fal-cdn.batuhan-941.workers.dev/files/zebra/r4NeCgXTkpzIBCF0DDHqJ.mp4'],
+                  ["A serene underwater scene featuring a sea turtle swimming through a coral reef.",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/tiger/cBHesvgUzjonNUEL1FQGw.mp4'],
+                  ["A futuristic hopeful busy city, purple and green color scheme",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/monkey/5-vUtY4_bHAhTtevZx16K.mp4']],
+        inputs=[textbox, dummy_video_output])
+
+    btn_list = [leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, regenerate_btn, clear_btn, ]
+
+    textbox.submit(
+        gen_func,
+        inputs=[state0, state1, textbox, model_selector_left, model_selector_right],
+        outputs=[state0, state1, chatbot_left, chatbot_right, model_selector_left, model_selector_right],
+        api_name="submit_btn_annony"
+    ).then(
+        enable_buttons_side_by_side,
+        inputs=None,
+        outputs=btn_list
+    )
+    send_btn.click(
+        gen_func,
+        inputs=[state0, state1, textbox, model_selector_left, model_selector_right],
+        outputs=[state0, state1, chatbot_left, chatbot_right, model_selector_left, model_selector_right],
+        api_name="send_btn_annony"
+    ).then(
+        enable_buttons_side_by_side,
+        inputs=None,
+        outputs=btn_list
+    )
+
+    clear_btn.click(
+        clear_history_side_by_side_anony,
+        inputs=None,
+        outputs=[state0, state1, textbox, chatbot_left, chatbot_right, model_selector_left, model_selector_right],
+        api_name="clear_btn_annony"
+    ).then(
+        disable_buttons_side_by_side,
+        inputs=None,
+        outputs=btn_list
+    )
+
+    regenerate_btn.click(
+        gen_func,
+        inputs=[state0, state1, textbox, model_selector_left, model_selector_right],
+        outputs=[state0, state1, chatbot_left, chatbot_right, model_selector_left, model_selector_right],
+        api_name="regenerate_btn_annony"
+    ).then(
+        enable_buttons_side_by_side,
+        inputs=None,
+        outputs=btn_list
+    )
+    dummy_left_model = gr.State("")
+    dummy_right_model = gr.State("")
+    leftvote_btn.click(
+        leftvote_last_response,
+        inputs=[state0, state1, dummy_left_model, dummy_right_model],
+        outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
+    )
+    rightvote_btn.click(
+        rightvote_last_response,
+        inputs=[state0, state1, dummy_left_model, dummy_right_model],
+        outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
+    )
+    tie_btn.click(
+        tievote_last_response,
+        inputs=[state0, state1, dummy_left_model, dummy_right_model],
+        outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
+    )
+    bothbad_btn.click(
+        bothbad_vote_last_response,
+        inputs=[state0, state1, dummy_left_model, dummy_right_model],
+        outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
+    )
+
+    # share_btn.click(
+    #     share_click,
+    #     inputs=[state0, state1, model_selector_left, model_selector_right],
+    #     outputs=[],
+    #     js=share_js
+    # )
+
+
+def build_side_by_side_ui_named_vg(models):
+    notice_markdown = """
+# ⚔️  GenAI-Arena ⚔️ : Benchmarking Visual Generative Models in the Wild
+| [GitHub](https://github.com/TIGER-AI-Lab/ImagenHub) | [Paper](https://arxiv.org/abs/2310.01596) | [Dataset](https://huggingface.co/ImagenHub) |
+
+## 📜 Rules
+- Generate with any two selected models side-by-side and vote!
+- Input prompt you want to generate.
+- Click "Send" to submit the prompt.
+- Click "Clear history" to start a new round.
+
+## 🤖 Choose two models to compare
+"""
+    model_list = models.model_vg_list
+
+    state0 = gr.State()
+    state1 = gr.State()
+    gen_func = partial(generate_vgm, models.generate_video_vg_parallel)
+    gr.Markdown(notice_markdown, elem_id="notice_markdown")
+
+    with gr.Group(elem_id="share-region-named"):
+        with gr.Row():
+            with gr.Column():
+                model_selector_left = gr.Dropdown(
+                    choices=model_list,
+                    value=model_list[0] if len(model_list) > 0 else "",
+                    interactive=True,
+                    show_label=False,
+                    container=False,
+                )
+            with gr.Column():
+                model_selector_right = gr.Dropdown(
+                    choices=model_list,
+                    value=model_list[1] if len(model_list) > 1 else "",
+                    interactive=True,
+                    show_label=False,
+                    container=False,
+                )
+        with gr.Row():
+            with gr.Accordion("🔍 Expand to see all model descriptions", open=False):
+                model_description_md = get_model_description_md(model_list)
+                gr.Markdown(model_description_md, elem_id="model_description_markdown")
+
+        with gr.Row():
+            with gr.Column():
+                chatbot_left = gr.Video(width=512, label="Model A", autoplay=True)
+            with gr.Column():
+                chatbot_right = gr.Video(width=512, label="Model B", autoplay=True)
+        with gr.Row():
+            leftvote_btn = gr.Button(
+                value="👈  A is better", visible=False, interactive=False
+            )
+            rightvote_btn = gr.Button(
+                value="👉  B is better", visible=False, interactive=False
+            )
+            tie_btn = gr.Button(value="🤝  Tie", visible=False, interactive=False)
+            bothbad_btn = gr.Button(
+                value="👎  Both are bad", visible=False, interactive=False
+            )
+
+    with gr.Row():
+        textbox = gr.Textbox(
+            show_label=False,
+            placeholder="👉 Enter your prompt and press ENTER",
+            elem_id="input_box"
+        )
+        send_btn = gr.Button(value="Send", variant="primary", scale=0)
+
+    with gr.Row():
+        clear_btn = gr.Button(value="🗑️  Clear history", interactive=False)
+        regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
+        # share_btn = gr.Button(value="📷  Share")
+
+    gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
+
+    dummy_video_output = gr.Video(width=512, visible=False)
+    gr.Examples(
+        examples=[["a cute dog is playing a ball",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/lion/BFcM30vpeyRHFfdhbfFoY.mp4'],
+                  ["Buildings on fire, old film still",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/zebra/r4NeCgXTkpzIBCF0DDHqJ.mp4'],
+                  ["A serene underwater scene featuring a sea turtle swimming through a coral reef.",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/tiger/cBHesvgUzjonNUEL1FQGw.mp4'],
+                  ["A futuristic hopeful busy city, purple and green color scheme",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/monkey/5-vUtY4_bHAhTtevZx16K.mp4']],
+        inputs=[textbox, dummy_video_output])
+
+    model_selector_left.change(clear_history_side_by_side, inputs=None,
+                               outputs=[state0, state1, textbox, chatbot_left, chatbot_right],
+                               api_name="model_selector_left_side_by_side")
+    model_selector_right.change(clear_history_side_by_side, inputs=None,
+                                outputs=[state0, state1, textbox, chatbot_left, chatbot_right],
+                                api_name="model_selector_right_side_by_side")
+
+    btn_list = [leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, regenerate_btn, clear_btn]
+
+    textbox.submit(
+        gen_func,
+        inputs=[state0, state1, textbox, model_selector_left, model_selector_right],
+        outputs=[state0, state1, chatbot_left, chatbot_right],
+        api_name="textbox_side_by_side"
+    ).then(
+        enable_buttons_side_by_side,
+        inputs=None,
+        outputs=btn_list
+    )
+
+    send_btn.click(
+        gen_func,
+        inputs=[state0, state1, textbox, model_selector_left, model_selector_right],
+        outputs=[state0, state1, chatbot_left, chatbot_right],
+        api_name="send_side_by_side"
+    ).then(
+        enable_buttons_side_by_side,
+        inputs=None,
+        outputs=btn_list
+    )
+    regenerate_btn.click(
+        gen_func,
+        inputs=[state0, state1, textbox, model_selector_left, model_selector_right],
+        outputs=[state0, state1, chatbot_left, chatbot_right],
+        api_name="regenerate_side_by_side"
+    ).then(
+        enable_buttons_side_by_side,
+        inputs=None,
+        outputs=btn_list
+    )
+
+    clear_btn.click(
+        clear_history_side_by_side,
+        inputs=None,
+        outputs=[state0, state1, textbox, chatbot_left, chatbot_right],
+        api_name="clear_btn_side_by_side"
+    ).then(
+        disable_buttons_side_by_side,
+        inputs=None,
+        outputs=btn_list
+    )
+
+    leftvote_btn.click(
+        leftvote_last_response,
+        inputs=[state0, state1, model_selector_left, model_selector_right],
+        outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
+    )
+    rightvote_btn.click(
+        rightvote_last_response,
+        inputs=[state0, state1, model_selector_left, model_selector_right],
+        outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
+    )
+    tie_btn.click(
+        tievote_last_response,
+        inputs=[state0, state1, model_selector_left, model_selector_right],
+        outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
+    )
+    bothbad_btn.click(
+        bothbad_vote_last_response,
+        inputs=[state0, state1, model_selector_left, model_selector_right],
+        outputs=[textbox, leftvote_btn, rightvote_btn, tie_btn, bothbad_btn, model_selector_left, model_selector_right]
+    )
+
+    # share_btn.click(
+    #     share_click,
+    #     inputs=[state0, state1, model_selector_left, model_selector_right],
+    #     outputs=[],
+    #     js=share_js
+    # )
+
+
+def build_single_model_ui_vg(models, add_promotion_links=False):
+    promotion = (
+        """
+- | [GitHub](https://github.com/TIGER-AI-Lab/ImagenHub) | [Paper](https://arxiv.org/abs/2310.01596) | [Dataset](https://huggingface.co/ImagenHub) |
+"""
+        if add_promotion_links
+        else ""
+    )
+
+    notice_markdown = f"""
+# 🏔️ Play with Video Generation Models
+{promotion}
+
+## 🤖 Choose any model to generate
+"""
+
+    state = gr.State()
+    gen_func = partial(generate_vg, models.generate_video_vg)
+    gr.Markdown(notice_markdown, elem_id="notice_markdown")
+
+    model_list = models.model_vg_list
+
+    with gr.Row(elem_id="model_selector_row"):
+        model_selector = gr.Dropdown(
+            choices=model_list,
+            value=model_list[0] if len(model_list) > 0 else "",
+            interactive=True,
+            show_label=False
+        )
+
+    with gr.Row():
+        with gr.Accordion(
+                "🔍 Expand to see all model descriptions",
+                open=False,
+                elem_id="model_description_accordion",
+        ):
+            model_description_md = get_model_description_md(model_list)
+            gr.Markdown(model_description_md, elem_id="model_description_markdown")
+
+    with gr.Row():
+        textbox = gr.Textbox(
+            show_label=False,
+            placeholder="👉 Enter your prompt and press ENTER",
+            elem_id="input_box"
+        )
+
+        send_btn = gr.Button(value="Send", variant="primary", scale=0)
+
+    with gr.Row():
+        chatbot = gr.Video(width=512, autoplay=True)
+
+    with gr.Row() as button_row:
+        upvote_btn = gr.Button(value="👍  Upvote", interactive=False)
+        downvote_btn = gr.Button(value="👎  Downvote", interactive=False)
+        flag_btn = gr.Button(value="⚠️  Flag", interactive=False)
+        regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
+        clear_btn = gr.Button(value="🗑️  Clear history", interactive=False)
+
+    if add_promotion_links:
+        gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
+
+    dummy_video_output = gr.Video(width=512, visible=False)
+    gr.Examples(
+        examples=[["a cute dog is playing a ball",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/lion/BFcM30vpeyRHFfdhbfFoY.mp4'],
+                  ["Buildings on fire, old film still",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/zebra/r4NeCgXTkpzIBCF0DDHqJ.mp4'],
+                  ["A serene underwater scene featuring a sea turtle swimming through a coral reef.",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/tiger/cBHesvgUzjonNUEL1FQGw.mp4'],
+                  ["A futuristic hopeful busy city, purple and green color scheme",
+                   'https://fal-cdn.batuhan-941.workers.dev/files/monkey/5-vUtY4_bHAhTtevZx16K.mp4']],
+        inputs=[textbox, dummy_video_output])
+
+    model_selector.change(clear_history, inputs=None, outputs=[state, textbox, chatbot],
+                          api_name="model_selector_single")
+
+    btn_list = [upvote_btn, downvote_btn, flag_btn, regenerate_btn, clear_btn]
+
+    textbox.submit(
+        gen_func,
+        inputs=[state, textbox, model_selector],
+        outputs=[state, chatbot],
+        api_name="submit_btn_single",
+        show_progress="full"
+    ).success(
+        enable_buttons,
+        inputs=None,
+        outputs=btn_list
+    )
+    send_btn.click(
+        gen_func,
+        inputs=[state, textbox, model_selector],
+        outputs=[state, chatbot],
+        api_name="send_btn_single",
+        show_progress="full"
+    ).success(
+        enable_buttons,
+        inputs=None,
+        outputs=btn_list
+    )
+
+    upvote_btn.click(
+        upvote_last_response,
+        inputs=[state, model_selector],
+        outputs=[textbox, upvote_btn, downvote_btn, flag_btn]
+    )
+
+    downvote_btn.click(
+        downvote_last_response,
+        inputs=[state, model_selector],
+        outputs=[textbox, upvote_btn, downvote_btn, flag_btn]
+    )
+    flag_btn.click(
+        flag_last_response,
+        inputs=[state, model_selector],
+        outputs=[textbox, upvote_btn, downvote_btn, flag_btn]
+    )
+    regenerate_btn.click(
+        gen_func,
+        inputs=[state, textbox, model_selector],
+        outputs=[state, chatbot],
+        api_name="regenerate_btn_single",
+        show_progress="full"
+    ).success(
+        enable_buttons,
+        inputs=None,
+        outputs=btn_list
+    )
+    clear_btn.click(
+        clear_history,
+        inputs=None,
+        outputs=[state, textbox, chatbot],
+        api_name="clear_history_single",
+        show_progress="full"
+    ).then(
+        disable_buttons,
+        inputs=None,
+        outputs=btn_list
+    )
+

serve/log_server.py

@@ -4,7 +4,7 @@ import json
 import os
 import aiofiles
 from .log_utils import build_logger
-from .constants import LOG_SERVER_SUBDOMAIN, APPEND_JSON, SAVE_IMAGE, SAVE_LOG
+from .constants import LOG_SERVER_SUBDOMAIN, APPEND_JSON, SAVE_IMAGE, SAVE_VIDEO, SAVE_LOG
 
 logger = build_logger("log_server", "log_server.log", add_remote_handler=False)
 
@@ -40,6 +40,20 @@ async def save_image(image: UploadFile = File(...), image_path: str = Form(...))
     logger.info(f"Image saved successfully at {image_path}")
     return {"message": f"Image saved successfully at {image_path}"}
 
+@app.post(f"/{SAVE_VIDEO}")
+async def save_video(video: UploadFile = File(...), video_path: str = Form(...)):
+    """
+    Saves an uploaded video to the specified path.
+    """
+    # Note: 'video_path' should include the file name and extension for the video to be saved.
+    if os.path.dirname(video_path):
+        os.makedirs(os.path.dirname(video_path), exist_ok=True)
+    async with aiofiles.open(video_path, mode='wb') as f:
+        content = await video.read()  # Read the content of the uploaded video
+        await f.write(content)  # Write the video content to a file
+    logger.info(f"Video saved successfully at {video_path}")
+    return {"message": f"Image saved successfully at {video_path}"}
+
 @app.post(f"/{SAVE_LOG}")
 async def save_log(message: str = Form(...), log_path: str = Form(...)):
     """

serve/utils.py

@@ -6,7 +6,7 @@ import numpy as np
 import gradio as gr
 from pathlib import Path
 from model.model_registry import *
-from .constants import LOGDIR, LOG_SERVER_ADDR, APPEND_JSON, SAVE_IMAGE, SAVE_LOG
+from .constants import LOGDIR, LOG_SERVER_ADDR, APPEND_JSON, SAVE_IMAGE, SAVE_VIDEO, SAVE_LOG
 from typing import Union
 
 
@@ -151,6 +151,17 @@ def save_image_file_on_log_server(image_file:str):
         response = requests.post(url, files={'image': f}, data={'image_path': image_file})
     return response
 
+def save_video_file_on_log_server(video_file:str):
+
+    video_file = Path(video_file).absolute().relative_to(os.getcwd())
+    video_file = str(video_file)
+    # Open the video file in binary mode
+    url = f"{LOG_SERVER_ADDR}/{SAVE_VIDEO}"
+    with open(video_file, 'rb') as f:
+        # Make the POST request, sending the video file and the video path
+        response = requests.post(url, files={'video': f}, data={'video_path': video_file})
+    return response
+
 def append_json_item_on_log_server(json_item: Union[dict, str], log_file: str):
     if isinstance(json_item, dict):
         json_item = json.dumps(json_item)

serve/vote_utils.py

@@ -7,12 +7,14 @@ import regex as re
 from pathlib import Path
 from .utils import *
 from .log_utils import build_logger
-from .constants import IMAGE_DIR
+from .constants import IMAGE_DIR, VIDEO_DIR
 
 ig_logger = build_logger("gradio_web_server_image_generation", "gr_web_image_generation.log") # ig = image generation, loggers for single model direct chat
 igm_logger = build_logger("gradio_web_server_image_generation_multi", "gr_web_image_generation_multi.log") # igm = image generation multi, loggers for side-by-side and battle
 ie_logger = build_logger("gradio_web_server_image_editing", "gr_web_image_editing.log") # ie = image editing, loggers for single model direct chat
 iem_logger = build_logger("gradio_web_server_image_editing_multi", "gr_web_image_editing_multi.log") # iem = image editing multi, loggers for side-by-side and battle
+vg_logger = build_logger("gradio_web_server_video_generation", "gr_web_video_generation.log") # vg = video generation, loggers for single model direct chat
+vgm_logger = build_logger("gradio_web_server_video_generation_multi", "gr_web_video_generation_multi.log") # vgm = video generation multi, loggers for side-by-side and battle
 
 def vote_last_response_ig(state, vote_type, model_selector, request: gr.Request):
     with open(get_conv_log_filename(), "a") as fout:
@@ -25,6 +27,10 @@ def vote_last_response_ig(state, vote_type, model_selector, request: gr.Request)
         }
         fout.write(json.dumps(data) + "\n")
         append_json_item_on_log_server(data, get_conv_log_filename())
+    output_file = f'{IMAGE_DIR}/generation/{state.conv_id}.jpg'
+    with open(output_file, 'w') as f:
+        state.output.save(f, 'JPEG')
+    save_image_file_on_log_server(output_file)
         
 def vote_last_response_igm(states, vote_type, model_selectors, request: gr.Request):
     with open(get_conv_log_filename(), "a") as fout:
@@ -83,6 +89,47 @@ def vote_last_response_iem(states, vote_type, model_selectors, request: gr.Reque
             state.source_image.save(sf, 'JPEG')
         save_image_file_on_log_server(output_file)
         save_image_file_on_log_server(source_file)
+
+
+def vote_last_response_vg(state, vote_type, model_selector, request: gr.Request):
+    with open(get_conv_log_filename(), "a") as fout:
+        data = {
+            "tstamp": round(time.time(), 4),
+            "type": vote_type,
+            "model": model_selector,
+            "state": state.dict(),
+            "ip": get_ip(request),
+        }
+        fout.write(json.dumps(data) + "\n")
+        append_json_item_on_log_server(data, get_conv_log_filename())
+
+    output_file = f'{VIDEO_DIR}/generation/{state.conv_id}.mp4'
+    os.makedirs(os.path.dirname(output_file), exist_ok=True)
+    r = requests.get(state.output)
+    with open(output_file, 'wb') as outfile:
+        outfile.write(r.content)
+    save_video_file_on_log_server(output_file)
+
+
+
+def vote_last_response_vgm(states, vote_type, model_selectors, request: gr.Request):
+    with open(get_conv_log_filename(), "a") as fout:
+        data = {
+            "tstamp": round(time.time(), 4),
+            "type": vote_type,
+            "models": [x for x in model_selectors],
+            "states": [x.dict() for x in states],
+            "ip": get_ip(request),
+        }
+        fout.write(json.dumps(data) + "\n")
+        append_json_item_on_log_server(data, get_conv_log_filename())
+    for state in states:
+        output_file = f'{VIDEO_DIR}/generation/{state.conv_id}.mp4'
+        os.makedirs(os.path.dirname(output_file), exist_ok=True)
+        r = requests.get(state.output)
+        with open(output_file, 'wb') as outfile:
+            outfile.write(r.content)
+        save_video_file_on_log_server(output_file)
             
 
 ## Image Generation (IG) Single Model Direct Chat
@@ -217,6 +264,67 @@ def bothbad_vote_last_response_iem(
     return names + ("", "", gr.Image(height=512, width=512, type="pil"), "") + (disable_btn,) * 4
 
 
+## Video Generation (VG) Single Model Direct Chat
+def upvote_last_response_vg(state, model_selector, request: gr.Request):
+    ip = get_ip(request)
+    vg_logger.info(f"upvote. ip: {ip}")
+    vote_last_response_vg(state, "upvote", model_selector, request)
+    return ("",) + (disable_btn,) * 3
+
+def downvote_last_response_vg(state, model_selector, request: gr.Request):
+    ip = get_ip(request)
+    vg_logger.info(f"downvote. ip: {ip}")
+    vote_last_response_vg(state, "downvote", model_selector, request)
+    return ("",) + (disable_btn,) * 3
+
+
+def flag_last_response_vg(state, model_selector, request: gr.Request):
+    ip = get_ip(request)
+    vg_logger.info(f"flag. ip: {ip}")
+    vote_last_response_vg(state, "flag", model_selector, request)
+    return ("",) + (disable_btn,) * 3
+
+## Image Generation Multi (IGM) Side-by-Side and Battle
+
+def leftvote_last_response_vgm(
+    state0, state1, model_selector0, model_selector1, request: gr.Request
+):
+    vgm_logger.info(f"leftvote (named). ip: {get_ip(request)}")
+    vote_last_response_vgm(
+        [state0, state1], "leftvote", [model_selector0, model_selector1], request
+    )
+    return ("",) + (disable_btn,) * 4 + (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+
+
+def rightvote_last_response_vgm(
+    state0, state1, model_selector0, model_selector1, request: gr.Request
+):
+    vgm_logger.info(f"rightvote (named). ip: {get_ip(request)}")
+    vote_last_response_vgm(
+        [state0, state1], "rightvote", [model_selector0, model_selector1], request
+    )
+    return ("",) + (disable_btn,) * 4 + (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+
+
+def tievote_last_response_vgm(
+    state0, state1, model_selector0, model_selector1, request: gr.Request
+):
+    vgm_logger.info(f"tievote (named). ip: {get_ip(request)}")
+    vote_last_response_vgm(
+        [state0, state1], "tievote", [model_selector0, model_selector1], request
+    )
+    return ("",) + (disable_btn,) * 4 + (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+
+
+def bothbad_vote_last_response_vgm(
+    state0, state1, model_selector0, model_selector1, request: gr.Request
+):
+    vgm_logger.info(f"bothbad_vote (named). ip: {get_ip(request)}")
+    vote_last_response_vgm(
+        [state0, state1], "bothbad_vote", [model_selector0, model_selector1], request
+    )
+    return ("",) + (disable_btn,) * 4 + (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+
 share_js = """
 function (a, b, c, d) {
     const captureElement = document.querySelector('#share-region-named');
@@ -288,6 +396,21 @@ class ImageStateIE:
             }
         return base
 
+class VideoStateVG:
+    def __init__(self, model_name):
+        self.conv_id = uuid.uuid4().hex
+        self.model_name = model_name
+        self.prompt = None
+        self.output = None
+
+    def dict(self):
+        base = {
+            "conv_id": self.conv_id,
+            "model_name": self.model_name,
+            "prompt": self.prompt
+            }
+        return base
+
 
 
 
@@ -659,4 +782,169 @@ def generate_iem_annoy(gen_func, state0, state1, source_text, target_text, instr
         with open(output_file, 'w') as f:
             state.output.save(f, 'JPEG')
         save_image_file_on_log_server(src_img_file)
-        save_image_file_on_log_server(output_file)
+        save_image_file_on_log_server(output_file)
+
+def generate_vg(gen_func, state, text, model_name, request: gr.Request):
+    if not text:
+        raise gr.Warning("Prompt cannot be empty.")
+    if not model_name:
+        raise gr.Warning("Model name cannot be empty.")
+    if state is None:
+        state = VideoStateVG(model_name)
+    ip = get_ip(request)
+    vg_logger.info(f"generate. ip: {ip}")
+    start_tstamp = time.time()
+    generated_video = gen_func(text, model_name)
+    state.prompt = text
+    state.output = generated_video
+    state.model_name = model_name
+
+    yield state, generated_video
+
+    finish_tstamp = time.time()
+
+    with open(get_conv_log_filename(), "a") as fout:
+        data = {
+            "tstamp": round(finish_tstamp, 4),
+            "type": "chat",
+            "model": model_name,
+            "gen_params": {},
+            "start": round(start_tstamp, 4),
+            "finish": round(finish_tstamp, 4),
+            "state": state.dict(),
+            "ip": get_ip(request),
+        }
+        fout.write(json.dumps(data) + "\n")
+        append_json_item_on_log_server(data, get_conv_log_filename())
+
+    output_file = f'{VIDEO_DIR}/generation/{state.conv_id}.mp4'
+    os.makedirs(os.path.dirname(output_file), exist_ok=True)
+    r = requests.get(state.output)
+    with open(output_file, 'wb') as outfile:
+        outfile.write(r.content)
+    save_video_file_on_log_server(output_file)
+
+def generate_vgm(gen_func, state0, state1, text, model_name0, model_name1, request: gr.Request):
+    if not text:
+        raise gr.Warning("Prompt cannot be empty.")
+    if not model_name0:
+        raise gr.Warning("Model name A cannot be empty.")
+    if not model_name1:
+        raise gr.Warning("Model name B cannot be empty.")
+    if state0 is None:
+        state0 = VideoStateVG(model_name0)
+    if state1 is None:
+        state1 = VideoStateVG(model_name1)
+    ip = get_ip(request)
+    igm_logger.info(f"generate. ip: {ip}")
+    start_tstamp = time.time()
+    # Remove ### Model (A|B): from model name
+    model_name0 = re.sub(r"### Model A: ", "", model_name0)
+    model_name1 = re.sub(r"### Model B: ", "", model_name1)
+    generated_video0, generated_video1 = gen_func(text, model_name0, model_name1)
+    state0.prompt = text
+    state1.prompt = text
+    state0.output = generated_video0
+    state1.output = generated_video1
+    state0.model_name = model_name0
+    state1.model_name = model_name1
+
+    yield state0, state1, generated_video0, generated_video1
+
+    finish_tstamp = time.time()
+    # logger.info(f"===output===: {output}")
+
+    with open(get_conv_log_filename(), "a") as fout:
+        data = {
+            "tstamp": round(finish_tstamp, 4),
+            "type": "chat",
+            "model": model_name0,
+            "gen_params": {},
+            "start": round(start_tstamp, 4),
+            "finish": round(finish_tstamp, 4),
+            "state": state0.dict(),
+            "ip": get_ip(request),
+        }
+        fout.write(json.dumps(data) + "\n")
+        append_json_item_on_log_server(data, get_conv_log_filename())
+        data = {
+            "tstamp": round(finish_tstamp, 4),
+            "type": "chat",
+            "model": model_name1,
+            "gen_params": {},
+            "start": round(start_tstamp, 4),
+            "finish": round(finish_tstamp, 4),
+            "state": state1.dict(),
+            "ip": get_ip(request),
+        }
+        fout.write(json.dumps(data) + "\n")
+        append_json_item_on_log_server(data, get_conv_log_filename())
+
+    for i, state in enumerate([state0, state1]):
+        output_file = f'{VIDEO_DIR}/generation/{state.conv_id}.mp4'
+        os.makedirs(os.path.dirname(output_file), exist_ok=True)
+        r = requests.get(state.output)
+        with open(output_file, 'wb') as outfile:
+            outfile.write(r.content)
+        save_video_file_on_log_server(output_file)
+
+
+def generate_vgm_annoy(gen_func, state0, state1, text, model_name0, model_name1, request: gr.Request):
+    if not text:
+        raise gr.Warning("Prompt cannot be empty.")
+    if state0 is None:
+        state0 = VideoStateVG(model_name0)
+    if state1 is None:
+        state1 = VideoStateVG(model_name1)
+    ip = get_ip(request)
+    vgm_logger.info(f"generate. ip: {ip}")
+    start_tstamp = time.time()
+    model_name0 = re.sub(r"### Model A: ", "", model_name0)
+    model_name1 = re.sub(r"### Model B: ", "", model_name1)
+    generated_video0, generated_video1, model_name0, model_name1 = gen_func(text, model_name0, model_name1)
+    state0.prompt = text
+    state1.prompt = text
+    state0.output = generated_video0
+    state1.output = generated_video1
+    state0.model_name = model_name0
+    state1.model_name = model_name1
+
+    yield state0, state1, generated_video0, generated_video1, \
+        gr.Markdown(f"### Model A: {model_name0}"), gr.Markdown(f"### Model B: {model_name1}")
+
+    finish_tstamp = time.time()
+    # logger.info(f"===output===: {output}")
+
+    with open(get_conv_log_filename(), "a") as fout:
+        data = {
+            "tstamp": round(finish_tstamp, 4),
+            "type": "chat",
+            "model": model_name0,
+            "gen_params": {},
+            "start": round(start_tstamp, 4),
+            "finish": round(finish_tstamp, 4),
+            "state": state0.dict(),
+            "ip": get_ip(request),
+        }
+        fout.write(json.dumps(data) + "\n")
+        append_json_item_on_log_server(data, get_conv_log_filename())
+        data = {
+            "tstamp": round(finish_tstamp, 4),
+            "type": "chat",
+            "model": model_name1,
+            "gen_params": {},
+            "start": round(start_tstamp, 4),
+            "finish": round(finish_tstamp, 4),
+            "state": state1.dict(),
+            "ip": get_ip(request),
+        }
+        fout.write(json.dumps(data) + "\n")
+        append_json_item_on_log_server(data, get_conv_log_filename())
+
+    for i, state in enumerate([state0, state1]):
+        output_file = f'{VIDEO_DIR}/generation/{state.conv_id}.mp4'
+        os.makedirs(os.path.dirname(output_file), exist_ok=True)
+        r = requests.get(state.output)
+        with open(output_file, 'wb') as outfile:
+            outfile.write(r.content)
+        save_video_file_on_log_server(output_file)

serve_local.sh

@@ -1,26 +0,0 @@
-LOG_SERVER_PORT=9929
-export LOG_SERVER="http://0.0.0.0:${LOG_SERVER_PORT}"
-export LOG_SERVER_SUBDOMAIN=""
-export LOGDIR="tmp_log/"
-export SERVER_PORT=10201 # gradio server port
-
-# start log server
-uvicorn serve.log_server:app --reload --port ${LOG_SERVER_PORT} --host 0.0.0.0 &
-PID1=$!
-# start gradio server
-python app.py &
-PID2=$!
-
-# Function to kill the background jobs if this script is terminated
-cleanup() {
-    echo "Cleaning up..."
-    kill $PID1 $PID2
-    exit 0
-}
-
-# Trap SIGINT (Ctrl+C) and SIGTERM signals and call cleanup function
-trap cleanup SIGINT SIGTERM
-
-# Wait for both background processes to finish
-wait $PID1
-wait $PID2