File size: 22,792 Bytes
b67275f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12eb1d7
 
 
 
b67275f
12eb1d7
 
 
b67275f
12eb1d7
b67275f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
"""
Phase 3: Native Reference Image Generator
==========================================

This module implements Vertex AI Native Reference Images using the capability model.
It provides true subject locking for character and location consistency across scenes.

Key Features:
- Service Account Authentication (required for native references)
- Vertex AI imagen-3.0-capability-001 model
- Native reference images with REFERENCE_TYPE_SUBJECT
- Subject type inference (PERSON, ANIMAL, PRODUCT)
- Fallback to Phase 2 when service account unavailable

WHY CAPABILITY MODEL IS USED:
-----------------------------
The imagen-3.0-capability-001 model supports native reference images, which allows
passing actual image data as reference subjects. This provides much stronger identity
consistency compared to prompt-based descriptions (Phase 2).

WHY API KEY IS NOT ALLOWED HERE:
--------------------------------
Native reference image features require the Vertex AI SDK with service account
authentication. The REST API with API key (used in Phase 2) does not support
the referenceImages parameter required for subject locking.

Authentication Requirements:
- Service account with roles/aiplatform.user
- GOOGLE_APPLICATION_CREDENTIALS environment variable pointing to service account JSON
"""

import os
import base64
import uuid
from typing import Optional, Dict, List, Any, Tuple
from PIL import Image

# =============================================================================
# CONFIGURATION
# =============================================================================

# Service account credentials path
GOOGLE_APPLICATION_CREDENTIALS = os.getenv("GOOGLE_APPLICATION_CREDENTIALS")

# GCP Project configuration
GCP_PROJECT_ID = os.getenv("GCP_PROJECT_ID")
GCP_LOCATION = os.getenv("GCP_LOCATION", "us-central1")

# Model configuration
CAPABILITY_MODEL_ID = "imagen-3.0-capability-001"

# Output directory (inherited from main app)
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
OUTPUT_DIR_IMG = os.path.join(BASE_DIR, "outputs", "images")

# Ensure output directory exists
os.makedirs(OUTPUT_DIR_IMG, exist_ok=True)

# =============================================================================
# SERVICE ACCOUNT INITIALIZATION
# =============================================================================
# 
# Supports two authentication methods:
# 
# 1. FILE-BASED (Local development):
#    Set GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
#
# 2. SECRET-BASED (Hugging Face Spaces / Cloud deployment):
#    Set GCP_SERVICE_ACCOUNT_JSON=<entire JSON content as string>
#    The module will automatically write this to a temp file
#
# =============================================================================

PHASE3_AVAILABLE = False
_vertex_client = None
_imagen_model = None
_temp_credentials_file = None  # Track temp file for cleanup


def setup_credentials_from_secret() -> Optional[str]:
    """
    Set up credentials from HF Spaces secret (GCP_SERVICE_ACCOUNT_JSON).
    
    This allows storing the entire service account JSON as an environment
    variable/secret, which is required for Hugging Face Spaces deployment.
    
    Returns:
        Path to temporary credentials file, or None if not configured
    """
    global _temp_credentials_file
    import json
    import tempfile
    
    # Check for JSON content in environment variable
    sa_json = os.getenv("GCP_SERVICE_ACCOUNT_JSON")
    
    if not sa_json:
        return None
    
    try:
        # Validate JSON structure
        creds = json.loads(sa_json)
        
        required_fields = ["type", "project_id", "private_key", "client_email"]
        missing = [f for f in required_fields if f not in creds]
        
        if missing:
            print(f"⚠️  GCP_SERVICE_ACCOUNT_JSON missing fields: {', '.join(missing)}")
            return None
        
        if creds.get("type") != "service_account":
            print(f"⚠️  Invalid credential type in GCP_SERVICE_ACCOUNT_JSON")
            return None
        
        # Write to temporary file
        fd, temp_path = tempfile.mkstemp(suffix=".json", prefix="gcp_sa_")
        with os.fdopen(fd, 'w') as f:
            json.dump(creds, f)
        
        _temp_credentials_file = temp_path
        
        # Set environment variable for Google libraries
        os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = temp_path
        
        print(f"πŸ” Phase 3: Credentials loaded from HF secret (GCP_SERVICE_ACCOUNT_JSON)")
        print(f"   Service account: {creds.get('client_email')}")
        
        return temp_path
        
    except json.JSONDecodeError as e:
        print(f"⚠️  Invalid JSON in GCP_SERVICE_ACCOUNT_JSON: {e}")
        return None
    except Exception as e:
        print(f"⚠️  Error setting up credentials from secret: {e}")
        return None


def check_service_account_credentials() -> Tuple[bool, str]:
    """
    Validate that service account credentials are properly configured.
    
    Supports both:
    - File-based: GOOGLE_APPLICATION_CREDENTIALS=/path/to/file.json
    - Secret-based: GCP_SERVICE_ACCOUNT_JSON=<json content> (for HF Spaces)
    
    Returns:
        Tuple of (is_valid, message)
    """
    # First, try to set up from HF Spaces secret if available
    if os.getenv("GCP_SERVICE_ACCOUNT_JSON"):
        temp_path = setup_credentials_from_secret()
        if temp_path:
            return True, f"Credentials loaded from HF secret"
    
    # Then check file-based credentials
    creds_path = os.getenv("GOOGLE_APPLICATION_CREDENTIALS")
    
    if not creds_path:
        return False, "No credentials configured (set GOOGLE_APPLICATION_CREDENTIALS or GCP_SERVICE_ACCOUNT_JSON)"
    
    if not os.path.isfile(creds_path):
        return False, f"Service account file not found: {creds_path}"
    
    # Validate JSON structure
    try:
        import json
        with open(creds_path, 'r') as f:
            creds = json.load(f)
        
        required_fields = ["type", "project_id", "private_key", "client_email"]
        missing = [f for f in required_fields if f not in creds]
        
        if missing:
            return False, f"Service account JSON missing fields: {', '.join(missing)}"
        
        if creds.get("type") != "service_account":
            return False, f"Invalid credential type: {creds.get('type')} (expected 'service_account')"
        
        return True, f"Service account validated: {creds.get('client_email')}"
        
    except json.JSONDecodeError as e:
        return False, f"Invalid JSON in service account file: {e}"
    except Exception as e:
        return False, f"Error reading service account file: {e}"


def init_phase3_vertex_ai() -> bool:
    """
    Initialize Vertex AI with service account authentication for Phase 3.
    
    This initialization is separate from Phase 2 (API key) and uses the
    full Vertex AI SDK with service account credentials.
    
    Returns:
        True if initialization successful, False otherwise
    """
    global PHASE3_AVAILABLE, _vertex_client, _imagen_model
    
    # Check credentials first
    is_valid, message = check_service_account_credentials()
    if not is_valid:
        print(f"⚠️  Phase 3 (Native References): {message}")
        print("   β†’ Falling back to Phase 2 (prompt-based consistency)")
        return False
    
    print(f"πŸ” Phase 3: {message}")
    
    # Check project ID
    project_id = os.getenv("GCP_PROJECT_ID")
    if not project_id:
        print("⚠️  Phase 3: GCP_PROJECT_ID not set")
        print("   β†’ Falling back to Phase 2 (prompt-based consistency)")
        return False
    
    try:
        # Import Vertex AI SDK
        import vertexai
        from vertexai.preview.vision_models import ImageGenerationModel
        
        # Initialize Vertex AI with service account
        location = os.getenv("GCP_LOCATION", "us-central1")
        vertexai.init(project=project_id, location=location)
        
        # Load the capability model
        print(f"πŸ”„ Loading Vertex AI model: {CAPABILITY_MODEL_ID}...")
        _imagen_model = ImageGenerationModel.from_pretrained(CAPABILITY_MODEL_ID)
        
        # Verify model loaded successfully
        if _imagen_model is None:
            raise RuntimeError("Model failed to load")
        
        PHASE3_AVAILABLE = True
        print(f"βœ… Phase 3 (Native References) initialized successfully")
        print(f"   Model: {CAPABILITY_MODEL_ID}")
        print(f"   Project: {project_id}")
        print(f"   Location: {location}")
        return True
        
    except ImportError as e:
        print(f"⚠️  Phase 3: Vertex AI SDK not installed - {e}")
        print("   Install with: pip install google-cloud-aiplatform vertexai")
        print("   β†’ Falling back to Phase 2 (prompt-based consistency)")
        return False
        
    except Exception as e:
        print(f"⚠️  Phase 3: Initialization failed - {e}")
        print("   β†’ Falling back to Phase 2 (prompt-based consistency)")
        return False


def is_phase3_available() -> bool:
    """Check if Phase 3 native reference generation is available."""
    return PHASE3_AVAILABLE and _imagen_model is not None


# =============================================================================
# SUBJECT TYPE INFERENCE
# =============================================================================

def infer_subject_type(ref_key: str, image_path: Optional[str] = None) -> str:
    """
    Infer the subject type from the reference key or image content.
    
    Subject types supported by Vertex AI:
    - PERSON: Human characters
    - ANIMAL: Animals, creatures
    - PRODUCT: Objects, items, products
    
    Args:
        ref_key: Reference key name (e.g., "hero", "cat", "sword")
        image_path: Optional path to image for content-based inference
    
    Returns:
        One of: "PERSON", "ANIMAL", "PRODUCT"
    """
    ref_lower = ref_key.lower() if ref_key else ""
    
    # Person indicators
    person_keywords = [
        "person", "human", "man", "woman", "boy", "girl", "child", "kid",
        "hero", "villain", "character", "protagonist", "antagonist",
        "father", "mother", "daughter", "son", "parent", "adult",
        "doctor", "nurse", "teacher", "student", "scientist", "professor",
        "warrior", "knight", "mage", "wizard", "witch",
        "narrator", "speaker", "presenter",
        "pediatrician", "medical"
    ]
    
    # Animal indicators
    animal_keywords = [
        "animal", "creature", "beast", "pet",
        "cat", "dog", "bird", "fish", "rabbit", "hamster",
        "lion", "tiger", "bear", "wolf", "fox", "deer",
        "dragon", "unicorn", "phoenix", "monster",
        "horse", "cow", "pig", "sheep", "goat",
        "elephant", "giraffe", "zebra", "monkey",
        "snake", "lizard", "frog", "turtle"
    ]
    
    # Product indicators (check first since some might conflict with defaults)
    product_keywords = [
        "object", "item", "product", "thing", "prop",
        "sword", "weapon", "shield", "armor", "helmet",
        "car", "vehicle", "ship", "boat", "plane", "aircraft",
        "building", "house", "tower", "castle",
        "book", "scroll", "map", "letter",
        "potion", "bottle", "cup", "glass", "bowl",
        "ring", "necklace", "crown", "jewel", "gem",
        "wand", "staff", "orb", "crystal",
        "food", "fruit", "vegetable", "bread", "cake",
        "furniture", "chair", "table", "bed", "desk",
        "tool", "hammer", "wrench", "screwdriver",
        "machine", "robot", "device", "gadget", "computer"
    ]
    
    # Check for product keywords first
    for keyword in product_keywords:
        if keyword in ref_lower:
            return "PRODUCT"
    
    # Check for person keywords
    for keyword in person_keywords:
        if keyword in ref_lower:
            return "PERSON"
    
    # Check for animal keywords
    for keyword in animal_keywords:
        if keyword in ref_lower:
            return "ANIMAL"
    
    # Default to PERSON for character references
    # Most character references in animation are people
    character_indicators = ["char", "actor", "role", "cast"]
    for indicator in character_indicators:
        if indicator in ref_lower:
            return "PERSON"
    
    # Default to PERSON for unrecognized references
    # (most common case in animation production)
    return "PERSON"


# =============================================================================
# REFERENCE IMAGE PREPARATION
# =============================================================================

def prepare_reference_image(image_path: str) -> Optional[bytes]:
    """
    Prepare a reference image for the Vertex AI API.
    
    Ensures image meets API requirements:
    - Supported formats (PNG, JPEG)
    - Reasonable file size
    
    Args:
        image_path: Path to the reference image
    
    Returns:
        Image bytes, or None on failure
    """
    if not image_path or not os.path.isfile(image_path):
        return None
    
    try:
        # Read and validate image
        with Image.open(image_path) as img:
            # Convert to RGB if necessary (remove alpha channel)
            if img.mode in ('RGBA', 'LA', 'P'):
                background = Image.new('RGB', img.size, (255, 255, 255))
                if img.mode == 'P':
                    img = img.convert('RGBA')
                background.paste(img, mask=img.split()[-1] if img.mode == 'RGBA' else None)
                img = background
            elif img.mode != 'RGB':
                img = img.convert('RGB')
            
            # Resize if too large (max 4096px on longest side)
            max_size = 4096
            if max(img.size) > max_size:
                ratio = max_size / max(img.size)
                new_size = (int(img.size[0] * ratio), int(img.size[1] * ratio))
                img = img.resize(new_size, Image.Resampling.LANCZOS)
            
            # Save to bytes
            import io
            buffer = io.BytesIO()
            img.save(buffer, format='PNG')
            return buffer.getvalue()
            
    except Exception as e:
        print(f"⚠️  Error preparing reference image {image_path}: {e}")
        return None


def load_reference_image_as_vertex_image(image_path: str):
    """
    Load a reference image as a Vertex AI Image object.
    
    Args:
        image_path: Path to the reference image
    
    Returns:
        Vertex AI Image object, or None on failure
    """
    if not image_path or not os.path.isfile(image_path):
        return None
    
    try:
        from vertexai.preview.vision_models import Image as VertexImage
        
        # Load image using Vertex AI's Image class
        return VertexImage.load_from_file(image_path)
        
    except Exception as e:
        print(f"⚠️  Error loading reference image as Vertex Image: {e}")
        return None


# =============================================================================
# NATIVE REFERENCE IMAGE GENERATION
# =============================================================================

def generate_with_native_references(
    prompt: str,
    scene_id: str,
    variant: str,
    ref_images: Dict[str, Any] = None,
    aspect_ratio: str = "9:16"
) -> Optional[str]:
    """
    Generate an image using Vertex AI native reference images.
    
    This function uses the imagen-3.0-capability-001 model with true subject
    locking via the referenceImages parameter.
    
    Args:
        prompt: The text prompt for image generation
        scene_id: Scene identifier for naming
        variant: "A" or "B" for the variant
        ref_images: Dict with:
            - 'character': Single character path (backward compat)
            - 'characters': List of (key, path) tuples
            - 'location': Location path
        aspect_ratio: Aspect ratio (default "9:16" for vertical)
    
    Returns:
        Path to generated image, or None on failure
    """
    global _imagen_model
    
    if not is_phase3_available():
        print("⚠️  Phase 3 not available, cannot generate with native references")
        return None
    
    try:
        from vertexai.preview.vision_models import Image as VertexImage
        
        print(f"🎨 Phase 3: Generating with native references for {scene_id}_{variant}...")
        
        # Build reference images list
        reference_images = []
        reference_id = 1
        
        # Process character references
        if ref_images:
            characters_list = ref_images.get("characters", [])
            
            # Add single character if no list provided
            if not characters_list and ref_images.get("character"):
                char_path = ref_images.get("character")
                if char_path and os.path.isfile(char_path):
                    characters_list = [("character", char_path)]
            
            # Process each character reference
            for key, path in characters_list:
                if path and os.path.isfile(path):
                    try:
                        ref_image = VertexImage.load_from_file(path)
                        subject_type = infer_subject_type(key, path)
                        
                        reference_images.append({
                            "referenceImage": ref_image,
                            "referenceType": "REFERENCE_TYPE_SUBJECT",
                            "referenceId": reference_id,
                            "subjectType": subject_type
                        })
                        
                        print(f"   πŸ“Ž Reference {reference_id}: {key} ({subject_type}) β†’ {os.path.basename(path)}")
                        reference_id += 1
                        
                    except Exception as e:
                        print(f"   ⚠️  Failed to load reference {key}: {e}")
            
            # Process location reference (as style/background reference)
            location_path = ref_images.get("location")
            if location_path and os.path.isfile(location_path):
                try:
                    loc_image = VertexImage.load_from_file(location_path)
                    
                    # Locations are treated as style references, not subjects
                    reference_images.append({
                        "referenceImage": loc_image,
                        "referenceType": "REFERENCE_TYPE_STYLE",
                        "referenceId": reference_id
                    })
                    
                    print(f"   πŸ“Ž Reference {reference_id}: location (STYLE) β†’ {os.path.basename(location_path)}")
                    reference_id += 1
                    
                except Exception as e:
                    print(f"   ⚠️  Failed to load location reference: {e}")
        
        # Build generation config
        generation_kwargs = {
            "prompt": f"Kurzgesagt style illustration, 2D flat vector animation, vibrant colors, minimalist design, {prompt}",
            "number_of_images": 1,
            "aspect_ratio": aspect_ratio,
        }
        
        # NOTE: The Vertex AI Python SDK's generate_images() does NOT support 
        # reference_images parameter directly. Native reference images require
        # using the REST API with the imagen-3.0-capability-001 model.
        # For now, we skip Phase 3 and use Phase 2 fallback.
        if reference_images:
            print(f"   ⚠️  SDK limitation: reference_images not supported in generate_images()")
            print(f"   β†’ Falling back to Phase 2 (prompt-based consistency)")
            return None
        
        # Generate image without references
        response = _imagen_model.generate_images(**generation_kwargs)
        
        if response.images and len(response.images) > 0:
            # Save generated image
            output_path = os.path.join(
                OUTPUT_DIR_IMG, 
                f"{scene_id}_{variant}_{uuid.uuid4().hex[:8]}.png"
            )
            
            response.images[0].save(output_path)
            print(f"βœ… Phase 3: Image generated: {output_path}")
            return output_path
        else:
            print(f"⚠️  Phase 3: No images in response")
            return None
            
    except Exception as e:
        print(f"❌ Phase 3 generation failed: {e}")
        import traceback
        traceback.print_exc()
        return None


# =============================================================================
# FALLBACK CHECK & HYBRID GENERATION
# =============================================================================

def generate_image_with_fallback(
    prompt: str,
    scene_id: str,
    variant: str,
    ref_images: Dict[str, Any] = None,
    phase2_fallback_fn = None,
    aspect_ratio: str = "9:16"
) -> Optional[str]:
    """
    Generate an image using Phase 3 (native references) if available,
    otherwise fall back to Phase 2 (prompt-based).
    
    This is the main entry point for hybrid Phase 2/3 generation.
    
    Args:
        prompt: The text prompt for image generation
        scene_id: Scene identifier for naming
        variant: "A" or "B" for the variant
        ref_images: Reference images dictionary
        phase2_fallback_fn: Fallback function for Phase 2 generation
        aspect_ratio: Aspect ratio for generation
    
    Returns:
        Path to generated image, or None on failure
    """
    # Try Phase 3 first if available
    if is_phase3_available():
        print(f"πŸ”· Using Phase 3 (Native Reference Images) for {scene_id}_{variant}")
        result = generate_with_native_references(
            prompt=prompt,
            scene_id=scene_id,
            variant=variant,
            ref_images=ref_images,
            aspect_ratio=aspect_ratio
        )
        
        if result:
            return result
        else:
            print(f"⚠️  Phase 3 generation failed, falling back to Phase 2")
    
    # Fall back to Phase 2
    if phase2_fallback_fn:
        print(f"πŸ”Ά Using Phase 2 (Prompt-based) for {scene_id}_{variant}")
        return phase2_fallback_fn(prompt, scene_id, variant, ref_images)
    
    return None


# =============================================================================
# INITIALIZATION ON MODULE LOAD
# =============================================================================

# Attempt to initialize Phase 3 on module import
# This is safe - if it fails, Phase 2 will be used as fallback
print("\n" + "=" * 60)
print("πŸ”„ Initializing Phase 3 (Native Reference Images)...")
print("=" * 60)
init_phase3_vertex_ai()
print("=" * 60 + "\n")