import random import gradio as gr from sample import (arg_parse, sampling, load_fontdiffuer_pipeline) def run_fontdiffuer(source_image, character, reference_image, sampling_step, guidance_scale, batch_size): args.character_input = False if source_image is not None else True args.content_character = character args.sampling_step = sampling_step args.guidance_scale = guidance_scale args.batch_size = batch_size args.seed = random.randint(0, 10000) out_image = sampling( args=args, pipe=pipe, content_image=source_image, style_image=reference_image) return out_image if __name__ == '__main__': args = arg_parse() args.demo = True args.ckpt_dir = 'ckpt' args.ttf_path = 'ttf/KaiXinSongA.ttf' args.device = 'cpu' # load fontdiffuer pipeline pipe = load_fontdiffuer_pipeline(args=args) with gr.Blocks() as demo: with gr.Row(): with gr.Column(scale=1): gr.HTML("""

FontDiffuser

Zhenhua Yang, Dezhi Peng, Yuxin Kong, Yuyi Zhang, Cong Yao, Lianwen Jin

South China University of Technology, Alibaba DAMO Academy

[arXiv] [Homepage] [Github]

1.We propose FontDiffuser, which is capable to generate unseen characters and styles, and it can be extended to the cross-lingual generation, such as Chinese to Korean.

2. FontDiffuser excels in generating complex character and handling large style variation. And it achieves state-of-the-art performance.

""") gr.Image('figures/result_vis.png') gr.Image('figures/demo_tips.png') with gr.Column(scale=1): with gr.Row(): source_image = gr.Image(width=320, label='[Option 1] Source Image', image_mode='RGB', type='pil') reference_image = gr.Image(width=320, label='Reference Image', image_mode='RGB', type='pil') with gr.Row(): character = gr.Textbox(value='隆', label='[Option 2] Source Character') with gr.Row(): fontdiffuer_output_image = gr.Image(height=200, label="FontDiffuser Output Image", image_mode='RGB', type='pil') sampling_step = gr.Slider(20, 50, value=20, step=10, label="Sampling Step", info="The sampling step by FontDiffuser.") guidance_scale = gr.Slider(1, 12, value=7.5, step=0.5, label="Scale of Classifier-free Guidance", info="The scale used for classifier-free guidance sampling") batch_size = gr.Slider(1, 4, value=1, step=1, label="Batch Size", info="The number of images to be sampled.") FontDiffuser = gr.Button('Run FontDiffuser') gr.Markdown("## Examples that You Can Choose Below⬇️") with gr.Row(): gr.Markdown("## Examples") with gr.Row(): with gr.Column(scale=1): gr.Markdown("## Example 1️⃣: Source Image and Reference Image") gr.Markdown("### In this mode, we provide both the source image and \ the reference image for you to try our demo!") gr.Examples( examples=[['figures/source_imgs/source_灨.jpg', 'figures/ref_imgs/ref_籍.jpg'], ['figures/source_imgs/source_鑻.jpg', 'figures/ref_imgs/ref_鹰.jpg'], ['figures/source_imgs/source_鑫.jpg', 'figures/ref_imgs/ref_壤.jpg'], ['figures/source_imgs/source_釅.jpg', 'figures/ref_imgs/ref_雕.jpg']], inputs=[source_image, reference_image] ) with gr.Column(scale=1): gr.Markdown("## Example 2️⃣: Character and Reference Image") gr.Markdown("### In this mode, we provide the content character and the reference image \ for you to try our demo!") gr.Examples( examples=[['龍', 'figures/ref_imgs/ref_鷢.jpg'], ['轉', 'figures/ref_imgs/ref_鲸.jpg'], ['懭', 'figures/ref_imgs/ref_籍_1.jpg'], ['識', 'figures/ref_imgs/ref_鞣.jpg']], inputs=[character, reference_image] ) with gr.Column(scale=1): gr.Markdown("## Example 3️⃣: Reference Image") gr.Markdown("### In this mode, we provide only the reference image, \ you can upload your own source image or you choose the character above \ to try our demo!") gr.Examples( examples=['figures/ref_imgs/ref_闡.jpg', 'figures/ref_imgs/ref_雕.jpg', 'figures/ref_imgs/ref_豄.jpg', 'figures/ref_imgs/ref_馨.jpg', 'figures/ref_imgs/ref_鲸.jpg', 'figures/ref_imgs/ref_檀.jpg', 'figures/ref_imgs/ref_鞣.jpg', 'figures/ref_imgs/ref_穗.jpg', 'figures/ref_imgs/ref_欟.jpg', 'figures/ref_imgs/ref_籍_1.jpg', 'figures/ref_imgs/ref_鷢.jpg', 'figures/ref_imgs/ref_媚.jpg', 'figures/ref_imgs/ref_籍.jpg', 'figures/ref_imgs/ref_壤.jpg', 'figures/ref_imgs/ref_蜓.jpg', 'figures/ref_imgs/ref_鹰.jpg'], examples_per_page=20, inputs=reference_image ) FontDiffuser.click( fn=run_fontdiffuer, inputs=[source_image, character, reference_image, sampling_step, guidance_scale, batch_size], outputs=fontdiffuer_output_image) demo.launch(debug=True)