""" Viral Images — Image Scoring System Upload an image, score it, and improve it. Deployed on HuggingFace Spaces: https://huggingface.co/spaces/Babajaan/viral-images """ import os import sys import time import traceback from pathlib import Path # ── Compatibility shim: gradio 5.0.0 + huggingface_hub >= 0.27.0 ── # HfFolder was removed in huggingface_hub 0.27+ but gradio 5.0.0 still imports it. # We inject a minimal stub directly into huggingface_hub's module dict BEFORE gradio loads. try: import sys import huggingface_hub hf_mod = sys.modules.get("huggingface_hub") or huggingface_hub if "HfFolder" not in hf_mod.__dict__: class _HfFolderCompat: @staticmethod def get_token(): return os.environ.get("HF_TOKEN", None) @staticmethod def save_token(token): pass @staticmethod def delete_token(): pass # Inject into the real module dict so `from huggingface_hub import HfFolder` works hf_mod.__dict__["HfFolder"] = _HfFolderCompat() if hasattr(hf_mod, "__all__") and "HfFolder" not in hf_mod.__all__: hf_mod.__all__.append("HfFolder") print("[ViralImages] HfFolder compatibility shim applied.") else: print("[ViralImages] HfFolder already present, no shim needed.") except Exception as e: print(f"[ViralImages] HfFolder shim failed (non-fatal): {e}") # Ensure our modules are importable sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) import gradio as gr import pandas as pd from PIL import Image from utils.preprocessing import preprocess_image, validate_image from utils.formatting import ( format_overall_score, format_sub_scores, format_strengths, format_weaknesses, format_suggestions, format_projected_improvement, format_comparison, ) from scoring.engine import ScoringEngine, ScoreResponse from recommendations.engine import RecommendationEngine, compute_improvement_potential from compare.comparator import compare_scores from models.loader import ModelLoader # ─────────────────────────────────────────────────────────────── # Global state (models loaded once at startup) # ─────────────────────────────────────────────────────────────── MODEL_LOADER = None SCORING_ENGINE = None REC_ENGINE = None def _init_engines(): """Initialize scoring engines on startup.""" global MODEL_LOADER, SCORING_ENGINE, REC_ENGINE print("[ViralImages] Initializing engines...") t0 = time.time() MODEL_LOADER = ModelLoader() # Try to find config files config_paths = { "scoring": None, "rules": None, "model": None, } for base in ["configs", "viral-images/configs", os.path.join(os.path.dirname(__file__), "configs")]: for key, fname in [("scoring", "scoring_weights.yaml"), ("rules", "suggestion_rules.yaml"), ("model", "model_config.yaml")]: if config_paths[key] is None: p = os.path.join(base, fname) if os.path.exists(p): config_paths[key] = p SCORING_ENGINE = ScoringEngine( model_loader=MODEL_LOADER, config_path=config_paths["scoring"] ) REC_ENGINE = RecommendationEngine( rules_path=config_paths["rules"] ) # Pre-warm (load CLIP if available) try: SCORING_ENGINE.warmup() print(f"[ViralImages] Engines ready in {time.time()-t0:.1f}s") except Exception as e: print(f"[ViralImages] Warmup failed: {e}") print("[ViralImages] Running in fallback mode (no CLIP).") # Initialize on module load _init_engines() # ─────────────────────────────────────────────────────────────── # Score Image Tab # ─────────────────────────────────────────────────────────────── def score_image(image_input, concept, audience, use_case, detail_level): """ Score an uploaded image. Returns: (overall_html, scores_df, strengths_text, weaknesses_text, suggestions_text, projected_html) """ if image_input is None: return ( "
⚠️ Please upload an image first.
", pd.DataFrame(), "", "", "", "
No image to analyze.
" ) try: img = preprocess_image(image_input) val_error = validate_image(img) if val_error: return ( f"
⚠️ {val_error}
", pd.DataFrame(), "", "", "", "
Image validation failed.
" ) # Score response = SCORING_ENGINE.score(img, concept or "", audience, use_case) # Generate suggestions suggestions = REC_ENGINE.generate( response.sub_scores, response.raw_features, use_case ) response.suggestions = [ { "priority": s.priority, "sub_score_target": s.sub_score_target, "message": s.message, "projected_gain": s.projected_gain, "trigger_feature": s.trigger_feature, "trigger_value": s.trigger_value, } for s in suggestions ] # Projected improvement projected = REC_ENGINE.estimate_improvement( response.sub_scores, suggestions, use_case ) response.projected_improvement = projected # Format outputs overall_html = format_overall_score(response.overall_score) scores_df = format_sub_scores(response.sub_scores) strengths_html = format_strengths(response.strengths) weaknesses_html = format_weaknesses(response.weaknesses) suggestions_html = format_suggestions(response.suggestions) projected_html = format_projected_improvement( response.overall_score, projected ) # Metadata footer meta = response.metadata meta_html = f"
⏱ {meta.get('processing_time_ms', '?')}ms | Mode: {meta.get('neural_richness_mode', '?')} | Confidence: {response.confidence:.0%}
" return ( overall_html + meta_html, scores_df, strengths_html, weaknesses_html, suggestions_html, projected_html, ) except Exception as e: traceback.print_exc() return ( f"
⚠️ Error during scoring:
{str(e)}
", pd.DataFrame(), "", "", "", "
Scoring failed.
" ) # ─────────────────────────────────────────────────────────────── # Compare Mode Tab # ─────────────────────────────────────────────────────────────── def compare_images(original_img, revised_img, concept, audience, use_case): """ Compare before/after images. Returns: (original_scores_label, revised_scores_label, comparison_html, next_steps_text) """ if original_img is None or revised_img is None: return ( {}, {}, "
⚠️ Please upload both images.
", "Upload both original and revised images to compare." ) try: # Score both original = SCORING_ENGINE.score(original_img, concept or "", audience, use_case) revised = SCORING_ENGINE.score(revised_img, concept or "", audience, use_case) # Compare result = compare_scores(original, revised) # Format orig_label = {k: v / 100 for k, v in original.sub_scores.items()} rev_label = {k: v / 100 for k, v in revised.sub_scores.items()} comp_html = format_comparison_result(result) next_steps = "\n".join( f"• {rec['message']}" for rec in result["next_recommendations"] ) if result["next_recommendations"] else "No further improvements needed — your revised image scores well!" return ( orig_label, rev_label, comp_html, next_steps, ) except Exception as e: traceback.print_exc() return ( {}, {}, f"
⚠️ Error:
{str(e)}
", "Comparison failed. Please try again." ) def format_comparison_result(result: dict) -> str: """Format comparison as HTML.""" orig = result["original_overall"] rev = result["revised_overall"] delta = result["delta_overall"] color = "#27ae60" if delta > 0 else "#e74c3c" if delta < 0 else "#999" emoji = "📈" if delta > 0 else "📉" if delta < 0 else "➡️" html = f"""
Original
{orig:.0f}
→
Revised
{rev:.0f}
{emoji} Overall: {abs(delta):.0f} point{'s' if abs(delta) != 1 else ''} {'gained' if delta > 0 else 'lost' if delta < 0 else 'unchanged'}
""" if result["improvements"]: html += "
✅ Improvements:
" if result["regressions"]: html += "
⚠️ Regressions:
" if result["next_recommendations"]: html += "
📋 Next Steps:
" for rec in result["next_recommendations"]: html += f"
{rec['message']}
" html += "
" html += "
" return html # ─────────────────────────────────────────────────────────────── # Gradio App # ─────────────────────────────────────────────────────────────── ABOUT_TEXT = """ ## 🧠 About Viral Images **Viral Images** scores any image across 8 dimensions and suggests concrete improvements. ### What It Measures | Dimension | Description | |-----------|-------------| | **Concept Match** | Does the image convey your intended topic? | | **Visual Focus** | Is there a clear focal point that draws the eye? | | **Readability** | Is text legible and well-sized? | | **Complexity Balance** | Is detail level appropriate — not empty, not cluttered? | | **Communication Clarity** | Is the visual hierarchy clear and well-organized? | | **Predicted Neural Richness** | How visually engaging is the image? | | **Memorability** | Will viewers remember it after one viewing? | | **Improvement Potential** | How much room for improvement remains? | ### How to Use 1. **Upload an image** (JPG, PNG, or any common format) 2. **Describe the concept** — what should this image communicate? 3. **Select audience and use case** 4. Click **"Score Image"** 5. Review scores, strengths, weaknesses, and suggestions ### Compare Mode Upload an **original** and a **revised** version to see what improved and what to focus on next. ### ⚠️ Important Disclaimer - Scores are **AI-predicted proxies**, not measurements of real brain activity. - Scores do **not guarantee engagement, virality, or aesthetic quality**. - Use as **directional feedback** to complement your own judgment. - Predicted Neural Richness is estimated via computational proxy models, not from actual fMRI data. ### Technical Details Models used: CLIP ViT-B/32 (concept matching), heuristic analysis (OpenCV), saliency heuristics, OCR heuristics, and aesthetic proxies. Neural richness is estimated via proxy, not measured. """ def build_app(): """Build and return the Gradio app.""" with gr.Blocks() as demo: gr.Markdown( """ # 🧠 Viral Images ### *Upload an image, score it, and improve it.* """ ) # ── TAB 1: Score Image ── with gr.Tab("📊 Score Image"): with gr.Row(): # Input column with gr.Column(scale=2): image_input = gr.Image( type="pil", label="📷 Upload Image", sources=["upload", "clipboard"], height=300, ) concept_input = gr.Textbox( label="💡 Concept / Theme", placeholder="e.g., photosynthesis process, product launch, data visualization...", lines=1, ) audience_input = gr.Dropdown( choices=[ "General", "Children (K-8)", "High School Students", "College Students", "Professionals", "Researchers", "General Public", "Social Media Audience", ], label="👥 Target Audience", value="General", ) usecase_input = gr.Dropdown( choices=[ ("Social Media Post", "social_media"), ("Thumbnail", "thumbnail"), ("Educational", "educational"), ("Scientific Figure", "scientific_figure"), ("Infographic", "infographic"), ("Presentation Slide", "presentation_slide"), ("Marketing", "marketing"), ("General", "default"), ], label="🎯 Use Case", value="social_media", ) score_btn = gr.Button( "🔍 Score Image", variant="primary", size="lg", ) with gr.Accordion("⚙️ Advanced Options", open=False): detail_slider = gr.Slider( 1, 5, value=3, step=1, label="Detail Level", ) # Output column with gr.Column(scale=3): overall_score = gr.HTML( label="Overall Score", value="
Upload an image and click 'Score Image' to see results.
" ) with gr.Row(): strengths_box = gr.HTML( label="✅ Strengths", value="", ) weaknesses_box = gr.HTML( label="⚠️ Weaknesses", value="", ) score_chart = gr.BarPlot( x="Sub-Score", y="Score", color="Sub-Score", title="Score Breakdown", y_lim=[0, 100], height=280, value=pd.DataFrame({"Sub-Score": [], "Score": []}), ) suggestions_box = gr.HTML( label="💡 Improvement Suggestions", value="", ) projected_html = gr.HTML( label="Projected Improvement", value="", ) score_btn.click( fn=score_image, inputs=[image_input, concept_input, audience_input, usecase_input, detail_slider], outputs=[overall_score, score_chart, strengths_box, weaknesses_box, suggestions_box, projected_html], concurrency_id="inference", concurrency_limit=1, ) # ── TAB 2: Compare Mode ── with gr.Tab("🔄 Compare Mode"): with gr.Row(): before_img = gr.Image( type="pil", label="📷 Original Image", sources=["upload"], height=250, ) after_img = gr.Image( type="pil", label="📷 Revised Image", sources=["upload"], height=250, ) with gr.Row(): compare_concept = gr.Textbox( label="💡 Concept", placeholder="Same concept as original...", lines=1, ) compare_audience = gr.Dropdown( choices=[ "General", "Children (K-8)", "High School Students", "College Students", "Professionals", "Researchers", "General Public", "Social Media Audience", ], label="👥 Audience", value="General", ) compare_usecase = gr.Dropdown( choices=[ ("Social Media Post", "social_media"), ("Thumbnail", "thumbnail"), ("Educational", "educational"), ("Scientific Figure", "scientific_figure"), ("Infographic", "infographic"), ("Presentation Slide", "presentation_slide"), ("Marketing", "marketing"), ("General", "default"), ], label="🎯 Use Case", value="social_media", ) compare_btn = gr.Button( "🔄 Compare", variant="primary", size="lg", ) with gr.Row(): before_scores = gr.Label( label="Original Scores", value={}, ) after_scores = gr.Label( label="Revised Scores", value={}, ) comparison_html = gr.HTML( label="Comparison Results", value="
Upload both images to compare.
" ) next_steps = gr.HTML( label="📋 Next Recommended Edits", value="", ) compare_btn.click( fn=compare_images, inputs=[before_img, after_img, compare_concept, compare_audience, compare_usecase], outputs=[before_scores, after_scores, comparison_html, next_steps], concurrency_id="inference", concurrency_limit=1, ) # ── TAB 3: About ── with gr.Tab("ℹ️ About"): gr.Markdown(ABOUT_TEXT) # Footer gr.Markdown( """
⚠️ Disclaimer: Scores are AI-predicted proxies, not measurements of brain activity or guaranteed engagement. Use as directional feedback, not absolute truth. TRIBE is used only as inspiration for the neural richness proxy.
""" ) return demo # ─────────────────────────────────────────────────────────────── # Launch # ─────────────────────────────────────────────────────────────── if __name__ == "__main__": demo = build_app() demo.queue(max_size=10, api_open=False).launch( show_api=False, server_name="0.0.0.0", server_port=int(os.environ.get("PORT", 7860)), )