Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import trafilatura | |
| from transformers import pipeline | |
| import pytesseract | |
| from PIL import Image | |
| import requests | |
| from io import BytesIO | |
| # Load model | |
| reviewer = pipeline("text2text-generation", model="google/flan-t5-base") | |
| # OCR | |
| def extract_text_from_image_url(img_url): | |
| try: | |
| response = requests.get(img_url) | |
| img = Image.open(BytesIO(response.content)) | |
| text = pytesseract.image_to_string(img) | |
| return text | |
| except Exception as e: | |
| return f"❌ OCR Error: {e}" | |
| # Extract blog | |
| def extract_text_from_url(url): | |
| downloaded = trafilatura.fetch_url(url) | |
| if downloaded: | |
| return trafilatura.extract(downloaded) | |
| else: | |
| return "❌ Blog Error: Could not fetch content from the URL." | |
| # Review line-by-line | |
| def review_lines(text): | |
| lines = text.strip().split('\n') | |
| suggestions = [] | |
| for line in lines: | |
| if line.strip() == "": | |
| continue | |
| prompt = f"Rewrite this to fix grammar, tone, and remove any offensive language:\n\n{line}" | |
| suggestion = reviewer(prompt, max_new_tokens=100)[0]['generated_text'] | |
| suggestions.append((line, suggestion.strip())) | |
| return suggestions | |
| # Finalize by combining accepted lines | |
| def finalize_text(originals, suggestions, decisions): | |
| final = [] | |
| for orig, sugg, keep in zip(originals, suggestions, decisions): | |
| final.append(sugg if keep else orig) | |
| return "\n".join(final) | |
| # Gradio UI | |
| with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue")) as app: | |
| gr.Markdown("## ✨ BlogChecker AI\nSmart AI reviewer for blog content, with interactive approval and OCR image support.") | |
| with gr.Row(): | |
| blog_url = gr.Textbox(label="📎 Blog URL") | |
| image_url = gr.Textbox(label="🖼️ Image URL (optional)") | |
| extract_btn = gr.Button("🔍 Extract") | |
| combined_text = gr.Textbox(label="📝 Combined Blog + OCR Text", lines=10) | |
| with gr.Row(): | |
| review_btn = gr.Button("🧠 Review Content") | |
| finalize_btn = gr.Button("✅ Finalize Clean Blog") | |
| line_outputs = [] | |
| decisions = [] | |
| originals = [] | |
| suggestions = [] | |
| with gr.Column() as dynamic_review_section: | |
| for i in range(5): # support 5 lines for demo; can be dynamic later | |
| orig = gr.Textbox(label=f"Original Line {i+1}", interactive=False) | |
| sugg = gr.Textbox(label=f"Suggested Line {i+1}", interactive=False) | |
| accept = gr.Checkbox(label="✅ Accept Suggestion") | |
| originals.append(orig) | |
| suggestions.append(sugg) | |
| decisions.append(accept) | |
| line_outputs.append((orig, sugg, accept)) | |
| final_output = gr.Textbox(label="📦 Final Clean Blog", lines=10) | |
| def extract_both(url, img_url): | |
| blog = extract_text_from_url(url) | |
| ocr = extract_text_from_image_url(img_url) if img_url else "" | |
| return blog + ("\n" + ocr if ocr else "") | |
| def process_review(text): | |
| results = review_lines(text) | |
| outputs = [] | |
| for i in range(len(line_outputs)): | |
| if i < len(results): | |
| outputs.extend([results[i][0], results[i][1], False]) # orig, suggestion, unselected | |
| else: | |
| outputs.extend(["", "", False]) # clear unused slots | |
| return outputs | |
| def collect_decisions(*args): | |
| count = len(args) // 3 | |
| origs = args[:count] | |
| suggs = args[count:2*count] | |
| accepts = args[2*count:] | |
| return finalize_text(origs, suggs, accepts) | |
| # Wire actions | |
| extract_btn.click(fn=extract_both, inputs=[blog_url, image_url], outputs=combined_text) | |
| review_btn.click(fn=process_review, inputs=combined_text, outputs=[el for group in line_outputs for el in group]) | |
| finalize_btn.click(fn=collect_decisions, inputs=originals + suggestions + decisions, outputs=final_output) | |
| app.launch() | |