Text Generation
video-editing
social-media
agent
tool-calling
sft
trl
viralcut
File size: 21,777 Bytes
0307f85
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
"""
ViralCut Agent - Runtime
========================
The actual agent that uses the fine-tuned model to edit videos autonomously.

This connects the trained model to real tools:
- FFmpeg for video editing
- DuckDuckGo for web search (free, no API key)
- Whisper for transcription
- PySceneDetect for shot detection

Usage:
    python agent.py --video raw_footage.mp4 --platform tiktok --niche food
    python agent.py --plan --niche "coffee shop" --platform tiktok
"""

import argparse
import json
import os
import re
import subprocess
import sys
import tempfile
from pathlib import Path

# ============================================================
# TOOL IMPLEMENTATIONS
# ============================================================

class FFmpegTool:
    """Execute FFmpeg commands for video/audio processing."""
    
    @staticmethod
    def run(command: str, description: str = "") -> str:
        """Execute an FFmpeg command and return result."""
        print(f"  🎬 FFmpeg: {description}")
        print(f"     $ {command}")
        try:
            result = subprocess.run(
                command, shell=True, capture_output=True, text=True, timeout=120
            )
            if result.returncode == 0:
                return json.dumps({"status": "success", "message": f"Command completed: {description}"})
            else:
                return json.dumps({"status": "error", "message": result.stderr[:500]})
        except subprocess.TimeoutExpired:
            return json.dumps({"status": "error", "message": "Command timed out after 120s"})
        except Exception as e:
            return json.dumps({"status": "error", "message": str(e)})


class WebSearchTool:
    """Search the web using DuckDuckGo (free, no API key needed)."""
    
    @staticmethod
    def search(query: str, search_type: str = "general") -> str:
        """Search the web and return results."""
        print(f"  πŸ” Searching: {query} (type: {search_type})")
        try:
            from duckduckgo_search import DDGS
            with DDGS() as ddgs:
                results = []
                for r in ddgs.text(query, max_results=5):
                    results.append({
                        "title": r.get("title", ""),
                        "url": r.get("href", ""),
                        "description": r.get("body", "")[:200]
                    })
                return json.dumps({"results": results})
        except ImportError:
            return json.dumps({"results": [{"title": "Install duckduckgo-search", "description": "pip install duckduckgo-search"}]})
        except Exception as e:
            return json.dumps({"results": [], "error": str(e)})


class VideoAnalyzer:
    """Analyze video files using ffprobe and PySceneDetect."""
    
    @staticmethod
    def analyze(video_path: str, analysis_type: str = "full") -> str:
        """Analyze a video file."""
        print(f"  πŸ“Š Analyzing: {video_path} ({analysis_type})")
        
        if not os.path.exists(video_path):
            return json.dumps({"error": f"File not found: {video_path}"})
        
        result = {}
        
        # Get basic info via ffprobe
        try:
            probe = subprocess.run(
                f'ffprobe -v quiet -print_format json -show_format -show_streams "{video_path}"',
                shell=True, capture_output=True, text=True
            )
            if probe.returncode == 0:
                info = json.loads(probe.stdout)
                fmt = info.get("format", {})
                result["duration"] = float(fmt.get("duration", 0))
                result["size_mb"] = round(int(fmt.get("size", 0)) / 1024 / 1024, 1)
                
                for stream in info.get("streams", []):
                    if stream.get("codec_type") == "video":
                        result["resolution"] = f"{stream.get('width')}x{stream.get('height')}"
                        result["fps"] = eval(stream.get("r_frame_rate", "30/1"))
                        result["codec"] = stream.get("codec_name")
                    elif stream.get("codec_type") == "audio":
                        result["audio_codec"] = stream.get("codec_name")
                        result["audio_channels"] = stream.get("channels")
        except Exception as e:
            result["probe_error"] = str(e)
        
        # Scene detection
        if analysis_type in ("full", "scenes"):
            try:
                from scenedetect import open_video, SceneManager
                from scenedetect.detectors import ContentDetector
                
                video = open_video(video_path)
                scene_manager = SceneManager()
                scene_manager.add_detector(ContentDetector(threshold=27))
                scene_manager.detect_scenes(video)
                scene_list = scene_manager.get_scene_list()
                
                result["scenes"] = []
                for i, (start, end) in enumerate(scene_list):
                    result["scenes"].append({
                        "scene": i + 1,
                        "start": round(start.get_seconds(), 2),
                        "end": round(end.get_seconds(), 2),
                        "duration": round((end - start).get_seconds(), 2)
                    })
            except ImportError:
                result["scenes_note"] = "Install scenedetect: pip install scenedetect[opencv]"
            except Exception as e:
                result["scenes_error"] = str(e)
        
        # Transcript via Whisper
        if analysis_type in ("full", "transcript", "audio"):
            try:
                import whisper
                model = whisper.load_model("base")
                transcript = model.transcribe(video_path)
                result["transcript"] = transcript.get("text", "")[:2000]
                result["segments"] = [
                    {"start": s["start"], "end": s["end"], "text": s["text"]}
                    for s in transcript.get("segments", [])[:50]
                ]
            except ImportError:
                result["transcript_note"] = "Install whisper: pip install openai-whisper"
            except Exception as e:
                result["transcript_error"] = str(e)
        
        return json.dumps(result)


class ViralityScorer:
    """Score video content for viral potential."""
    
    @staticmethod
    def score(video_path: str, platform: str, niche: str = "") -> str:
        """Score a video's viral potential based on heuristics."""
        print(f"  πŸ“ˆ Scoring virality: {video_path} for {platform}")
        
        # Get video info
        try:
            probe = subprocess.run(
                f'ffprobe -v quiet -print_format json -show_format -show_streams "{video_path}"',
                shell=True, capture_output=True, text=True
            )
            info = json.loads(probe.stdout) if probe.returncode == 0 else {}
        except:
            info = {}
        
        duration = float(info.get("format", {}).get("duration", 0))
        has_audio = any(s.get("codec_type") == "audio" for s in info.get("streams", []))
        
        # Platform-specific optimal durations
        optimal_ranges = {
            "tiktok": (7, 30),
            "instagram_reels": (15, 30),
            "youtube_shorts": (30, 60)
        }
        opt_min, opt_max = optimal_ranges.get(platform, (15, 60))
        
        # Score components
        scores = {}
        
        # Length score
        if opt_min <= duration <= opt_max:
            scores["length_optimal"] = 90
        elif duration < opt_min:
            scores["length_optimal"] = max(50, 90 - (opt_min - duration) * 5)
        else:
            scores["length_optimal"] = max(40, 90 - (duration - opt_max) * 3)
        
        # Audio presence
        scores["audio_match"] = 80 if has_audio else 30
        
        # Resolution check
        for s in info.get("streams", []):
            if s.get("codec_type") == "video":
                h = int(s.get("height", 0))
                w = int(s.get("width", 0))
                if h >= 1920 or w >= 1080:
                    scores["visual_quality"] = 85
                elif h >= 1080:
                    scores["visual_quality"] = 75
                else:
                    scores["visual_quality"] = 55
                # Vertical check
                if h > w:
                    scores["format_match"] = 90
                else:
                    scores["format_match"] = 50
        
        scores.setdefault("visual_quality", 60)
        scores.setdefault("format_match", 60)
        scores["hook_strength"] = 70  # Can't assess without content analysis
        scores["pacing"] = 70
        scores["trend_alignment"] = 65
        
        overall = round(sum(scores.values()) / len(scores))
        
        suggestions = []
        if scores.get("format_match", 0) < 70:
            suggestions.append("Convert to 9:16 vertical format for better reach")
        if scores.get("length_optimal", 0) < 70:
            suggestions.append(f"Adjust length to {opt_min}-{opt_max}s for {platform}")
        if not has_audio:
            suggestions.append("Add audio - videos without sound get 40% less reach")
        
        return json.dumps({
            "overall_score": overall,
            "breakdown": scores,
            "suggestions": suggestions
        })


class CaptionGenerator:
    """Generate platform-optimized captions."""
    
    @staticmethod
    def generate(video_description: str, platform: str, tone: str = "casual", include_cta: bool = True) -> str:
        """Generate a caption (using the model itself for this in production)."""
        print(f"  ✍️ Generating caption for {platform}")
        
        hashtag_sets = {
            "tiktok": ["#fyp", "#viral", "#foryou", "#trending"],
            "instagram": ["#reels", "#explore", "#instagood", "#trending"],
            "youtube": ["#shorts", "#subscribe", "#viral"]
        }
        
        base_tags = hashtag_sets.get(platform, ["#viral"])
        
        # Extract keywords from description for niche hashtags
        words = video_description.lower().split()
        niche_tags = [f"#{w}" for w in words if len(w) > 3 and w.isalpha()][:3]
        
        posting_times = {
            "tiktok": "7-9am, 12-1pm, or 7-9pm in your audience timezone",
            "instagram": "6-9am, 12-2pm, or 5-7pm EST",
            "youtube": "2-4pm or 8-10pm EST"
        }
        
        return json.dumps({
            "caption": f"[AI will generate based on: {video_description}]",
            "hashtags": " ".join(base_tags + niche_tags),
            "posting_time": posting_times.get(platform, "Check your analytics"),
            "tip": "Reply to every comment in the first hour - algorithm loves engagement"
        })


class AIDetector:
    """Detect AI-generated content."""
    
    @staticmethod
    def detect(content_path: str, check_type: str = "video") -> str:
        """Basic AI content detection heuristics."""
        print(f"  πŸ”¬ Checking for AI artifacts: {content_path}")
        
        if not os.path.exists(content_path):
            return json.dumps({"error": f"File not found: {content_path}"})
        
        # Basic file analysis (real detection would use a classifier model)
        size = os.path.getsize(content_path)
        
        return json.dumps({
            "file_analyzed": content_path,
            "check_type": check_type,
            "file_size_mb": round(size / 1024 / 1024, 2),
            "note": "Full AI detection requires DeMamba or VideoScore2 model. Basic file analysis only.",
            "recommendations": [
                "Check for morphing objects between frames",
                "Look for impossible reflections or shadows",
                "Verify text is readable and consistent",
                "Check if camera movement is unnaturally smooth"
            ]
        })


# ============================================================
# AGENT CORE
# ============================================================

TOOL_MAP = {
    "ffmpeg_cmd": lambda args: FFmpegTool.run(**args),
    "web_search": lambda args: WebSearchTool.search(**args),
    "analyze_video": lambda args: VideoAnalyzer.analyze(**args),
    "score_virality": lambda args: ViralityScorer.score(**args),
    "generate_caption": lambda args: CaptionGenerator.generate(**args),
    "detect_ai_slop": lambda args: AIDetector.detect(**args),
}


class ViralCutAgent:
    """The main agent that orchestrates video editing using the fine-tuned model."""
    
    def __init__(self, model_id="ryu34/viralcut-agent", device="auto"):
        print(f"Loading ViralCut Agent from {model_id}...")
        
        from transformers import AutoModelForCausalLM, AutoTokenizer
        
        self.tokenizer = AutoTokenizer.from_pretrained(model_id)
        self.model = AutoModelForCausalLM.from_pretrained(
            model_id,
            device_map=device,
            torch_dtype="auto",
        )
        self.model.eval()
        
        # Tool definitions for the chat template
        self.tools = [
            {"type": "function", "function": {"name": "ffmpeg_cmd", "description": "Execute FFmpeg command for video/audio processing.", "parameters": {"type": "object", "properties": {"command": {"type": "string"}, "description": {"type": "string"}}, "required": ["command", "description"]}}},
            {"type": "function", "function": {"name": "web_search", "description": "Search web for royalty-free assets and trends.", "parameters": {"type": "object", "properties": {"query": {"type": "string"}, "search_type": {"type": "string", "enum": ["royalty_free_music", "sound_effects", "trending_content", "general"]}}, "required": ["query", "search_type"]}}},
            {"type": "function", "function": {"name": "analyze_video", "description": "Analyze video for scenes, audio, transcript, quality.", "parameters": {"type": "object", "properties": {"video_path": {"type": "string"}, "analysis_type": {"type": "string", "enum": ["full", "scenes", "audio", "transcript", "quality", "pacing"]}}, "required": ["video_path", "analysis_type"]}}},
            {"type": "function", "function": {"name": "score_virality", "description": "Score video viral potential 0-100.", "parameters": {"type": "object", "properties": {"video_path": {"type": "string"}, "platform": {"type": "string", "enum": ["tiktok", "instagram_reels", "youtube_shorts"]}, "niche": {"type": "string"}}, "required": ["video_path", "platform"]}}},
            {"type": "function", "function": {"name": "generate_caption", "description": "Generate platform-optimized caption with hashtags.", "parameters": {"type": "object", "properties": {"video_description": {"type": "string"}, "platform": {"type": "string", "enum": ["tiktok", "instagram", "youtube"]}, "tone": {"type": "string"}, "include_cta": {"type": "boolean"}}, "required": ["video_description", "platform"]}}},
            {"type": "function", "function": {"name": "detect_ai_slop", "description": "Check content for AI-generated artifacts.", "parameters": {"type": "object", "properties": {"content_path": {"type": "string"}, "check_type": {"type": "string", "enum": ["video", "image", "text", "audio"]}}, "required": ["content_path", "check_type"]}}}
        ]
        
        print("Agent ready!")
    
    def run(self, user_message: str, max_turns: int = 15):
        """Run the agent on a user request, executing tool calls autonomously."""
        
        messages = [
            {"role": "system", "content": "You are ViralCut Agent, an autonomous AI video editor and social media content strategist. You transform raw video footage into professional, viral-worthy social media content. Use your tools to analyze, edit, search, and optimize. Think step-by-step. Always use royalty-free content."},
            {"role": "user", "content": user_message}
        ]
        
        print(f"\n{'='*60}")
        print(f"🎬 ViralCut Agent")
        print(f"{'='*60}")
        print(f"User: {user_message}\n")
        
        for turn in range(max_turns):
            # Generate response
            text = self.tokenizer.apply_chat_template(
                messages, tools=self.tools, tokenize=False, add_generation_prompt=True
            )
            inputs = self.tokenizer(text, return_tensors="pt").to(self.model.device)
            
            with __import__("torch").no_grad():
                outputs = self.model.generate(
                    **inputs,
                    max_new_tokens=1024,
                    temperature=0.7,
                    top_p=0.9,
                    do_sample=True,
                )
            
            response = self.tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=False)
            
            # Parse response for tool calls or plain text
            tool_calls = self._parse_tool_calls(response)
            
            if tool_calls:
                # Add assistant message with tool calls
                messages.append({"role": "assistant", "tool_calls": tool_calls})
                
                # Execute each tool call
                for tc in tool_calls:
                    func_name = tc["function"]["name"]
                    try:
                        args = json.loads(tc["function"]["arguments"])
                    except:
                        args = {}
                    
                    print(f"\n  πŸ”§ Calling: {func_name}")
                    
                    if func_name in TOOL_MAP:
                        result = TOOL_MAP[func_name](args)
                    else:
                        result = json.dumps({"error": f"Unknown tool: {func_name}"})
                    
                    messages.append({"role": "tool", "name": func_name, "content": result})
                    print(f"  βœ… Result: {result[:200]}...")
            else:
                # Plain text response - agent is done
                clean = self._clean_response(response)
                messages.append({"role": "assistant", "content": clean})
                print(f"\nπŸ€– Agent: {clean}")
                break
        
        return messages
    
    def _parse_tool_calls(self, response: str) -> list:
        """Parse tool calls from model output."""
        tool_calls = []
        
        # Qwen tool call format: <tool_call>{"name": "...", "arguments": {...}}</tool_call>
        pattern = r'<tool_call>\s*(\{.*?\})\s*</tool_call>'
        matches = re.findall(pattern, response, re.DOTALL)
        
        for match in matches:
            try:
                data = json.loads(match)
                tool_calls.append({
                    "type": "function",
                    "function": {
                        "name": data.get("name", ""),
                        "arguments": json.dumps(data.get("arguments", {}))
                    }
                })
            except json.JSONDecodeError:
                continue
        
        return tool_calls
    
    def _clean_response(self, response: str) -> str:
        """Clean up model response."""
        # Remove special tokens
        for token in ["<|endoftext|>", "<|im_end|>", "<|im_start|>"]:
            response = response.replace(token, "")
        return response.strip()


# ============================================================
# CLI
# ============================================================

def main():
    parser = argparse.ArgumentParser(description="ViralCut Agent - AI Video Editor")
    parser.add_argument("--video", type=str, help="Path to raw video file")
    parser.add_argument("--platform", type=str, default="tiktok", 
                       choices=["tiktok", "instagram", "youtube"],
                       help="Target platform")
    parser.add_argument("--niche", type=str, default="", help="Content niche")
    parser.add_argument("--plan", action="store_true", help="Generate content plan only (no video needed)")
    parser.add_argument("--model", type=str, default="ryu34/viralcut-agent", help="Model ID")
    parser.add_argument("--check-slop", type=str, nargs="+", help="Check files for AI-generated content")
    
    args = parser.parse_args()
    
    if args.check_slop:
        # Quick AI slop check without loading the full model
        for f in args.check_slop:
            result = AIDetector.detect(f, "video")
            print(json.dumps(json.loads(result), indent=2))
        return
    
    agent = ViralCutAgent(model_id=args.model)
    
    if args.plan:
        niche = args.niche or "general"
        agent.run(f"Research current {args.platform} trends for the '{niche}' niche and create a detailed 7-day content plan with hooks, posting times, and viral strategies.")
    elif args.video:
        if not os.path.exists(args.video):
            print(f"Error: Video file not found: {args.video}")
            sys.exit(1)
        niche_str = f" in the {args.niche} niche" if args.niche else ""
        agent.run(f"I have raw footage at {args.video}. Transform it into a professional, viral {args.platform} video{niche_str}. Analyze it, find the best moments, add trending music, professional edits, and optimize for maximum engagement.")
    else:
        # Interactive mode
        print("ViralCut Agent - Interactive Mode")
        print("Type your request (or 'quit' to exit):\n")
        while True:
            try:
                user_input = input("You: ").strip()
                if user_input.lower() in ("quit", "exit", "q"):
                    break
                if user_input:
                    agent.run(user_input)
            except (KeyboardInterrupt, EOFError):
                break


if __name__ == "__main__":
    main()