AI-Talent-Force/dev_caio
0
1"""2ShortSmith v2 - Gradio Application3 4Hugging Face Space interface for video highlight extraction.5Features:6- Multi-modal analysis (visual + audio + motion)7- Domain-optimized presets8- Person-specific filtering (optional)9- Scene-aware clip cutting10- Batch testing with parameter variations11"""12 13import os14import sys15import tempfile16import shutil17import json18import zipfile19from pathlib import Path20import time21import traceback22from typing import List, Dict, Any, Optional23 24import gradio as gr25import pandas as pd26 27# Add project root to path28sys.path.insert(0, str(Path(__file__).parent))29 30# Initialize logging31try:32 from utils.logger import setup_logging, get_logger33 setup_logging(log_level="INFO", log_to_console=True)34 logger = get_logger("app")35except Exception:36 import logging37 logging.basicConfig(level=logging.INFO)38 logger = logging.getLogger("app")39 40 41# =============================================================================42# Shared Utilities43# =============================================================================44 45def build_metrics_output(result, domain: str, custom_prompt: Optional[str] = None) -> str:46 """47 Build formatted metrics output for testing and evaluation.48 49 Args:50 result: PipelineResult object51 domain: Content domain used for processing52 custom_prompt: Custom prompt used (if any)53 54 Returns:55 Formatted string with all metrics56 """57 lines = []58 lines.append("=" * 50)59 lines.append("AUTOMATED METRICS (System-Generated)")60 lines.append("=" * 50)61 lines.append("")62 63 # Processing Metrics64 lines.append("PROCESSING METRICS")65 lines.append("-" * 30)66 lines.append(f"processing_time_seconds: {result.processing_time:.2f}")67 lines.append(f"frames_analyzed: {len(result.visual_features)}")68 lines.append(f"scenes_detected: {len(result.scenes)}")69 lines.append(f"audio_segments_analyzed: {len(result.audio_features)}")70 lines.append(f"domain: {domain}")71 lines.append(f"custom_prompt: {custom_prompt if custom_prompt else 'none'}")72 73 # Count hooks from scores (estimate based on high-scoring segments)74 hooks_detected = sum(1 for s in result.scores if s.combined_score > 0.7) if result.scores else 075 lines.append(f"hooks_detected: {hooks_detected}")76 77 if result.metadata:78 lines.append(f"video_duration_seconds: {result.metadata.duration:.2f}")79 lines.append(f"video_resolution: {result.metadata.resolution}")80 lines.append(f"video_fps: {result.metadata.fps:.2f}")81 82 lines.append("")83 84 # Per Clip Metrics85 lines.append("PER CLIP METRICS")86 lines.append("-" * 30)87 88 for i, clip in enumerate(result.clips):89 lines.append("")90 lines.append(f"[Clip {i + 1}]")91 lines.append(f" clip_id: {i + 1}")92 lines.append(f" start_time: {clip.start_time:.2f}")93 lines.append(f" end_time: {clip.end_time:.2f}")94 lines.append(f" duration: {clip.duration:.2f}")95 lines.append(f" hype_score: {clip.hype_score:.4f}")96 lines.append(f" visual_score: {clip.visual_score:.4f}")97 lines.append(f" audio_score: {clip.audio_score:.4f}")98 lines.append(f" motion_score: {clip.motion_score:.4f}")99 100 # Hook info - derive from segment scores if available101 hook_type = "none"102 hook_confidence = 0.0103 104 # Find matching segment score for this clip105 for score in result.scores:106 if abs(score.start_time - clip.start_time) < 1.0:107 if score.combined_score > 0.7:108 hook_confidence = score.combined_score109 # Infer hook type based on dominant score110 if score.audio_score > score.visual_score and score.audio_score > score.motion_score:111 hook_type = "audio_peak"112 elif score.motion_score > score.visual_score:113 hook_type = "motion_spike"114 else:115 hook_type = "visual_highlight"116 break117 118 lines.append(f" hook_type: {hook_type}")119 lines.append(f" hook_confidence: {hook_confidence:.4f}")120 121 if clip.person_detected:122 lines.append(f" person_detected: True")123 lines.append(f" person_screen_time: {clip.person_screen_time:.4f}")124 125 lines.append("")126 lines.append("=" * 50)127 lines.append("END METRICS")128 lines.append("=" * 50)129 130 return "\n".join(lines)131 132 133# =============================================================================134# Single Video Processing135# =============================================================================136 137def process_video(138 video_file,139 domain,140 num_clips,141 clip_duration,142 reference_image,143 custom_prompt,144 progress=gr.Progress()145):146 """147 Main video processing function for single video mode.148 149 Args:150 video_file: Uploaded video file path151 domain: Content domain for scoring weights152 num_clips: Number of clips to extract153 clip_duration: Duration of each clip in seconds154 reference_image: Optional reference image for person filtering155 custom_prompt: Optional custom instructions156 progress: Gradio progress tracker157 158 Returns:159 Tuple of (status_message, clip1, clip2, clip3, log_text, metrics_text)160 """161 if video_file is None:162 return "Please upload a video first.", None, None, None, "", ""163 164 log_messages = []165 166 def log(msg):167 log_messages.append(f"[{time.strftime('%H:%M:%S')}] {msg}")168 logger.info(msg)169 170 try:171 video_path = Path(video_file)172 log(f"Processing video: {video_path.name}")173 progress(0.05, desc="Validating video...")174 175 # Import pipeline components176 from utils.helpers import validate_video_file, validate_image_file, format_duration177 from pipeline.orchestrator import PipelineOrchestrator178 179 # Validate video180 validation = validate_video_file(video_file)181 if not validation.is_valid:182 return f"Error: {validation.error_message}", None, None, None, "\n".join(log_messages), ""183 184 log(f"Video size: {validation.file_size / (1024*1024):.1f} MB")185 186 # Validate reference image if provided187 ref_path = None188 if reference_image is not None:189 ref_validation = validate_image_file(reference_image)190 if ref_validation.is_valid:191 ref_path = reference_image192 log(f"Reference image: {Path(reference_image).name}")193 else:194 log(f"Warning: Invalid reference image - {ref_validation.error_message}")195 196 # Map domain string to internal value197 domain_map = {198 "Sports": "sports",199 "Vlogs": "vlogs",200 "Music Videos": "music",201 "Podcasts": "podcasts",202 "Gaming": "gaming",203 "General": "general",204 }205 domain_value = domain_map.get(domain, "general")206 log(f"Domain: {domain_value}")207 208 # Create output directory209 output_dir = Path(tempfile.mkdtemp(prefix="shortsmith_output_"))210 log(f"Output directory: {output_dir}")211 212 # Progress callback to update UI during processing213 def on_progress(pipeline_progress):214 stage = pipeline_progress.stage.value215 pct = pipeline_progress.progress216 msg = pipeline_progress.message217 log(f"[{stage}] {msg}")218 # Map pipeline progress (0-1) to our range (0.1-0.9)219 mapped_progress = 0.1 + (pct * 0.8)220 progress(mapped_progress, desc=f"{stage}: {msg}")221 222 # Initialize pipeline223 progress(0.1, desc="Initializing AI models...")224 log("Initializing pipeline...")225 pipeline = PipelineOrchestrator(progress_callback=on_progress)226 227 # Process video228 progress(0.15, desc="Starting analysis...")229 log(f"Processing: {int(num_clips)} clips @ {int(clip_duration)}s each")230 231 result = pipeline.process(232 video_path=video_path,233 num_clips=int(num_clips),234 clip_duration=float(clip_duration),235 domain=domain_value,236 reference_image=ref_path,237 custom_prompt=custom_prompt.strip() if custom_prompt else None,238 )239 240 progress(0.9, desc="Extracting clips...")241 242 # Handle result243 if result.success:244 log(f"Processing complete in {result.processing_time:.1f}s")245 246 clip_paths = []247 for i, clip in enumerate(result.clips):248 if clip.clip_path.exists():249 output_path = output_dir / f"highlight_{i+1}.mp4"250 shutil.copy2(clip.clip_path, output_path)251 clip_paths.append(str(output_path))252 log(f"Clip {i+1}: {format_duration(clip.start_time)} - {format_duration(clip.end_time)} (score: {clip.hype_score:.2f})")253 254 status = f"Successfully extracted {len(clip_paths)} highlight clips!\nProcessing time: {result.processing_time:.1f}s"255 256 # Build metrics output257 metrics_output = build_metrics_output(result, domain_value, custom_prompt.strip() if custom_prompt else None)258 259 pipeline.cleanup()260 progress(1.0, desc="Done!")261 262 # Return up to 3 clips263 clip1 = clip_paths[0] if len(clip_paths) > 0 else None264 clip2 = clip_paths[1] if len(clip_paths) > 1 else None265 clip3 = clip_paths[2] if len(clip_paths) > 2 else None266 267 return status, clip1, clip2, clip3, "\n".join(log_messages), metrics_output268 else:269 log(f"Processing failed: {result.error_message}")270 pipeline.cleanup()271 return f"Error: {result.error_message}", None, None, None, "\n".join(log_messages), ""272 273 except Exception as e:274 error_msg = f"Unexpected error: {str(e)}"275 log(error_msg)276 log(traceback.format_exc())277 logger.exception("Pipeline error")278 return error_msg, None, None, None, "\n".join(log_messages), ""279 280 281# =============================================================================282# Batch Testing Functions283# =============================================================================284 285def generate_test_queue(286 videos: List[str],287 domains: List[str],288 durations: List[int],289 num_clips: int,290 ref_image: Optional[str],291 prompts: List[str],292 include_no_prompt: bool293) -> List[Dict[str, Any]]:294 """Generate all parameter combinations to test (cartesian product)."""295 # Build prompt list296 prompt_list = []297 if include_no_prompt:298 prompt_list.append(None) # No prompt baseline299 prompt_list.extend([p.strip() for p in prompts if p and p.strip()])300 301 # If no prompts at all, use just None302 if not prompt_list:303 prompt_list = [None]304 305 # Map domain display names to internal values306 domain_map = {307 "Sports": "sports",308 "Vlogs": "vlogs",309 "Music Videos": "music",310 "Podcasts": "podcasts",311 "Gaming": "gaming",312 "General": "general",313 }314 315 queue = []316 test_id = 1317 for video in videos:318 video_name = Path(video).name if video else "unknown"319 for domain in domains:320 domain_value = domain_map.get(domain, "general")321 for duration in durations:322 for prompt in prompt_list:323 queue.append({324 "test_id": test_id,325 "video_path": video,326 "video_name": video_name,327 "domain": domain,328 "domain_value": domain_value,329 "clip_duration": duration,330 "num_clips": num_clips,331 "reference_image": ref_image,332 "custom_prompt": prompt,333 })334 test_id += 1335 return queue336 337 338def run_single_batch_test(config: Dict[str, Any], output_base_dir: Path) -> Dict[str, Any]:339 """Run a single test from the batch queue."""340 from utils.helpers import validate_video_file341 from pipeline.orchestrator import PipelineOrchestrator342 343 test_id = config["test_id"]344 video_path = config["video_path"]345 video_name = config["video_name"]346 domain_value = config["domain_value"]347 duration = config["clip_duration"]348 num_clips = config["num_clips"]349 ref_image = config["reference_image"]350 custom_prompt = config["custom_prompt"]351 352 # Create unique output folder for this test353 prompt_suffix = "no_prompt" if not custom_prompt else f"prompt_{hash(custom_prompt) % 1000}"354 test_folder = f"{Path(video_name).stem}_{domain_value}_{duration}s_{prompt_suffix}"355 output_dir = output_base_dir / test_folder356 output_dir.mkdir(parents=True, exist_ok=True)357 358 result_data = {359 "test_id": test_id,360 "video_name": video_name,361 "domain": domain_value,362 "clip_duration": duration,363 "custom_prompt": custom_prompt if custom_prompt else "none",364 "num_clips": num_clips,365 "status": "failed",366 "error": None,367 "processing_time": 0,368 "frames_analyzed": 0,369 "scenes_detected": 0,370 "hooks_detected": 0,371 "clips": [],372 "clip_paths": [],373 }374 375 try:376 # Validate video377 validation = validate_video_file(video_path)378 if not validation.is_valid:379 result_data["error"] = validation.error_message380 return result_data381 382 # Initialize and run pipeline383 pipeline = PipelineOrchestrator()384 result = pipeline.process(385 video_path=video_path,386 num_clips=num_clips,387 clip_duration=float(duration),388 domain=domain_value,389 reference_image=ref_image,390 custom_prompt=custom_prompt,391 )392 393 if result.success:394 result_data["status"] = "success"395 result_data["processing_time"] = round(result.processing_time, 2)396 result_data["frames_analyzed"] = len(result.visual_features)397 result_data["scenes_detected"] = len(result.scenes)398 result_data["hooks_detected"] = sum(1 for s in result.scores if s.combined_score > 0.7) if result.scores else 0399 400 # Copy clips and collect data401 for i, clip in enumerate(result.clips):402 if clip.clip_path.exists():403 clip_output = output_dir / f"clip_{i+1}.mp4"404 shutil.copy2(clip.clip_path, clip_output)405 result_data["clip_paths"].append(str(clip_output))406 407 # Find hook type for this clip408 hook_type = "none"409 hook_confidence = 0.0410 for score in result.scores:411 if abs(score.start_time - clip.start_time) < 1.0:412 if score.combined_score > 0.7:413 hook_confidence = score.combined_score414 if score.audio_score > score.visual_score and score.audio_score > score.motion_score:415 hook_type = "audio_peak"416 elif score.motion_score > score.visual_score:417 hook_type = "motion_spike"418 else:419 hook_type = "visual_highlight"420 break421 422 result_data["clips"].append({423 "clip_id": i + 1,424 "start_time": round(clip.start_time, 2),425 "end_time": round(clip.end_time, 2),426 "duration": round(clip.duration, 2),427 "hype_score": round(clip.hype_score, 4),428 "visual_score": round(clip.visual_score, 4),429 "audio_score": round(clip.audio_score, 4),430 "motion_score": round(clip.motion_score, 4),431 "hook_type": hook_type,432 "hook_confidence": round(hook_confidence, 4),433 })434 else:435 result_data["error"] = result.error_message436 437 pipeline.cleanup()438 439 except Exception as e:440 result_data["error"] = str(e)441 logger.exception(f"Batch test {test_id} failed")442 443 return result_data444 445 446def results_to_dataframe(results: List[Dict[str, Any]]) -> pd.DataFrame:447 """Convert batch results to a pandas DataFrame for display."""448 rows = []449 for r in results:450 row = {451 "Test ID": r["test_id"],452 "Video": r["video_name"],453 "Domain": r["domain"],454 "Duration": f"{r['clip_duration']}s",455 "Prompt": r["custom_prompt"][:20] + "..." if len(r["custom_prompt"]) > 20 else r["custom_prompt"],456 "Status": r["status"],457 "Time (s)": r["processing_time"],458 "Frames": r["frames_analyzed"],459 "Hooks": r["hooks_detected"],460 }461 # Add clip scores462 for i, clip in enumerate(r.get("clips", [])[:3]):463 row[f"Clip {i+1} Hype"] = clip.get("hype_score", 0)464 rows.append(row)465 return pd.DataFrame(rows)466 467 468def results_to_csv(results: List[Dict[str, Any]]) -> str:469 """Convert results to CSV format."""470 rows = []471 for r in results:472 row = {473 "test_id": r["test_id"],474 "video_name": r["video_name"],475 "domain": r["domain"],476 "clip_duration": r["clip_duration"],477 "custom_prompt": r["custom_prompt"],478 "num_clips": r["num_clips"],479 "status": r["status"],480 "error": r.get("error", ""),481 "processing_time": r["processing_time"],482 "frames_analyzed": r["frames_analyzed"],483 "scenes_detected": r["scenes_detected"],484 "hooks_detected": r["hooks_detected"],485 }486 # Add per-clip data487 for i in range(3):488 if i < len(r.get("clips", [])):489 clip = r["clips"][i]490 row[f"clip_{i+1}_start"] = clip["start_time"]491 row[f"clip_{i+1}_end"] = clip["end_time"]492 row[f"clip_{i+1}_hype"] = clip["hype_score"]493 row[f"clip_{i+1}_visual"] = clip["visual_score"]494 row[f"clip_{i+1}_audio"] = clip["audio_score"]495 row[f"clip_{i+1}_motion"] = clip["motion_score"]496 row[f"clip_{i+1}_hook_type"] = clip["hook_type"]497 else:498 row[f"clip_{i+1}_start"] = ""499 row[f"clip_{i+1}_end"] = ""500 row[f"clip_{i+1}_hype"] = ""501 row[f"clip_{i+1}_visual"] = ""502 row[f"clip_{i+1}_audio"] = ""503 row[f"clip_{i+1}_motion"] = ""504 row[f"clip_{i+1}_hook_type"] = ""505 rows.append(row)506 507 df = pd.DataFrame(rows)508 return df.to_csv(index=False)509 510 511def results_to_json(results: List[Dict[str, Any]]) -> str:512 """Convert results to JSON format."""513 # Remove clip_paths from export (they're temp files)514 export_results = []515 for r in results:516 r_copy = r.copy()517 r_copy.pop("clip_paths", None)518 export_results.append(r_copy)519 return json.dumps(export_results, indent=2)520 521 522def create_clips_zip(results: List[Dict[str, Any]]) -> Optional[str]:523 """Create a ZIP file of all extracted clips."""524 zip_path = Path(tempfile.mkdtemp()) / "batch_clips.zip"525 526 with zipfile.ZipFile(zip_path, 'w', zipfile.ZIP_DEFLATED) as zf:527 for r in results:528 if r["status"] == "success":529 folder_name = f"{Path(r['video_name']).stem}_{r['domain']}_{r['clip_duration']}s"530 if r["custom_prompt"] != "none":531 folder_name += f"_prompt"532 for clip_path in r.get("clip_paths", []):533 if Path(clip_path).exists():534 arcname = f"{folder_name}/{Path(clip_path).name}"535 zf.write(clip_path, arcname)536 537 return str(zip_path) if zip_path.exists() else None538 539 540# Batch state (module level for simplicity)541batch_state = {542 "is_running": False,543 "should_cancel": False,544 "results": [],545 "output_dir": None,546}547 548 549def run_batch_tests(550 videos,551 domains,552 durations,553 num_clips,554 reference_image,555 include_no_prompt,556 prompt1,557 prompt2,558 prompt3,559 progress=gr.Progress()560):561 """Main batch testing function."""562 global batch_state563 564 # Validate inputs565 if not videos:566 return "Please upload at least one video.", None, "", "", None, None, None567 568 if not domains:569 return "Please select at least one domain.", None, "", "", None, None, None570 571 if not durations:572 return "Please select at least one duration.", None, "", "", None, None, None573 574 # Collect prompts575 prompts = [p for p in [prompt1, prompt2, prompt3] if p and p.strip()]576 577 # Generate test queue578 queue = generate_test_queue(579 videos=videos,580 domains=domains,581 durations=durations,582 num_clips=int(num_clips),583 ref_image=reference_image,584 prompts=prompts,585 include_no_prompt=include_no_prompt,586 )587 588 if not queue:589 return "No tests to run. Please check your configuration.", None, "", "", None, None, None590 591 # Initialize batch state592 batch_state["is_running"] = True593 batch_state["should_cancel"] = False594 batch_state["results"] = []595 batch_state["output_dir"] = Path(tempfile.mkdtemp(prefix="shortsmith_batch_"))596 597 total_tests = len(queue)598 log_messages = []599 600 def log(msg):601 log_messages.append(f"[{time.strftime('%H:%M:%S')}] {msg}")602 logger.info(msg)603 604 log(f"Starting batch testing: {total_tests} tests")605 log(f"Videos: {len(videos)}, Domains: {len(domains)}, Durations: {len(durations)}, Prompts: {len(prompts) + (1 if include_no_prompt else 0)}")606 607 # Run tests sequentially608 for i, test_config in enumerate(queue):609 if batch_state["should_cancel"]:610 log("Batch cancelled by user")611 break612 613 test_id = test_config["test_id"]614 video_name = test_config["video_name"]615 domain = test_config["domain_value"]616 duration = test_config["clip_duration"]617 prompt = test_config["custom_prompt"] or "no-prompt"618 619 log(f"[{i+1}/{total_tests}] Testing: {video_name} | {domain} | {duration}s | {prompt[:30]}...")620 progress((i + 1) / total_tests, desc=f"Test {i+1}/{total_tests}: {video_name}")621 622 # Run the test623 result = run_single_batch_test(test_config, batch_state["output_dir"])624 batch_state["results"].append(result)625 626 if result["status"] == "success":627 log(f" ✓ Completed in {result['processing_time']}s")628 else:629 log(f" ✗ Failed: {result.get('error', 'Unknown error')}")630 631 # Finalize632 batch_state["is_running"] = False633 completed = len([r for r in batch_state["results"] if r["status"] == "success"])634 failed = len([r for r in batch_state["results"] if r["status"] == "failed"])635 636 log(f"Batch complete: {completed} succeeded, {failed} failed")637 638 # Generate outputs639 results_df = results_to_dataframe(batch_state["results"])640 csv_content = results_to_csv(batch_state["results"])641 json_content = results_to_json(batch_state["results"])642 643 # Save CSV and JSON to files for download644 csv_path = batch_state["output_dir"] / "results.csv"645 json_path = batch_state["output_dir"] / "results.json"646 csv_path.write_text(csv_content)647 json_path.write_text(json_content)648 649 # Create ZIP of clips650 zip_path = create_clips_zip(batch_state["results"])651 652 status = f"Batch complete: {completed}/{total_tests} tests succeeded"653 654 return (655 status,656 results_df,657 "\n".join(log_messages),658 json_content,659 str(csv_path),660 str(json_path),661 zip_path,662 )663 664 665def cancel_batch():666 """Cancel the running batch."""667 global batch_state668 batch_state["should_cancel"] = True669 return "Cancelling batch... (will stop after current test completes)"670 671 672def calculate_queue_size(videos, domains, durations, include_no_prompt, prompt1, prompt2, prompt3):673 """Calculate and display the queue size."""674 num_videos = len(videos) if videos else 0675 num_domains = len(domains) if domains else 0676 num_durations = len(durations) if durations else 0677 678 prompts = [p for p in [prompt1, prompt2, prompt3] if p and p.strip()]679 num_prompts = len(prompts) + (1 if include_no_prompt else 0)680 if num_prompts == 0:681 num_prompts = 1 # Default to no-prompt if nothing selected682 683 total = num_videos * num_domains * num_durations * num_prompts684 685 return f"Queue: {num_videos} video(s) × {num_domains} domain(s) × {num_durations} duration(s) × {num_prompts} prompt(s) = **{total} tests**"686 687 688# =============================================================================689# Build Gradio Interface690# =============================================================================691 692with gr.Blocks(693 title="ShortSmith v2",694 theme=gr.themes.Soft(),695 css="""696 .container { max-width: 1200px; margin: auto; }697 .output-video { min-height: 200px; }698 """699) as demo:700 701 gr.Markdown("""702 # ShortSmith v2703 ### AI-Powered Video Highlight Extractor704 705 Upload a video and automatically extract the most engaging highlight clips using AI analysis.706 """)707 708 with gr.Tabs():709 # =================================================================710 # Tab 1: Single Video711 # =================================================================712 with gr.TabItem("Single Video"):713 with gr.Row():714 # Left column - Inputs715 with gr.Column(scale=1):716 gr.Markdown("### Input")717 718 video_input = gr.Video(719 label="Upload Video",720 sources=["upload"],721 )722 723 with gr.Accordion("Settings", open=True):724 domain_dropdown = gr.Dropdown(725 choices=["Sports", "Vlogs", "Music Videos", "Podcasts", "Gaming", "General"],726 value="General",727 label="Content Domain",728 info="Select the type of content for optimized scoring"729 )730 731 with gr.Row():732 num_clips_slider = gr.Slider(733 minimum=1,734 maximum=3,735 value=3,736 step=1,737 label="Number of Clips",738 info="How many highlight clips to extract"739 )740 duration_slider = gr.Slider(741 minimum=5,742 maximum=30,743 value=15,744 step=1,745 label="Clip Duration (seconds)",746 info="Target duration for each clip"747 )748 749 with gr.Accordion("Person Filtering (Optional)", open=False):750 reference_image = gr.Image(751 label="Reference Image",752 type="filepath",753 sources=["upload"],754 )755 gr.Markdown("*Upload a photo of a person to prioritize clips featuring them.*")756 757 with gr.Accordion("Custom Instructions (Optional)", open=False):758 custom_prompt = gr.Textbox(759 label="Additional Instructions",760 placeholder="E.g., 'Focus on crowd reactions' or 'Prioritize action scenes'",761 lines=2,762 )763 764 process_btn = gr.Button(765 "Extract Highlights",766 variant="primary",767 size="lg"768 )769 770 # Right column - Outputs771 with gr.Column(scale=1):772 gr.Markdown("### Output")773 774 status_output = gr.Textbox(775 label="Status",776 lines=2,777 interactive=False778 )779 780 gr.Markdown("#### Extracted Clips")781 clip1_output = gr.Video(label="Clip 1", elem_classes=["output-video"])782 clip2_output = gr.Video(label="Clip 2", elem_classes=["output-video"])783 clip3_output = gr.Video(label="Clip 3", elem_classes=["output-video"])784 785 with gr.Accordion("Processing Log", open=True):786 log_output = gr.Textbox(787 label="Log",788 lines=10,789 interactive=False,790 show_copy_button=True791 )792 793 with gr.Accordion("Automated Metrics (System-Generated)", open=True):794 metrics_output = gr.Textbox(795 label="Metrics for Testing",796 lines=20,797 interactive=False,798 show_copy_button=True,799 info="Copy these metrics for evaluation spreadsheets"800 )801 802 # Connect single video processing803 process_btn.click(804 fn=process_video,805 inputs=[806 video_input,807 domain_dropdown,808 num_clips_slider,809 duration_slider,810 reference_image,811 custom_prompt812 ],813 outputs=[814 status_output,815 clip1_output,816 clip2_output,817 clip3_output,818 log_output,819 metrics_output820 ],821 show_progress="full"822 )823 824 # =================================================================825 # Tab 2: Batch Testing826 # =================================================================827 with gr.TabItem("Batch Testing"):828 with gr.Row():829 # Left column - Configuration830 with gr.Column(scale=1):831 gr.Markdown("### Batch Configuration")832 833 batch_videos = gr.File(834 label="Upload Video(s)",835 file_count="multiple",836 file_types=["video"],837 )838 839 gr.Markdown("#### Domains to Test")840 batch_domains = gr.CheckboxGroup(841 choices=["Sports", "Vlogs", "Music Videos", "Podcasts", "Gaming", "General"],842 value=["General"],843 label="Select domains",844 )845 846 gr.Markdown("#### Clip Durations to Test")847 batch_durations = gr.CheckboxGroup(848 choices=[10, 15, 20, 30],849 value=[15],850 label="Select durations (seconds)",851 )852 853 batch_num_clips = gr.Slider(854 minimum=1,855 maximum=3,856 value=3,857 step=1,858 label="Number of Clips per Test",859 )860 861 with gr.Accordion("Custom Prompts", open=True):862 batch_no_prompt = gr.Checkbox(863 label="Include no-prompt baseline",864 value=True,865 info="Test without any custom prompt for comparison"866 )867 batch_prompt1 = gr.Textbox(868 label="Prompt 1",869 placeholder="E.g., 'Focus on action moments'",870 lines=1,871 )872 batch_prompt2 = gr.Textbox(873 label="Prompt 2",874 placeholder="E.g., 'Find crowd reactions'",875 lines=1,876 )877 batch_prompt3 = gr.Textbox(878 label="Prompt 3",879 placeholder="E.g., 'Prioritize emotional moments'",880 lines=1,881 )882 883 with gr.Accordion("Reference Image (Optional)", open=False):884 batch_ref_image = gr.Image(885 label="Reference Image (applies to all tests)",886 type="filepath",887 sources=["upload"],888 )889 890 # Queue size indicator891 queue_info = gr.Markdown("Queue: 0 tests")892 893 with gr.Row():894 batch_start_btn = gr.Button(895 "Start Batch",896 variant="primary",897 size="lg"898 )899 batch_cancel_btn = gr.Button(900 "Cancel",901 variant="secondary",902 size="lg"903 )904 905 # Right column - Results906 with gr.Column(scale=1):907 gr.Markdown("### Results")908 909 batch_status = gr.Textbox(910 label="Status",911 lines=2,912 interactive=False913 )914 915 batch_results_table = gr.Dataframe(916 label="Test Results",917 headers=["Test ID", "Video", "Domain", "Duration", "Prompt", "Status", "Time (s)", "Frames", "Hooks"],918 interactive=False,919 )920 921 with gr.Accordion("Processing Log", open=True):922 batch_log = gr.Textbox(923 label="Log",924 lines=15,925 interactive=False,926 show_copy_button=True927 )928 929 with gr.Accordion("Full Results (JSON)", open=False):930 batch_json = gr.Textbox(931 label="JSON Output",932 lines=10,933 interactive=False,934 show_copy_button=True935 )936 937 gr.Markdown("#### Download Results")938 with gr.Row():939 csv_download = gr.File(label="CSV Results")940 json_download = gr.File(label="JSON Results")941 zip_download = gr.File(label="All Clips (ZIP)")942 943 # Update queue size when inputs change944 queue_inputs = [batch_videos, batch_domains, batch_durations, batch_no_prompt, batch_prompt1, batch_prompt2, batch_prompt3]945 for inp in queue_inputs:946 inp.change(947 fn=calculate_queue_size,948 inputs=queue_inputs,949 outputs=queue_info950 )951 952 # Connect batch processing953 batch_start_btn.click(954 fn=run_batch_tests,955 inputs=[956 batch_videos,957 batch_domains,958 batch_durations,959 batch_num_clips,960 batch_ref_image,961 batch_no_prompt,962 batch_prompt1,963 batch_prompt2,964 batch_prompt3,965 ],966 outputs=[967 batch_status,968 batch_results_table,969 batch_log,970 batch_json,971 csv_download,972 json_download,973 zip_download,974 ],975 show_progress="full"976 )977 978 batch_cancel_btn.click(979 fn=cancel_batch,980 inputs=[],981 outputs=[batch_status]982 )983 984 gr.Markdown("""985 ---986 **ShortSmith v2** | Powered by Qwen2-VL, InsightFace, and Librosa |987 [GitHub](https://github.com) | Built with Gradio988 """)989 990# Launch the app991if __name__ == "__main__":992 demo.queue()993 demo.launch(994 server_name="0.0.0.0",995 server_port=7860,996 show_error=True997 )998else:999 # For HuggingFace Spaces1000 demo.queue()1001 demo.launch()1002