Create app.py

app.py

@@ -0,0 +1,71 @@
+import gradio as gr
+import os
+import torch
+from PIL import Image
+from inference.flovd_demo import generate_video
+import requests
+import shutil
+
+# --------- SETUP CODE: Download models if not found ---------
+FVSM_PATH = "./ckpt/FVSM/FloVD_FVSM_Controlnet.pt"
+OMSM_PATH = "./ckpt/OMSM"
+POSE_DIR = "./assets/manual_poses"
+EXAMPLE_POSE = os.path.join(POSE_DIR, "example.txt")
+
+def download_if_missing():
+    os.makedirs("ckpt/FVSM", exist_ok=True)
+    os.makedirs("ckpt/OMSM", exist_ok=True)
+    os.makedirs("assets/manual_poses", exist_ok=True)
+    os.makedirs("output/generated_videos", exist_ok=True)
+
+    # Download FVSM model
+    if not os.path.exists(FVSM_PATH):
+        print("Downloading FVSM model...")
+        url = "https://huggingface.co/datasets/mutqa/FloVD-HF-Assets/resolve/main/FloVD_FVSM_Controlnet.pt"
+        r = requests.get(url, stream=True)
+        with open(FVSM_PATH, 'wb') as f:
+            shutil.copyfileobj(r.raw, f)
+
+    # Download OMSM weights (as folder with .safetensors inside)
+    if not os.listdir(OMSM_PATH):
+        print("Cloning OMSM weights...")
+        os.system("git clone https://huggingface.co/datasets/mutqa/FloVD-HF-OMSM ckpt/OMSM")
+
+    # Download example camera pose
+    if not os.path.exists(EXAMPLE_POSE):
+        print("Downloading example pose...")
+        url = "https://huggingface.co/datasets/mutqa/FloVD-HF-Assets/resolve/main/example.txt"
+        r = requests.get(url)
+        with open(EXAMPLE_POSE, "w") as f:
+            f.write(r.text)
+
+# --------- UI Function ---------
+def run_flovd(prompt, image, cam_pose_name):
+    download_if_missing()
+    image_path = "./temp_input.png"
+    image.save(image_path)
+
+    generate_video(
+        prompt=prompt,
+        fvsm_path=FVSM_PATH,
+        omsm_path=OMSM_PATH,
+        image_path=image_path,
+        cam_pose_name=cam_pose_name,
+        output_path="./output/",
+        dtype=torch.float16,
+    )
+    return "./output/generated_videos/your_video.mp4"
+
+# --------- Launch Gradio ---------
+iface = gr.Interface(
+    fn=run_flovd,
+    inputs=[
+        gr.Textbox(label="Prompt"),
+        gr.Image(type="pil", label="Input Image"),
+        gr.Textbox(label="Camera Pose File Name", value="example.txt"),
+    ],
+    outputs=gr.Video(label="Generated Video"),
+    title="FloVD - Optical Flow Video Diffusion with Camera Motion",
+)
+
+iface.launch()