app.py

import gradio as gr
from transformers import pipeline
import numpy as np
from diffusers import DiffusionPipeline


prompt_writer = pipeline('text-generation', model='toloka/gpt2-large-rl-prompt-writing')
prompt_reward_model = pipeline('text-classification', model='toloka/prompts_reward_model')
pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")


def write_prompt(img_desc):
        prompts = [p['generated_text'] for p in prompt_writer(img_desc + '</s>', max_new_tokens=100, num_return_sequences=2)]
        scores = [p['score'] for p in prompt_reward_model(prompts, function_to_apply='none')]
        return prompts[np.argmax(scores)].split('</s>')[1].strip()


def generate(text):
    prompt = write_prompt(text)
    img = pipe(prompt=prompt, num_inference_steps=50).images[0]
    return img, prompt


with gr.Blocks() as demo:
    with gr.Column(variant="panel"):
        with gr.Row(variant="compact"):
            text = gr.Textbox(
                label="Enter your image description, e.g., \"a cat\"",
                show_label=False,
                max_lines=1,
                placeholder="Enter your image description, e.g., \"a cat\"",
            ).style(
                container=False,
            )
            btn = gr.Button("Generate image").style(full_width=False)

        written_prompt = gr.outputs.Textbox(label="AI-written prompt")
        gen_img = gr.outputs.Image(type="pil",
            label="Generated image",
        ).style(object_fit="contain", height=512)

    btn.click(generate, text, [gen_img, written_prompt])

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