app.py

import os
import torch
import gradio as gr
from PIL import Image
from inference.flovd_demo import generate_video
from huggingface_hub import snapshot_download

# -----------------------------------
# Step 1: Download model checkpoints
# -----------------------------------
hf_token = os.getenv("HF_TOKEN", None)

snapshot_download(
    repo_id="roll-ai/FloVD-weights",
    repo_type="dataset",
    local_dir="./",
    allow_patterns="ckpt/**",
    token=hf_token,
)

# -----------------------------------
# Step 2: Setup paths and config
# -----------------------------------
BASE_DIR = "app"
FVSM_PATH = os.path.join(BASE_DIR, "ckpt/FVSM/FloVD_FVSM_Controlnet.pt")
OMSM_PATH = os.path.join(BASE_DIR, "ckpt/OMSM/")
DEPTH_CKPT_PATH = os.path.join(BASE_DIR, "ckpt/others/depth_anything_v2_metric_hypersim_vitb.pth")
OUTPUT_PATH = os.path.join(BASE_DIR, "results")
GEN_VID_DIR = os.path.join(OUTPUT_PATH, "generated_videos")

POSE_TYPE = "re10k"
CONTROLNET_GUIDANCE_END = 0.4
SPEED = 1.0
NUM_FRAMES = 81
FPS = 16
INFER_STEPS = 50

os.makedirs(GEN_VID_DIR, exist_ok=True)

# -----------------------------------
# Helper Functions
# -----------------------------------

def list_generated_videos():
    try:
        return sorted([
            f for f in os.listdir(GEN_VID_DIR)
            if f.endswith(".mp4")
        ])
    except Exception as e:
        print(f"[⚠️] Could not list contents: {str(e)}")
        return []

def run_flovd(prompt, image, cam_pose_name):
    try:
        print("\n----------------------------")
        print("🚀 Starting video generation")
        print("----------------------------")

        image_path = os.path.join(BASE_DIR, "temp_input.png")
        image.save(image_path)
        print(f"📸 Image saved at {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_PATH,
            controlnet_guidance_end=CONTROLNET_GUIDANCE_END,
            pose_type=POSE_TYPE,
            speed=SPEED,
            use_flow_integration=True,
            depth_ckpt_path=DEPTH_CKPT_PATH,
            dtype=torch.float16,
            num_frames=NUM_FRAMES,
            fps=FPS,
            num_inference_steps=INFER_STEPS,
        )

        prompt_short = prompt[:30].strip().replace(" ", "_").replace(".", "").replace(",", "")
        video_filename = f"{prompt_short}_{cam_pose_name}.mp4"
        video_path = os.path.join(GEN_VID_DIR, video_filename)

        print(f"\n📍 Looking for generated video at: {video_path}")
        if os.path.exists(video_path) and os.path.getsize(video_path) > 0:
            print("✅ Video file found.")
            return video_path, "✅ Video generated successfully.", list_generated_videos(), None
        else:
            print("❌ File missing or empty.")
            return None, f"❌ File not found at: {video_path}", list_generated_videos(), None

    except Exception as e:
        print(f"🔥 Exception occurred: {str(e)}")
        return None, f"🔥 Exception: {str(e)}", list_generated_videos(), None

def get_video_file(video_name):
    video_path = os.path.join(GEN_VID_DIR, video_name)
    if os.path.exists(video_path):
        return video_path
    else:
        return None

# -----------------------------------
# Step 3: Launch Gradio Interface
# -----------------------------------

with gr.Blocks(title="FloVD + CogVideoX") as demo:
    gr.Markdown("## 🎥 FloVD - Camera Motion Guided Video Generation + Downloader")

    with gr.Row():
        with gr.Column():
            prompt = gr.Textbox(label="Prompt")
            image = gr.Image(type="pil", label="Input Image")
            pose_file = gr.Textbox(label="Camera Pose Filename (e.g. abc.txt)")
            run_btn = gr.Button("Generate Video")
        with gr.Column():
            output_video = gr.Video(label="Generated Video")
            status_text = gr.Markdown(label="Status / Logs")  # shows errors/success
            video_selector = gr.Dropdown(choices=list_generated_videos(), label="Select Video to Download")
            download_btn = gr.Button("Download Selected")
            file_output = gr.File(label="Download Link")

    run_btn.click(
        fn=run_flovd,
        inputs=[prompt, image, pose_file],
        outputs=[output_video, status_text, video_selector, file_output],
    )

    download_btn.click(
        fn=get_video_file,
        inputs=[video_selector],
        outputs=[file_output],
    )

demo.launch(server_name="0.0.0.0", server_port=7860)