app.py

import gradio as gr
import numpy as np
import random, json, spaces, torch, time
from diffusers import AutoencoderKL, FlowMatchEulerDiscreteScheduler
from videox_fun.pipeline import ZImageControlPipeline
from videox_fun.models import ZImageControlTransformer2DModel
from transformers import AutoTokenizer, Qwen3ForCausalLM
from diffusers import AutoencoderKL
from utils.image_utils import get_image_latent, rescale_image
from utils.prompt_utils import polish_prompt
# from controlnet_aux import HEDdetector, MLSDdetector, OpenposeDetector, CannyDetector, MidasDetector
from controlnet_aux.processor import Processor


# MODEL_REPO = "Tongyi-MAI/Z-Image-Turbo"
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1280

# git clone https://huggingface.co/Tongyi-MAI/Z-Image-Turbo
MODEL_LOCAL = "models/Z-Image-Turbo/"
# curl -L -o Z-Image-Turbo-Fun-Controlnet-Union.safetensors https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union/resolve/main/Z-Image-Turbo-Fun-Controlnet-Union.safetensors
TRANSFORMER_LOCAL = "models/Z-Image-Turbo-Fun-Controlnet-Union.safetensors"

weight_dtype = torch.bfloat16

# load transformer
transformer = ZImageControlTransformer2DModel.from_pretrained(
    MODEL_LOCAL,
    subfolder="transformer",
    transformer_additional_kwargs={
        "control_layers_places": [0, 5, 10, 15, 20, 25],
        "control_in_dim": 16
    },
).to("cuda", torch.bfloat16)

if TRANSFORMER_LOCAL is not None:
    print(f"From checkpoint: {TRANSFORMER_LOCAL}")
    from safetensors.torch import load_file, safe_open
    state_dict = load_file(TRANSFORMER_LOCAL)
    state_dict = state_dict["state_dict"] if "state_dict" in state_dict else state_dict

    m, u = transformer.load_state_dict(state_dict, strict=False)
    print(f"missing keys: {len(m)}, unexpected keys: {len(u)}")

# load ZImageControlPipeline
vae = AutoencoderKL.from_pretrained(
    MODEL_LOCAL,
    subfolder="vae",
    device_map="cuda"
).to(weight_dtype)

tokenizer = AutoTokenizer.from_pretrained(
    MODEL_LOCAL, 
    subfolder="tokenizer"
)
text_encoder = Qwen3ForCausalLM.from_pretrained(
    MODEL_LOCAL, 
    subfolder="text_encoder", 
    torch_dtype=weight_dtype,
)
# scheduler = FlowMatchEulerDiscreteScheduler(num_train_timesteps=1000, shift=3)
scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(
    MODEL_LOCAL, 
    subfolder="scheduler"
)
pipe = ZImageControlPipeline(
    vae=vae,
    tokenizer=tokenizer,
    text_encoder=text_encoder,
    transformer=transformer,
    scheduler=scheduler,
)
pipe.to("cuda", torch.bfloat16)
# pipe.transformer = transformer
# pipe.to("cuda")

# ======== AoTI compilation + FA3 ========
# pipe.transformer.layers._repeated_blocks = ["ZImageTransformerBlock"]
# spaces.aoti_blocks_load(pipe.transformer.layers, "zerogpu-aoti/Z-Image", variant="fa3")

def prepare(prompt, is_polish_prompt):
    if not is_polish_prompt: return prompt, False
    polished_prompt = polish_prompt(prompt)
    return polished_prompt, True

@spaces.GPU
def inference(
    prompt,
    negative_prompt,
    input_image,
    image_scale=1.0,
    control_mode='Canny',
    control_context_scale = 0.75,
    seed=42,
    randomize_seed=True,
    guidance_scale=1.5,
    num_inference_steps=8,
    progress=gr.Progress(track_tqdm=True),
):
    timestamp = time.time()
    print(f"timestamp: {timestamp}")

    # process image
    print("DEBUG: process image")
    if input_image is None:
        print("Error: input_image is empty.")
        return None
    
    # input_image, width, height = scale_image(input_image, image_scale)
    # control_mode='HED'
    processor_id = 'canny'
    if control_mode == 'HED':
        processor_id = 'softedge_hed'
    if control_mode =='Depth':
        processor_id = 'depth_midas'
    if control_mode =='MLSD':
        processor_id = 'mlsd'
    if control_mode =='Pose':
        processor_id = 'openpose_full'

    print(f"DEBUG: processor_id={processor_id}")
    processor = Processor(processor_id)

    # Width must be divisible by 16
    control_image, width, height = rescale_image(input_image, image_scale, 16)
    control_image = control_image.resize((1024, 1024))

    print("DEBUG: processor running")
    control_image = processor(control_image, to_pil=True)
    control_image = control_image.resize((width, height))

    print("DEBUG: control_image_torch")
    control_image_torch = get_image_latent(control_image, sample_size=[height, width])[:, :, 0]

    # generation
    if randomize_seed: seed = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed)

    image = pipe(
        prompt=prompt,
        negative_prompt = negative_prompt,
        height=height,
        width=width,
        generator=generator,
        guidance_scale=guidance_scale,
        control_image=control_image_torch,
        num_inference_steps=num_inference_steps,
        control_context_scale=control_context_scale,
    ).images[0]

    return image, seed, control_image


def read_file(path: str) -> str:
    with open(path, 'r', encoding='utf-8') as f:
        content = f.read()
    return content


css = """
#col-container {
    margin: 0 auto;
    max-width: 960px;
}
"""

with open('static/data.json', 'r') as file:
    data = json.load(file)
examples = data['examples']

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        with gr.Column():
            gr.HTML(read_file("static/header.html"))
        with gr.Row():
            with gr.Column():
                input_image = gr.Image(
                    height=290, sources=['upload', 'clipboard'], 
                    image_mode='RGB', 
                    # elem_id="image_upload", 
                    type="pil", label="Upload")
                
                prompt = gr.Textbox(
                    label="Prompt",
                    show_label=False,
                    lines=2,
                    placeholder="Enter your prompt",
                    # container=False,
                )
                is_polish_prompt = gr.Checkbox(label="Polish prompt", value=True)
                control_mode = gr.Radio(
                    choices=["Canny", "Depth", "HED", "MLSD", "Pose"],
                    value="Canny",
                    label="Control Mode"
                )
                run_button = gr.Button("Generate", variant="primary")
                with gr.Accordion("Advanced Settings", open=False):
                    
                    negative_prompt = gr.Textbox(
                        label="Negative prompt",
                        lines=2,
                        container=False,
                        placeholder="Enter your negative prompt",
                        value="blurry ugly bad"
                    )
                    with gr.Row():
                        num_inference_steps = gr.Slider(
                            label="Steps",
                            minimum=1,
                            maximum=30,
                            step=1,
                            value=9,
                        )
                        control_context_scale = gr.Slider(
                            label="Context scale",
                            minimum=0.0,
                            maximum=1.0,
                            step=0.01,
                            value=0.75,
                        )

                    with gr.Row():
                        guidance_scale = gr.Slider(
                            label="Guidance scale",
                            minimum=0.0,
                            maximum=10.0,
                            step=0.1,
                            value=1.0,
                        )

                        image_scale = gr.Slider(
                            label="Image scale",
                            minimum=0.5,
                            maximum=2.0,
                            step=0.1,
                            value=1.0,
                        )

                    seed = gr.Slider(
                        label="Seed",
                        minimum=0,
                        maximum=MAX_SEED,
                        step=1,
                        value=42,
                    )
                    randomize_seed = gr.Checkbox(label="Randomize seed", value=False)

            with gr.Column():
                output_image = gr.Image(label="Generated image", show_label=False)
                polished_prompt = gr.Textbox(label="Polished prompt", interactive=False)

                with gr.Accordion("Preprocessor output", open=False):
                    control_image = gr.Image(label="Control image", show_label=False)
                    

        
        gr.Examples(examples=examples, inputs=[input_image, prompt, control_mode])
        gr.Markdown(read_file("static/footer.md"))

    run_button.click(
        fn=prepare,
        inputs=[prompt, is_polish_prompt],
        outputs=[polished_prompt, is_polish_prompt]
        # outputs=gr.State(),  # Pass to the next function, not to UI at this step
    ).then(
        fn=inference,
        inputs=[
            polished_prompt,
            negative_prompt,
            input_image,
            image_scale,
            control_mode,
            control_context_scale,
            seed,
            randomize_seed,
            guidance_scale,
            num_inference_steps,
        ],
        outputs=[output_image, seed, control_image],
    )

if __name__ == "__main__":
    demo.launch(mcp_server=True)