app.py

import torch
import gradio as gr
import spaces
import random
import os
from diffusers.utils import export_to_video
from diffusers import AutoencoderKLWan, WanPipeline
from diffusers.schedulers.scheduling_unipc_multistep import UniPCMultistepScheduler
from diffusers.schedulers.scheduling_flow_match_euler_discrete import FlowMatchEulerDiscreteScheduler
from huggingface_hub import hf_hub_download
from lycoris import create_lycoris_from_weights

# Define model options
MODEL_OPTIONS = {
    "Wan2.1-T2V-1.3B": "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
    "Wan2.1-T2V-14B": "Wan-AI/Wan2.1-T2V-14B-Diffusers"
}

# Define scheduler options
SCHEDULER_OPTIONS = {
    "UniPCMultistepScheduler": UniPCMultistepScheduler,
    "FlowMatchEulerDiscreteScheduler": FlowMatchEulerDiscreteScheduler
}

def download_adapter(repo_id, weight_name=None):
    """
    Download the adapter file from the Hugging Face Hub.
    If weight_name is not provided, attempts to use pytorch_lora_weights.safetensors
    """
    adapter_filename = weight_name if weight_name else "pytorch_lora_weights.safetensors"
    cache_dir = os.environ.get('HF_PATH', os.path.expanduser('~/.cache/huggingface/hub/models'))
    cleaned_adapter_path = repo_id.replace("/", "_").replace("\\", "_").replace(":", "_")
    path_to_adapter = os.path.join(cache_dir, cleaned_adapter_path)
    os.makedirs(path_to_adapter, exist_ok=True)
    
    try:
        path_to_adapter_file = hf_hub_download(
            repo_id=repo_id, 
            filename=adapter_filename, 
            local_dir=path_to_adapter
        )
        return path_to_adapter_file
    except Exception as e:
        # If specific file not found, try to get a list of available safetensors files
        if weight_name is None:
            raise ValueError(f"Could not download default adapter file: {str(e)}\nPlease specify the exact weight file name.")
        else:
            raise ValueError(f"Could not download adapter file {weight_name}: {str(e)}")

@spaces.GPU(duration=300)
def generate_video(
    model_choice,
    prompt,
    negative_prompt,
    lycoris_id,
    lycoris_weight_name,
    lycoris_scale,
    scheduler_type,
    flow_shift,
    height,
    width,
    num_frames,
    guidance_scale,
    num_inference_steps,
    output_fps,
    seed
):
    # Get model ID from selection
    model_id = MODEL_OPTIONS[model_choice]
    
    # Set seed for reproducibility
    if seed == -1 or seed is None or seed == "":
        seed = random.randint(0, 2147483647)
    else:
        seed = int(seed)
    
    # Set the seed
    torch.manual_seed(seed)
    
    # Load model
    vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
    pipe = WanPipeline.from_pretrained(model_id, vae=vae, torch_dtype=torch.float16)
    
    # Set scheduler
    if scheduler_type == "UniPCMultistepScheduler":
        pipe.scheduler = UniPCMultistepScheduler.from_config(
            pipe.scheduler.config,
            flow_shift=flow_shift
        )
    else:
        pipe.scheduler = FlowMatchEulerDiscreteScheduler(shift=flow_shift)
    
    # Move to GPU
    pipe.to("cuda")
    
    # Load LyCORIS weights if provided
    if lycoris_id and lycoris_id.strip():
        try:
            # Download the adapter file
            adapter_file_path = download_adapter(repo_id=lycoris_id, weight_name=lycoris_weight_name if lycoris_weight_name and lycoris_weight_name.strip() else None)
            
            # Apply LyCORIS adapter
            wrapper, *_ = create_lycoris_from_weights(lycoris_scale, adapter_file_path, pipe.transformer)
            wrapper.merge_to()
            
        except ValueError as e:
            # Return informative error if there are issues loading the adapter
            if "more than one weights file" in str(e) or "Could not download default adapter file" in str(e):
                return f"Error: The repository '{lycoris_id}' may contain multiple weight files. Please specify a weight name using the 'LyCORIS Weight Name' field.", seed
            else:
                return f"Error loading LyCORIS weights: {str(e)}", seed
    
    # Enable CPU offload for low VRAM
    pipe.enable_model_cpu_offload()
    
    # Generate video
    output = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        height=height,
        width=width,
        num_frames=num_frames,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        generator=torch.Generator("cuda").manual_seed(seed)
    ).frames[0]
    
    # Export to video
    temp_file = "output.mp4"
    export_to_video(output, temp_file, fps=output_fps)
    
    return temp_file, seed

# Create the Gradio interface
with gr.Blocks() as demo:
    gr.HTML("""
    <p align="center">
        <svg version="1.1" viewBox="0 0 1200 295" xmlns="http://www.w3.org/2000/svg" xmlns:v="https://vecta.io/nano" width="400">
            <path d='M195.43 109.93a.18.18 0 0 1-.27-.03l-.17-.26a.13.12-43.2 0 1 .01-.16q21.87-21.98 42.02-37.95c17.57-13.93 36.61-26.36 57.23-34.01q13.26-4.92 24.23-4.1c10.91.81 18.01 8.05 19.12 18.82.79 7.76-.64 15.73-3.29 23.37q-5.87 16.92-15.46 32.47a1.05 1.02-40.3 0 0-.09.91q11.79 32.3 2.48 63.6c-13.06 43.95-53.78 72.83-99.49 71.06q-16.32-.64-30.92-6.27a1.88 1.88 0 0 0-1.65.15c-12.54 7.69-26.93 14.55-41.16 17.68-4.99 1.1-9.54.6-16.14.62a6.31 6.04 50.1 0 1-1.35-.15c-19.69-4.51-18.03-25.43-12.71-40.65q5.95-17.05 15.9-32.34a.89.88 50.1 0 0 .07-.82c-13.76-32.93-7.8-71.64 14.32-99.05q20.75-25.73 51.75-33.99 29.88-7.96 59.38 2.25a.18.18 0 0 1 .04.32l-11.17 7.52a1.18 1.17-55 0 1-.93.17q-26.82-6.3-51.8 2.83c-24.65 9.02-43.94 28.53-52.73 53.27q-8.3 23.33-2.99 48.49.7 3.32 3.15 10.35c7.87 22.55 24.1 40.02 45.62 50.35a.52.51 43.2 0 1 .03.91l-10.31 5.96a.7.68-44.3 0 1-.69 0q-27.53-15.57-41.3-43.86a.1.09-44.1 0 0-.17 0q-6.59 11.46-11.16 23.67c-4.39 11.73-11.35 38.04 10.45 38.5q6.38.13 11.55-1.22c17.2-4.51 32.86-13.33 47.66-23.11q24.23-16.04 49.48-39.51c28.57-26.55 55.11-56.92 74.11-90.87q5.96-10.65 10.32-23.77c2.38-7.15 5.16-20.81-.67-27.03q-3.67-3.92-8.08-4.48-10.87-1.39-22.3 2.5-11.06 3.77-18.33 7.43c-30.93 15.54-58.11 36.9-83.59 60.43zm100.11-36.16q10.18 9.98 17.31 22.67a.6.59 46.1 0 1-.01.59l-6.17 10.69a.34.33-42.8 0 1-.59-.02q-7.06-14.92-18.34-25.98-11.28-11.06-26.34-17.82a.34.33-48.4 0 1-.03-.59l10.56-6.38a.6.59 42.7 0 1 .59-.02q12.83 6.88 23.02 16.86zm430.69 78.36h10a1.13 1.12-74.2 0 0 .96-.54q11.1-17.8 23.93-38.19 1.8-2.85 5.39-3.1 3.42-.23 11.73-.11 1.88.02 2.61.66a2.23 2.08-54 0 1 .34 2.77q-12.01 19.41-29.2 46.98a.86.85 45.5 0 0 0 .93q12.8 20.23 28.66 46.12.22.36.29 1.09a1.98 1.97 85.7 0 1-1.85 2.16q-7.51.45-13.91-.05c-2.52-.2-4.25-2.09-5.57-4.24q-8.26-13.5-22.19-36.26a.74.72-15 0 0-.63-.35H726.2a.62.61 0 0 0-.62.61q-.03 31.31.02 35.17.07 5.25-5.02 5.31-7.87.1-10.26-.19-3.72-.45-3.71-4.69.05-35.54-.02-135.71c0-3.87.67-6.1 4.77-6.12q7.55-.05 10.37.06c3.4.13 3.94 3.29 3.92 6.34q-.22 43.68-.02 80.75a.6.6 0 0 0 .6.6zM191.9 74.2l1.52.2a.33.33 0 0 1 0 .65l-1.52.22a.33.33 0 0 0-.28.29l-.2 1.52a.33.33 0 0 1-.65 0l-.22-1.52a.33.33 0 0 0-.29-.28l-1.52-.2a.33.33 0 0 1 0-.65l1.52-.22a.33.33 0 0 0 .28-.29l.2-1.52a.33.33 0 0 1 .65 0l.22 1.52a.33.33 0 0 0 .29.28zm64.12 2.96l2.78.37a.33.33 0 0 1 0 .65l-2.78.39a.33.33 0 0 0-.28.29l-.37 2.78a.33.33 0 0 1-.65 0l-.39-2.78a.33.33 0 0 0-.29-.28l-2.78-.37a.33.33 0 0 1 0-.65l2.78-.39a.33.33 0 0 0 .28-.29l.37-2.78a.33.33 0 0 1 .65 0l.39 2.78a.33.33 0 0 0 .29.28zm150.31 27.5q-.37 2.75-.37 5.59.1 55.28.07 95.29-.01 5.69-4.97 5.57-4.23-.1-9.06-.04-4.92.06-4.92-4.56-.03-47.02-.01-122.76c0-4.45 2.16-5.1 6.27-5.1q17.99.04 19.41.04c4.14-.01 5.19 2.26 6.29 6.06q6.68 23.18 28.31 98.03a.22.22 0 0 0 .42-.01q26.48-97.12 26.96-99.03 1.26-5.01 5.72-5.05 9.8-.09 22.27.08c3.29.04 3.71 3.1 3.7 5.88q-.07 58.05-.18 118.23c-.01 4.45.51 7.92-4.69 8.13q-3.32.14-7.06.02c-5.17-.17-7.18.03-7.18-6.18q.01-66.62.01-100.28a.141.141 0 0 0-.28-.04q-20.97 79.04-26.98 101.79-1.27 4.8-5.99 4.76-4.96-.03-13.08-.01c-4.08.01-5.44-1.81-6.54-5.76q-17.37-62.23-27.98-100.66a.07.07 0 0 0-.14.01zm-172.34 1.26l2.49.36a.32.32 0 0 1 0 .63l-2.49.37a.32.32 0 0 0-.27.27l-.36 2.49a.32.32 0 0 1-.63 0l-.37-2.49a.32.32 0 0 0-.27-.27l-2.49-.36a.32.32 0 0 1 0-.63l2.49-.37a.32.32 0 0 0 .27-.27l.36-2.49a.32.32 0 0 1 .63 0l.37 2.49a.32.32 0 0 0 .27.27zm46.8 2.66l2.5.41a.34.34 0 0 1 0 .67l-2.5.4a.34.34 0 0 0-.28.28l-.41 2.5a.34.34 0 0 1-.67 0l-.4-2.5a.34.34 0 0 0-.28-.28l-2.5-.41a.34.34 0 0 1 0-.67l2.5-.4a.34.34 0 0 0 .28-.28l.41-2.5a.34.34 0 0 1 .67 0l.4 2.5a.34.34 0 0 0 .28.28zm311.55 38.61q.18-12.55-.16-15.38c-.44-3.69-4.05-3.64-6.93-3.64q-18.53.03-52.88-.01c-3.01 0-3.94-2.46-3.95-5.32q-.01-5.13.04-8.35.08-4.18 4.29-4.26 5.99-.11 59.86-.05c13.05.02 18.7 5.13 18.7 18.33q.01 61.89.01 63.24c.03 15.73-6.4 19.32-21.07 19.32q-30.49.01-43.5-.02c-14.08-.04-20.24-3.83-20.35-18.8q-.12-16.77.08-27.74c.21-12.16 6.75-16.58 18.27-16.58q22.02.01 46.85-.01a.74.74 0 0 0 .74-.73zm-.66 18.75q-26.48-.05-41.68.04-4.64.03-4.77 4.52-.22 8.08-.06 16.74c.12 6.3 2.92 5.93 8.53 5.94q14.25.01 34.08-.03c4.89-.02 4.57-4.12 4.6-8.32q.08-11.72-.06-18.26a.65.64 89.1 0 0-.64-.63zm45.34 45.09c-2.82.02-4.77-1.44-4.77-4.22q-.01-73.72.01-78.31c.05-12.96 5.29-18.22 18.24-18.3q11.13-.06 38.25-.02 4.76.01 4.86 4.11.13 4.95.08 8.97c-.05 3.99-1.85 4.93-5.85 4.92q-24.69-.01-26.97 0c-5.43 0-9.69-.88-9.69 5.79q-.01 66 .02 70.71.02 3.45-.84 4.58-1.47 1.96-4.11 1.92-9.21-.15-9.23-.15zm200.71-17.86q16.73 0 18.8-.04c5.15-.13 4.81-3.81 4.8-8.15q-.01-10.05.01-70.24 0-4.51 4.92-4.56 7.92-.08 10 .09 3.9.31 3.9 4.48.02 60.37-.01 77.78c-.02 15.09-6.99 18.68-20.65 18.54q-.42 0-21.77 0-21.35 0-21.77 0c-13.66.14-20.63-3.45-20.65-18.54q-.03-17.41-.01-77.78 0-4.17 3.9-4.48 2.08-.17 10-.09 4.92.05 4.92 4.56.02 60.19.01 70.24c-.01 4.34-.35 8.02 4.79 8.14q2.08.05 18.81.05zm63.52 13.64q-.01-73.72.01-78.31c.05-12.96 5.29-18.22 18.24-18.3q11.13-.06 38.25-.02 4.76.01 4.86 4.11.13 4.95.08 8.97c-.05 3.99-1.85 4.93-5.85 4.92q-24.69-.01-26.97 0c-5.43 0-9.69-.88-9.69 5.79q-.01 66 .02 70.71.02 3.45-.84 4.58-1.47 1.96-4.11 1.92-9.21-.15-9.23-.15c-2.82.02-4.77-1.44-4.77-4.22zm109.52-25.57q18.81-60.07 20.5-65.73 1.57-5.27 6.17-5.34 3.67-.05 10.57.07c2.6.05 3.39 2.42 2.65 4.66q-11.68 35.64-41.85 128.15-1.51 4.61-6.26 4.65-4.38.03-7.66-.07c-3.83-.12-3.5-3.08-2.61-6.07q3.74-12.68 8.78-29.76a1.97 1.91-47.2 0 0-.03-1.2q-19.47-53.62-34.46-94.88c-.96-2.65-1.38-5.23 2.18-5.46q1.66-.11 11.3-.09 4.8 0 6.58 4.96 13.98 38.91 23.84 66.12a.16.16 0 0 0 .3-.01zm-767.05-7.85q-.03-28.67.01-36.09.03-4-1.39-6.63c-2.4-4.47-7.99-5.48-11.98-2.79q-4.8 3.23-4.86 9.16-.17 16.69-.02 24.3a.47.47 0 0 1-.47.48h-10.36a.46.46 0 0 1-.46-.46q.02-21.54 0-23.86-.03-2.66-.94-5.2c-2.63-7.33-11.73-8.07-15.95-1.8q-1.99 2.96-1.98 7.75.05 21.93.01 22.99a.59.58-89 0 1-.58.57h-10.14a.57.57 0 0 1-.57-.57v-43.73a1.04 1.04 0 0 1 1.04-1.04h8.41a1.25 1.24-2.5 0 1 1.24 1.13l.3 3.59a.27.27 0 0 0 .44.19q.06-.06 1.28-1.44c2.92-3.32 7.69-4.52 12.01-4.15q8.1.68 11.68 7.78a.37.37 0 0 0 .64.03c4.8-7.6 13.82-9.19 22.01-6.61 8.67 2.72 11.91 11.02 11.96 19.52q.07 14.44.05 24.75a1.29 1.26 66.4 0 1-.32.85l-10.44 11.52a.36.36 0 0 1-.62-.24zm-87.22-51.13l1.95.24a.34.34 0 0 1 0 .67l-1.96.22a.34.34 0 0 0-.3.3l-.24 1.95a.34.34 0 0 1-.67 0l-.22-1.96a.34.34 0 0 0-.3-.3l-1.95-.24a.34.34 0 0 1 0-.67l1.96-.22a.34.34 0 0 0 .3-.3l.24-1.95a.34.34 0 0 1 .67 0l.22 1.96a.34.34 0 0 0 .3.3zm13.88 41.55q.18 0 .43.34a.87.82 79.1 0 0 .45.33l.6.17a.362.362 0 0 1-.14.71q-.38-.05-.48-.08-.04-.02-.4.44-.26.33-.45.33-.19 0-.45-.32-.37-.46-.41-.44-.1.03-.48.09a.363.363 0 0 1-.15-.71l.6-.18a.87.82-79.8 0 0 .45-.33q.24-.35.43-.35zm25.81 35.05q.01.16-.38.69a.6.59 33.3 0 1-.79.16q-1.23-.74-1.23-.8 0-.05 1.19-.85a.6.59-36.1 0 1 .79.12q.42.51.42.68z' fill="#100c6d" />
        </svg>
    </p>
    """)
    gr.Markdown("# Wan 2.1 T2V with LyCORIS")
    
    with gr.Row():
        with gr.Column(scale=1):
            model_choice = gr.Dropdown(
                choices=list(MODEL_OPTIONS.keys()),
                value="Wan2.1-T2V-1.3B",
                label="Model"
            )
            
            prompt = gr.Textbox(
                label="Prompt",
                value="",
                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",
                lines=3
            )
            
            with gr.Row():
                lycoris_id = gr.Textbox(
                    label="Adapter Repo (e.g., markury/wan-st)",
                    value="markury/wan-st"
                )
            
            with gr.Row():
                lycoris_weight_name = gr.Textbox(
                    label="Adapter Path in Repo",
                    value="pytorch_lora_weights.safetensors",
                    info="Specify for repos with multiple .safetensors files, e.g.: adapter_model.safetensors, pytorch_lora_weights.safetensors, etc."
                )
                lycoris_scale = gr.Slider(
                    label="Adapter Scale",
                    minimum=0.0,
                    maximum=2.0,
                    value=1.00,
                    step=0.05
                )
            
            with gr.Row():
                scheduler_type = gr.Dropdown(
                    choices=list(SCHEDULER_OPTIONS.keys()),
                    value="UniPCMultistepScheduler",
                    label="Scheduler"
                )
                flow_shift = gr.Slider(
                    label="Flow Shift",
                    minimum=1.0,
                    maximum=12.0,
                    value=3.0,
                    step=0.5,
                    info="2.0-5.0 for smaller videos, 7.0-12.0 for larger videos"
                )
            
            with gr.Row():
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=1024,
                    value=832,
                    step=32
                )
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=1792,
                    value=480,
                    step=30
                )
            
            with gr.Row():
                num_frames = gr.Slider(
                    label="Number of Frames (4k+1 is recommended, e.g. 33)",
                    minimum=17,
                    maximum=129,
                    value=33,
                    step=4
                )
                output_fps = gr.Slider(
                    label="Output FPS",
                    minimum=8,
                    maximum=30,
                    value=16,
                    step=1
                )
            
            with gr.Row():
                guidance_scale = gr.Slider(
                    label="Guidance Scale (CFG)",
                    minimum=1.0,
                    maximum=15.0,
                    value=4.0,
                    step=0.5
                )
                num_inference_steps = gr.Slider(
                    label="Inference Steps",
                    minimum=10,
                    maximum=100,
                    value=20,
                    step=1
                )
            
            seed = gr.Number(
                label="Seed (-1 for random)",
                value=-1,
                precision=0,
                info="Set a specific seed for deterministic results"
            )
            
            generate_btn = gr.Button("Generate Video")
        
        with gr.Column(scale=1):
            output_video = gr.Video(label="Generated Video")
            used_seed = gr.Number(label="Seed", precision=0)
            
    generate_btn.click(
        fn=generate_video,
        inputs=[
            model_choice,
            prompt,
            negative_prompt,
            lycoris_id,
            lycoris_weight_name,
            lycoris_scale,
            scheduler_type,
            flow_shift,
            height,
            width,
            num_frames,
            guidance_scale,
            num_inference_steps,
            output_fps,
            seed
        ],
        outputs=[output_video, used_seed]
    )
    
    gr.Markdown("""
    ## Tips for best results:
    - For smaller resolution videos, try lower values of flow shift (2.0-5.0)
    - For larger resolution videos, try higher values of flow shift (7.0-12.0)
    - Number of frames should be of the form 4k+1 (e.g., 33, 81)
    - Stick to lower frame counts. Even at 480p, an 81 frame sequence at 30 steps will nearly time out the request in this ZeroGPU space.
    
    """)

demo.launch()