app.py

import os
import io
import base64
import json
from typing import Callable, Any
from PIL import Image
import gradio as gr
from common import utils as posex
project_dir = os.path.dirname(os.path.abspath(__file__))
print(project_dir)

if '__file__' in globals():
    posex.set_save_dir(os.path.join(os.path.dirname(__file__), '', 'saved_poses'))
else:
    # cf. https://stackoverflow.com/a/53293924
    import inspect
    posex.set_save_dir(os.path.join(os.path.dirname(inspect.getfile(lambda: None)), '', 'saved_poses'))


def js2py(
        name: str,
        id: Callable[[str], str],
        js: Callable[[str], str],
        sink: gr.components.IOComponent,
) -> gr.Textbox:
    v_set = gr.Button(elem_id=id(f'{name}_set'),visible=False)
    v = gr.Textbox(elem_id=id(name),visible=False)
    v_sink = gr.Textbox(visible=False)
    v_set.click(fn=None, _js=js(name), outputs=[v, v_sink])
    v_sink.change(fn=None, _js=js(f'{name}_after'), outputs=[sink])
    return v


def py2js(
        name: str,
        fn: Callable[[], str],
        id: Callable[[str], str],
        js: Callable[[str], str],
        sink: gr.components.IOComponent,
) -> None:
    v_fire = gr.Button(elem_id=id(f'{name}_get'),visible=False)
    v_sink = gr.Textbox(visible=False)
    v_sink2 = gr.Textbox(visible=False)
    v_fire.click(fn=wrap_api(fn), outputs=[v_sink, v_sink2])
    v_sink2.change(fn=None, _js=js(name), inputs=[v_sink], outputs=[sink])


def jscall(
        name: str,
        fn: Callable[[str], str],
        id: Callable[[str], str],
        js: Callable[[str], str],
        sink: gr.components.IOComponent,
) -> None:
    v_args_set = gr.Button(elem_id=id(f'{name}_args_set'), visible=False)

    v_args = gr.JSON(elem_id=id(f'{name}_args'), visible=False)
    v_args_sink = gr.JSON(visible=False)
    v_args_set.click(fn=None, _js=js(f'{name}_args'), outputs=[v_args, v_args_sink])
    v_args_sink.change(fn=None, _js=js(f'{name}_args_after'), outputs=[sink])

    v_fire = gr.Button(elem_id=id(f'{name}_get'),visible=False)
    v_sink = gr.Textbox(visible=False)
    v_sink2 = gr.Textbox(visible=False)
    v_fire.click(fn=wrap_api(fn), inputs=[v_args], outputs=[v_sink, v_sink2])
    v_sink2.change(fn=None, _js=js(name), inputs=[v_sink], outputs=[sink])

def generatecall(
        name: str,
        fn: Callable[[str], str],
        id: Callable[[str], str],
        js: Callable[[str], str],
        sink: gr.components.IOComponent,
        prompt,
        prompt_n,
        output_img,
) -> None:
    v_args_set = gr.Button(elem_id=id(f'{name}_args_set'), visible=False)
    v_args = gr.JSON(elem_id=id(f'{name}_args'), visible=False)
    v_args_sink = gr.JSON(visible=False)
    v_args_set.click(fn=None, _js=js(f'{name}_args'), outputs=[v_args, v_args_sink])
    v_args_sink.change(fn=None, _js=js(f'{name}_args_after'), outputs=[sink])

    v_fire = gr.Button(elem_id=id(f'{name}_get'),visible=False)
    v_sink = gr.Textbox(visible=False)
    v_sink2 = gr.Textbox(visible=False)
    v_fire.click(fn=fn, inputs=[v_args,prompt,prompt_n], outputs=[output_img])
    v_sink2.change(fn=None, _js=js(name), inputs=[v_sink], outputs=[sink])


def get_self_extension():
    if '__file__' in globals():
        filepath = __file__
    else:
        import inspect
        filepath = inspect.getfile(lambda: None)



# APIs

def wrap_api(fn):
    _r = 0

    def f(*args, **kwargs):
        nonlocal _r
        _r += 1
        v = fn(*args, **kwargs)
        return v, str(_r)

    return f


def all_pose():
    return json.dumps(list(posex.all_poses()))


def delete_pose(args):
    posex.delete_pose(json.loads(args)[0])
    return ''


def save_pose(args):
    posex.save_pose(json.loads(args)[0])
    return ''


def load_pose(args):
    return json.dumps(posex.load_pose(json.loads(args)[0]))

# def get_imgs(args):
#     return posex.get_img(args)

def generate_imgs(data, image_prompt, image_n_prompt):
    return posex.generate_img(data, image_prompt, image_n_prompt)
def get_image_sketch(image_prompt, image_n_prompt, image):
    return posex.get_image_sketch(image, image_prompt, image_n_prompt)


def javascript_html():
    script_js = f'script.js?{os.path.getmtime(os.path.join(project_dir,"script.js"))}'
    path7 = f'javascript/posex-webui.js?{os.path.getmtime(os.path.join(project_dir,"javascript/posex-webui.js"))}'
    head = f'<script type="text/javascript" src="file={script_js}"></script>\n'
    head += f'<script type="text/javascript" src="file={path7}"></script>\n'

    return head


def css_html():
    head = f'<link rel="stylesheet" property="stylesheet" href="file=style.css">'

    return head


def reload_javascript():
    js = javascript_html()
    css = css_html()

    def template_response(*args, **kwargs):
        res = GradioTemplateResponseOriginal(*args, **kwargs)
        res.body = res.body.replace(b'</head>', f'{js}</head>'.encode("utf8"))
        res.body = res.body.replace(b'</body>', f'{css}</body>'.encode("utf8"))
        res.init_headers()
        return res
    gr.routes.templates.TemplateResponse = template_response

GradioTemplateResponseOriginal = gr.routes.templates.TemplateResponse

introduce = """
# ControlnetOverMask
### Controlnet
### Inpainting
### Stable diffusion

Use controlnet to generate sketches and characters, and keep the background unchanged.

Usually when we generate a very perfect background image, we want to add image elements, but using controlnet directly will affect the original background. This project aims to add elements to the page while keeping the background unchanged, and can directly operate on the original background.

Support skeletal character generation and sketch generation.

Optimize an inpaint model ( ./models/Any-inpainting  ) for the general domain against the stablediffusion-inpaint model.

### Two modes, openpose control and manuscript control
#### openpose control
#### manuscript control

"""

openpose_intro = """
Add character skeletons to a forest scene generated by SD, and keep the background unchanged to generate controllable characters
( ./models/control_any5_openpose )

<img src="https://raw.githubusercontent.com/hiroLinGoing/datasets/main/person.png"  width="400" height="300">

<video width="640" height="320" controls>
  <source src="https://raw.githubusercontent.com/hiroLinGoing/datasets/main/person_control.mp4" type="video/mp4">
</video>
"""

manuscript_intro = """
Added manuscript houses to an SD generated forest scene
Added manuscript chairs to an SD generated snow scene

<img src="https://raw.githubusercontent.com/hiroLinGoing/datasets/main/house.png"  width="600" height="300">
<img src="https://raw.githubusercontent.com/hiroLinGoing/datasets/main/chair.png"  width="600" height="300">

<video width="640" height="320" controls>
  <source src="https://raw.githubusercontent.com/hiroLinGoing/datasets/main/house.mp4" type="video/mp4">
</video>


"""

if __name__ == '__main__':
    reload_javascript()
    app = gr.Blocks()
    with app:
        is_img2img = []
        # id = lambda s: f'posex-{["t2i", "i2i"]}-{s}'
        id = lambda s: f'posex-t2i-{s}'
        js = lambda s: f'globalThis["{id(s)}"]'

        ext = get_self_extension()

        xpath = os.path.join(project_dir,"javascript/lazyload/posex-webui.js")
        js_ = [project_dir, f'{xpath}?{os.path.getmtime(xpath)}']
        print(js_)

        with gr.Blocks() as demo:
            with gr.Row():
                gr.Markdown(introduce)
            with gr.Tab("sketch"):
                with gr.Row():
                    gr.Markdown(<img src="https://raw.githubusercontent.com/hiroLinGoing/datasets/main/chair.png"  width="600" height="300">)
                with gr.Row():
                    image_prompt_sketch = gr.Textbox(label="image prompt", value='', elem_id=id('image_prompt'))
                    image_n_prompt_sketch = gr.Textbox(label="negative prompt", value='',
                                                       elem_id=id('image_n_prompt'))
                with gr.Row():
                    image_sketch = gr.Image(interactive=True,source='upload', type="numpy", tool="sketch").style(height=512, width=512)
                    result_img = gr.Gallery(label='Output', show_label=False, elem_id="gallery").style(grid=2,
                                                                                                   height='auto')
                with gr.Row():
                    generate_result_button = gr.Button(value='generate')
                    generate_result_button.click(fn=get_image_sketch,inputs=[image_prompt_sketch, image_n_prompt_sketch, image_sketch],
                                          outputs=[result_img])

            with gr.Tab("openpose", elem_id=id('tab')):
                with gr.Row():
                    gr.Markdown(openpose_intro)
                # with gr.Row():
                #     enabled = gr.Checkbox(value=False, label='Send this image to ControlNet.', elem_id=id('enabled'))
                #     cn_num = gr.Number(value=0, precision=0, label='Target ControlNet number', visible=True)
                with gr.Row():
                    # segment_prompt = gr.Textbox(label="segment prompt", value='',
                    #                             elem_id=id('segment_prompt'))
                    image_prompt = gr.Textbox(label="image prompt", value='', elem_id=id('image_prompt'))
                    image_n_prompt = gr.Textbox(label="negative prompt", value='',
                                                elem_id=id('image_n_prompt'))

                gr.HTML(value='\n'.join(js_), elem_id=id('js'), visible=False)

                gr.HTML(value='', elem_id=id('html'))

                generate_button = gr.Button(elem_id=id(f'generate'), value='generate')

                with gr.Column():
                    result_img = gr.Gallery(label='Output', show_label=False, elem_id="gallery").style(grid=2,
                                                                                                           height='auto')

                with gr.Group(visible=False):
                    sink = gr.HTML(value='', visible=False)  # to suppress error in javascript
                    base64 = js2py('base64', id, js, sink)
                    py2js('allposes', all_pose, id, js, sink)
                    jscall('delpose', delete_pose, id, js, sink)
                    jscall('savepose', save_pose, id, js, sink)
                    jscall('loadpose', load_pose, id, js, sink)
                    # jscall('getimgs', get_imgs, id, js, sink)
                    generatecall('getimgs', generate_imgs, id, js, sink, image_prompt, image_n_prompt, result_img)
    app.launch()