app.py

import gradio as gr
import io
from PIL import Image
import torch
from clip_interrogator import Config, Interrogator

config = Config()
config.device = 'cuda' if torch.cuda.is_available() else 'cpu'
config.blip_offload = False if torch.cuda.is_available() else True
config.chunk_size = 2048
config.flavor_intermediate_count = 512
config.blip_num_beams = 64

ci = Interrogator(config)

def inference(input_images, mode, best_max_flavors):
    prompt_results = []
    for image_bytes in input_images:
        image = Image.open(io.BytesIO(image_bytes)).convert('RGB')
        if mode == 'best':
            prompt_result = ci.interrogate(image, max_flavors=int(best_max_flavors))
        elif mode == 'classic':
            prompt_result = ci.interrogate_classic(image)
        else:
            prompt_result = ci.interrogate_fast(image)
        prompt_results.append(prompt_result)
    return "\n\n".join(prompt_results)

title = """
    <div style="text-align: center; max-width: 500px; margin: 0 auto;">
        <h1 style="font-weight: 600; margin-bottom: 7px;">
            CLIP Interrogator 2.1
        </h1>
        <p style="margin-bottom: 10px;font-size: 94%;font-weight: 100;line-height: 1.5em;">
        Want to figure out what a good prompt might be to create new images like an existing one? 
        <br />The CLIP Interrogator is here to get you answers!
        <br />This version is specialized for producing nice prompts for use with Stable Diffusion 2.0 using the ViT-H-14 OpenCLIP model!
        </p>
    </div>
"""

article = """
<div style="text-align: center; max-width: 500px; margin: 0 auto;font-size: 94%;">
    <p>
    Server busy? You can also run on <a href="https://colab.research.google.com/github/pharmapsychotic/clip-interrogator/blob/open-clip/clip_interrogator.ipynb">Google Colab</a>
    </p>
    <p>
    Has this been helpful to you? Follow Pharma on twitter 
    <a href="https://twitter.com/pharmapsychotic">@pharmapsychotic</a> 
    and check out more tools at his
    <a href="https://pharmapsychotic.com/tools.html">Ai generative art tools list</a>
    </p>
</div>
"""

css = '''
#col-container {width: 80%; margin-left: auto; margin-right: auto;}
a {text-decoration-line: underline; font-weight: 600;}
'''

with gr.Blocks(css=css) as block:
    with gr.Column(elem_id="col-container"):
        gr.HTML(title)

        input_image = gr.Files(label="Inputs", file_count="multiple", type='binary', elem_id='inputs')
        with gr.Row():
            mode_input = gr.Radio(['best', 'classic', 'fast'], label='Select mode', value='best')
            flavor_input = gr.Slider(minimum=2, maximum=24, step=2, value=4, label='Best mode max flavors')
        
        submit_btn = gr.Button("Submit")
        
        output_text = gr.Textbox(label="Output Prompts", lines=10, elem_id="output_text")

        with gr.Group(elem_id="share-btn-container"):
            gr.HTML(article)

    submit_btn.click(fn=inference, inputs=[input_image, mode_input, flavor_input], outputs=[output_text], api_name="clipi2")

block.queue().launch()