import os import json from PIL import Image, ImageDraw import io import requests openai.organization = os.getenv("API_ORG") openai.api_key = os.getenv("API_KEY") app_password = os.getenv("APP_PASSWORD") app_username = os.getenv("APP_USERNAME") canvas_width = 500 canvas_height = 400 html = f"""

""" scripts = """ async () => { let isSelecting = false; let startX, startY, endX, endY; const canvas = document.getElementById('canvas-root'); const ctx = canvas.getContext('2d'); const canvasRect = canvas.getBoundingClientRect(); const selectRect = document.getElementById('selectRect'); const coordinatesElement = document.querySelector('#rectangle textarea'); function handleMouseDown(event) { startX = event.clientX - canvasRect.left; startY = event.clientY - canvasRect.top; if (startX >= 0 && startY >= 0 && startX <= canvasRect.width && startY <= canvasRect.height) { isSelecting = true; } } function handleMouseMove(event) { if (isSelecting) { endX = Math.min(event.clientX - canvasRect.left, canvasRect.width); endY = Math.min(event.clientY - canvasRect.top, canvasRect.height); endX = Math.max(0, endX); endY = Math.max(0, endY); const left = Math.min(startX, endX); const top = Math.min(startY, endY); const width = Math.abs(endX - startX); const height = Math.abs(endY - startY); selectRect.style.left = left + 'px'; selectRect.style.top = top + 'px'; selectRect.style.width = width + 'px'; selectRect.style.height = height + 'px'; coordinatesElement.value = `{"left": ${left}, "top": ${top}, "width": ${width}, "height": ${height}}`; coordinatesElement.dispatchEvent(new CustomEvent("input")) } } function handleMouseUp() { isSelecting = false; } document.addEventListener('mousedown', handleMouseDown); document.addEventListener('mousemove', handleMouseMove); document.addEventListener('mouseup', handleMouseUp); } """ image_change = """ async () => { const canvas = document.getElementById('canvas-root'); const ctx= canvas.getContext('2d'); const canvasRect = canvas.getBoundingClientRect(); const selectRect = document.getElementById('selectRect'); selectRect.style.left = 0; selectRect.style.top = 0; selectRect.style.width = 0; selectRect.style.height = 0; ctx.clearRect(0, 0, canvasRect.width, canvasRect.height); var img = document.querySelector('#input_image img'); img.onload = function(){ if ((img.naturalWidth / canvasRect.width) > (img.naturalHeight / canvasRect.height)) { width = canvasRect.width; height = img.naturalHeight * (width / img.naturalWidth); } else { height = canvasRect.height; width = img.naturalWidth * (height / img.naturalHeight); } ctx.drawImage(img, 0, 0, width, height); } } """ def pil_to_bytes(pil_image, format='PNG'): image_bytes = io.BytesIO() pil_image.save(image_bytes, format=format) return image_bytes.getvalue() def expand2square(image, background_color): width, height = image.size longest = max(width, height) result = Image.new(image.mode, (longest, longest), background_color) result.paste(image, (0, 0)) return result.resize((2048, 2048)) def gen_mask(image, left, top, right, bottom): mask = Image.new("RGBA", image.size, (0, 0, 0, 255)) width = image.size[0] height = image.size[1] draw = ImageDraw.Draw(mask) draw.rectangle( [(left*width, top*height), (right*width, bottom*height)], fill=(255, 255, 255, 0) ) return mask def create_edit(image, rect, prompt): rect = json.loads(rect) image.putalpha(alpha=255) square_image = expand2square(image, "black") left, top, width, height = rect["left"], rect["top"], rect["width"], rect["height"] left, top, right, bottom = left / canvas_width, top / canvas_height, (left + width) / canvas_width, (top + height) / canvas_height response = openai.Image.create_edit( image=pil_to_bytes(square_image), mask=pil_to_bytes(gen_mask(square_image, left, top, right, bottom)), prompt=prompt, n=1, size="512x512" ) edited_image_url = response['data'][0]['url'] edited_image = requests.get(edited_image_url) edited_image = Image.open(io.BytesIO(edited_image.content)) raw_width, raw_height = image.size raw_longest = max(raw_width, raw_height) crop_width = raw_width * edited_image.size[0] / raw_longest crop_height = raw_height * edited_image.size[1] / raw_longest croped_edited_image = edited_image.crop((0,0,crop_width, crop_height)) return croped_edited_image with gr.Blocks() as demo: with gr.Column(): with gr.Row(): with gr.Column(): prompt_text = gr.Textbox(label="Prompt") prompt_examples = gr.Examples( examples=[ "White plate.", "A cherry on top of the pasta.", "Curry.", ], inputs=[prompt_text], outputs=None, ) in_image = gr.Image(label="Input", elem_id="input_image", type="pil") image_examples = gr.Examples( examples=[ "./001.png", "./002.png", "./003.png", ], inputs=[in_image], outputs=None, ) out_image = gr.Image(label="Output") with gr.Column(): gr.Markdown( """ # Edit領域の指定ドラッグで編集対象のマスクの領域を指定してください。 """) input_mic = gr.HTML(html) btn = gr.Button(value="Image Edit") rect_text = gr.Textbox(elem_id="rectangle", visible=False) in_image.change(None, inputs=None, outputs=None, _js=image_change) btn.click(create_edit, inputs=[in_image, rect_text, prompt_text], outputs=[out_image]) demo.load(_js=scripts) demo.launch(share=False, auth=(app_username, app_password))