DF.py

import os
import time
import uuid
import torch
import gradio as gr
from diffusers import WanPipeline, AutoencoderKLWan
from diffusers.utils import export_to_video
from dfloat11 import DFloat11Model
import spaces

os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:128"
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"

@spaces.GPU(enable_queue=True)
def generate_video(prompt, negative_prompt, width, height, num_frames,
                   guidance_scale, guidance_scale_2, num_inference_steps, fps, cpu_offload):

    start_time = time.time()
    torch.cuda.empty_cache()

    # Load VAE and Wan2.2 pipeline
    vae = AutoencoderKLWan.from_pretrained(
        "Wan-AI/Wan2.2-T2V-A14B-Diffusers",
        subfolder="vae",
        torch_dtype=torch.float32,
    )

    pipe = WanPipeline.from_pretrained(
        "Wan-AI/Wan2.2-T2V-A14B-Diffusers",
        vae=vae,
        torch_dtype=torch.bfloat16,
    )

    # Only apply second-stage DFloat11 model
    DFloat11Model.from_pretrained(
        "DFloat11/Wan2.2-T2V-A14B-2-DF11",
        device="cpu",
        cpu_offload=cpu_offload,
        bfloat16_model=pipe.transformer_2,
    )

    pipe.enable_model_cpu_offload()

    # Generate video frames
    output_frames = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        height=height,
        width=width,
        num_frames=num_frames,
        guidance_scale=guidance_scale,
        guidance_scale_2=guidance_scale_2,
        num_inference_steps=num_inference_steps,
    ).frames[0]

    # Export to video
    output_path = f"/tmp/{uuid.uuid4().hex}_t2v.mp4"
    export_to_video(output_frames, output_path, fps=fps)

    elapsed = time.time() - start_time
    print(f"✅ Generated in {elapsed:.2f}s, saved to {output_path}")
    return output_path


# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## 🎬 Wan2.2 + DFloat11 (Stage 2 only) - Text to Video Generator")

    with gr.Row():
        prompt = gr.Textbox(label="Prompt", value="A serene koi pond at night, with glowing lanterns reflecting on the rippling water. Ethereal fireflies dance above as cherry blossoms gently fall.", lines=3)
        negative_prompt = gr.Textbox(label="Negative Prompt", value="色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走", lines=3)

    with gr.Row():
        width = gr.Slider(256, 1280, value=1280, step=64, label="Width")
        height = gr.Slider(256, 720, value=720, step=64, label="Height")
        fps = gr.Slider(8, 30, value=16, step=1, label="FPS")

    with gr.Row():
        num_frames = gr.Slider(8, 81, value=81, step=1, label="Frames")
        num_inference_steps = gr.Slider(10, 60, value=40, step=1, label="Inference Steps")

    with gr.Row():
        guidance_scale = gr.Slider(1.0, 10.0, value=4.0, step=0.1, label="Guidance Scale (Stage 1)")
        guidance_scale_2 = gr.Slider(1.0, 10.0, value=3.0, step=0.1, label="Guidance Scale (Stage 2)")
        cpu_offload = gr.Checkbox(label="Enable CPU Offload", value=True)

    with gr.Row():
        btn = gr.Button("🚀 Generate Video")
        output_video = gr.Video(label="Generated Video")

    btn.click(
        generate_video,
        inputs=[prompt, negative_prompt, width, height, num_frames, guidance_scale, guidance_scale_2, num_inference_steps, fps, cpu_offload],
        outputs=[output_video]
    )

demo.launch()