app.py

import random
from pathlib import Path

import spaces
import torch
import gradio as gr
from huggingface_hub import snapshot_download
from transformers import AutoTokenizer, T5EncoderModel

from minit2i_pipeline import (
    MiniT2IMMJiTModel,
    MiniT2IFlowMatchScheduler,
    MiniT2ITextToImagePipeline,
)

REPO_ID = "MiniT2I/MiniT2I"
TEXT_ENCODER = "google/flan-t5-large"
DTYPE = torch.bfloat16
MAX_SEED = 2**31 - 1

MODELS = {
    "MiniT2I-B/16 (base)": "minit2i-b-16",
    "MiniT2I-L/16 (large)": "minit2i-l-16",
}

root = Path(snapshot_download(REPO_ID))

# Shared T5 text encoder — both variants use google/flan-t5-large
tokenizer = AutoTokenizer.from_pretrained(TEXT_ENCODER)
text_encoder = T5EncoderModel.from_pretrained(
    TEXT_ENCODER, torch_dtype=torch.float32
).to("cuda")


def _load(model_dir):
    transformer = MiniT2IMMJiTModel.from_pretrained(
        root / model_dir / "transformer", torch_dtype=DTYPE
    )
    scheduler = MiniT2IFlowMatchScheduler.from_pretrained(root / model_dir / "scheduler")
    pipe = MiniT2ITextToImagePipeline(
        transformer=transformer,
        scheduler=scheduler,
        tokenizer=tokenizer,
        text_encoder=text_encoder,
        text_encoder_name=TEXT_ENCODER,
    )
    pipe.to("cuda")
    return pipe


# Preload BOTH models in the global context
PIPES = {label: _load(model_dir) for label, model_dir in MODELS.items()}


@spaces.GPU(duration=60)
def generate(prompt, model_label, guidance_scale, num_inference_steps, seed, randomize_seed,
             progress=gr.Progress(track_tqdm=True)):
    if not prompt or not prompt.strip():
        raise gr.Error("Please enter a prompt.")
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator(device="cuda").manual_seed(int(seed))
    pipe = PIPES[model_label]
    image = pipe(
        prompt,
        guidance_scale=guidance_scale,
        num_inference_steps=int(num_inference_steps),
        generator=generator,
        progress=True,
    ).images[0]
    return image, seed


EXAMPLES = [
    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
    "A lonely astronaut standing on a quiet beach under two moons.",
    "A cozy cabin in a snowy forest at dusk, warm light in the windows.",
    "A bowl of ramen with steam rising, photorealistic, top-down view.",
]

with gr.Blocks() as demo:
    gr.Markdown(
        """
        # 🍊 MiniT2I
        A minimalist text-to-image model — pick the **B/16** or **L/16** variant below.
        Both are preloaded. Images are generated at 512×512.
        """
    )
    with gr.Row():
        with gr.Column(scale=1):
            prompt = gr.Textbox(
                label="Prompt",
                placeholder="Describe the image you want to generate…",
                lines=3,
            )
            model_label = gr.Radio(
                choices=list(MODELS.keys()),
                value=list(MODELS.keys())[0],
                label="Model",
            )
            run_btn = gr.Button("Generate", variant="primary")
            with gr.Accordion("Advanced settings", open=False):
                guidance_scale = gr.Slider(
                    label="Guidance scale", minimum=1.0, maximum=10.0, step=0.1, value=2.5
                )
                num_inference_steps = gr.Slider(
                    label="Inference steps", minimum=10, maximum=150, step=1, value=100
                )
                with gr.Row():
                    seed = gr.Slider(
                        label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0
                    )
                    randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
        with gr.Column(scale=1):
            output = gr.Image(label="Result", height=512)

    gr.Examples(examples=EXAMPLES, inputs=prompt)

    inputs = [prompt, model_label, guidance_scale, num_inference_steps, seed, randomize_seed]
    run_btn.click(generate, inputs=inputs, outputs=[output, seed])
    prompt.submit(generate, inputs=inputs, outputs=[output, seed])

demo.launch(theme=gr.themes.Citrus())