Spaces:
Running
on
Zero
Running
on
Zero
| """ | |
| Configuration file for Phramer AI | |
| By Pariente AI, for MIA TV Series | |
| Multimodal tool with BAGEL integration and professional photographic prompt optimization | |
| """ | |
| import os | |
| import torch | |
| from typing import Dict, Any | |
| # Application Configuration | |
| APP_CONFIG = { | |
| "title": "Phramer AI", | |
| "description": "Multimodal tool that reads images and turns them into refined, photo-realistic prompts. Ready for Midjourney, Flux or any generative engine.", | |
| "version": "2.0.0", | |
| "author": "Pariente AI", | |
| "project": "MIA TV Series", | |
| "tagline": "By Pariente AI, for MIA TV Series", | |
| "logline": "Phramer AI is a multimodal tool that reads an image and turns it into a refined, photo-realistic prompt. Ready for Midjourney, Flux or any generative engine." | |
| } | |
| # BAGEL Model Configuration | |
| BAGEL_CONFIG = { | |
| "model_repo": "ByteDance-Seed/BAGEL-7B-MoT", | |
| "local_model_path": "./model", | |
| "cache_dir": "./model/cache", | |
| "download_patterns": ["*.json", "*.safetensors", "*.bin", "*.py", "*.md", "*.txt"], | |
| # Model parameters | |
| "dtype": torch.bfloat16, | |
| "device_map_strategy": "auto", | |
| "max_memory_per_gpu": "80GiB", | |
| "offload_buffers": True, | |
| "force_hooks": True, | |
| # Image processing | |
| "vae_transform_size": (1024, 512, 16), | |
| "vit_transform_size": (980, 224, 14), | |
| # Inference parameters | |
| "max_new_tokens": 512, | |
| "temperature": 0.7, | |
| "top_p": 0.9, | |
| "do_sample": True | |
| } | |
| # Device Configuration for ZeroGPU | |
| def get_device_config() -> Dict[str, Any]: | |
| """Determine optimal device configuration for BAGEL""" | |
| device_config = { | |
| "device": "cpu", | |
| "use_gpu": False, | |
| "gpu_count": 0, | |
| "memory_efficient": True | |
| } | |
| if torch.cuda.is_available(): | |
| gpu_count = torch.cuda.device_count() | |
| device_config.update({ | |
| "device": "cuda", | |
| "use_gpu": True, | |
| "gpu_count": gpu_count, | |
| "gpu_memory_gb": torch.cuda.get_device_properties(0).total_memory / 1e9, | |
| "multi_gpu": gpu_count > 1 | |
| }) | |
| elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available(): | |
| device_config.update({ | |
| "device": "mps", | |
| "use_gpu": True, | |
| "gpu_count": 1 | |
| }) | |
| return device_config | |
| # BAGEL Device Mapping Configuration | |
| def get_bagel_device_map(gpu_count: int) -> Dict[str, str]: | |
| """Configure device mapping for BAGEL model""" | |
| # Same device modules that need to be on the same GPU | |
| same_device_modules = [ | |
| 'language_model.model.embed_tokens', | |
| 'time_embedder', | |
| 'latent_pos_embed', | |
| 'vae2llm', | |
| 'llm2vae', | |
| 'connector', | |
| 'vit_pos_embed' | |
| ] | |
| device_map = {} | |
| if gpu_count == 1: | |
| # Single GPU configuration | |
| for module in same_device_modules: | |
| device_map[module] = "cuda:0" | |
| else: | |
| # Multi-GPU configuration - keep critical modules on same device | |
| first_device = "cuda:0" | |
| for module in same_device_modules: | |
| device_map[module] = first_device | |
| return device_map | |
| # Processing Configuration | |
| PROCESSING_CONFIG = { | |
| "max_image_size": 1024, | |
| "image_quality": 95, | |
| "supported_formats": [".jpg", ".jpeg", ".png", ".webp"], | |
| "batch_size": 1, | |
| "timeout_seconds": 120 # Increased for BAGEL processing | |
| } | |
| # Prompt Optimization Rules for Multi-Engine Compatibility | |
| FLUX_RULES = { | |
| "remove_patterns": [ | |
| r',\s*trending on artstation', | |
| r',\s*trending on [^,]+', | |
| r',\s*\d+k\s*', | |
| r',\s*\d+k resolution', | |
| r',\s*artstation', | |
| r',\s*concept art', | |
| r',\s*digital art', | |
| r',\s*by greg rutkowski', | |
| ], | |
| "camera_configs": { | |
| "portrait": ", Shot on Hasselblad X2D 100C, 85mm f/1.4 lens at f/2.8, ISO 200, rule of thirds composition, professional portrait photography", | |
| "landscape": ", Shot on Phase One XT, 24-70mm f/4 lens at f/8, ISO 100, hyperfocal distance, leading lines composition, epic landscape photography", | |
| "street": ", Shot on Leica M11, 35mm f/1.4 lens at f/2.8, ISO 800, decisive moment, candid composition, documentary street photography", | |
| "cinematic": ", Shot on ARRI Alexa LF, 35mm anamorphic lens at f/2.8, ISO 400, cinematic framing, film grain, dramatic composition", | |
| "architectural": ", Shot on Canon EOS R5, 24-70mm f/2.8 lens at f/8, ISO 100, symmetrical composition, perspective correction, architectural photography", | |
| "commercial": ", Shot on Hasselblad X2D 100C, 90mm f/2.5 lens at f/4, ISO 100, centered composition, product photography", | |
| "default": ", Shot on Phase One XF IQ4, 80mm f/2.8 lens at f/4, ISO 200, balanced composition, professional photography" | |
| }, | |
| "lighting_enhancements": { | |
| "dramatic": ", dramatic key lighting with rim light, chiaroscuro lighting ratio 4:1", | |
| "portrait": ", soft key light with fill light, 3-point lighting setup, lighting ratio 2:1", | |
| "cinematic": ", moody cinematic lighting with practical lights, dramatic shadows, film noir lighting", | |
| "natural": ", natural window lighting, soft diffused light, even lighting ratio", | |
| "default": ", professional lighting setup, balanced exposure" | |
| }, | |
| "style_enhancements": { | |
| "photorealistic": ", photorealistic rendering, ultra-detailed, sharp focus, high resolution", | |
| "cinematic": ", cinematic composition, film grain, anamorphic bokeh, color grading", | |
| "commercial": ", commercial photography style, clean composition, product focus", | |
| "editorial": ", editorial photography style, storytelling composition, magazine quality", | |
| "fine_art": ", fine art photography, artistic composition, gallery quality" | |
| }, | |
| # Text condensation patterns | |
| "condensation_patterns": { | |
| "remove_redundant_phrases": [ | |
| r'This image depicts', | |
| r'The image shows', | |
| r'appears to be', | |
| r'seems to be', | |
| r'which adds', | |
| r'The setting appears to be', | |
| r'The scene is', | |
| r'The composition is carefully', | |
| r'The technical photographic qualities include', | |
| r'The artistic elements include' | |
| ], | |
| "simplify_lighting_descriptions": [ | |
| (r'The sun is setting, casting a warm, orange glow over the scene', 'golden hour lighting'), | |
| (r'soft and warm.*?light', 'soft warm lighting'), | |
| (r'dramatic contrast with the dark, shadowy areas', 'dramatic shadows'), | |
| (r'long shadows and creating a sense of depth', 'deep shadows') | |
| ], | |
| "condense_mood_descriptions": [ | |
| (r'The mood is somber and melancholic.*?contrast to the dark, desolate environment', 'melancholic mood with dramatic contrast'), | |
| (r'sense of contemplation or longing', 'contemplative mood'), | |
| (r'sense of desolation and abandonment', 'desolate atmosphere') | |
| ] | |
| }, | |
| # Mandatory generative keywords | |
| "mandatory_keywords": { | |
| "quality": ["photorealistic", "ultra-detailed", "sharp focus"], | |
| "technical": ["professional photography", "high resolution"], | |
| "style_by_scene": { | |
| "cinematic": ["cinematic composition", "film grain"], | |
| "portrait": ["shallow depth of field", "bokeh"], | |
| "landscape": ["wide shot", "epic scale"], | |
| "street": ["candid moment", "documentary style"], | |
| "architectural": ["geometric composition", "clean lines"] | |
| } | |
| } | |
| } | |
| # Enhanced Scoring Configuration with Professional Photography Criteria | |
| SCORING_CONFIG = { | |
| "max_score": 100, | |
| "score_weights": { | |
| "prompt_quality": 0.25, | |
| "technical_details": 0.25, | |
| "professional_photography": 0.25, | |
| "multi_engine_optimization": 0.25 | |
| }, | |
| "grade_thresholds": { | |
| 95: {"grade": "LEGENDARY", "color": "#059669"}, | |
| 90: {"grade": "EXCELLENT", "color": "#10b981"}, | |
| 85: {"grade": "VERY GOOD", "color": "#22c55e"}, | |
| 75: {"grade": "GOOD", "color": "#84cc16"}, | |
| 65: {"grade": "FAIR", "color": "#f59e0b"}, | |
| 50: {"grade": "NEEDS WORK", "color": "#f97316"}, | |
| 0: {"grade": "POOR", "color": "#ef4444"} | |
| }, | |
| "professional_criteria": { | |
| "camera_equipment": ["Canon", "Sony", "Leica", "Hasselblad", "Phase One", "ARRI"], | |
| "lens_specifications": ["f/", "mm", "anamorphic", "telephoto", "wide-angle"], | |
| "technical_settings": ["ISO", "aperture", "f/", "shutter speed", "exposure"], | |
| "lighting_techniques": ["key light", "rim light", "fill light", "3-point lighting", "lighting ratio"], | |
| "composition_rules": ["rule of thirds", "leading lines", "depth of field", "bokeh", "symmetrical", "centered"], | |
| "cinematography_terms": ["cinematic", "dramatic", "film grain", "anamorphic bokeh", "color grading"] | |
| } | |
| } | |
| # Environment Configuration | |
| ENVIRONMENT = { | |
| "is_spaces": os.getenv("SPACE_ID") is not None, | |
| "is_local": os.getenv("SPACE_ID") is None, | |
| "log_level": os.getenv("LOG_LEVEL", "INFO"), | |
| "debug_mode": os.getenv("DEBUG", "false").lower() == "true", | |
| "space_id": os.getenv("SPACE_ID", ""), | |
| "space_author": os.getenv("SPACE_AUTHOR_NAME", "") | |
| } | |
| # Enhanced BAGEL Prompts for Professional Analysis | |
| BAGEL_PROMPTS = { | |
| "multimodal_analysis": """Analyze this image for professional prompt generation. Provide exactly two sections: | |
| 1. DESCRIPTION: Create a detailed, flowing paragraph describing the image, including: | |
| - Image type (photograph, illustration, artwork, scene) | |
| - Subject matter and composition elements | |
| - Color palette, lighting conditions, and mood | |
| - Visual style, artistic elements, and photographic techniques | |
| - Any cinematic or dramatic qualities | |
| 2. CAMERA_SETUP: Recommend professional camera and lens setup based on scene analysis: | |
| - For portraits: High-end camera with portrait lens (85mm-135mm f/1.4-f/2.8) | |
| - For landscapes: Medium format with wide lens (24-40mm f/4-f/8) | |
| - For street/documentary: Compact system with standard lens (35mm-50mm f/1.4-f/2.8) | |
| - For cinematic scenes: Cinema camera with appropriate lens for mood | |
| - Include specific aperture, focal length, and shooting considerations | |
| Focus on creating prompts suitable for multiple generative engines (Flux, Midjourney, etc.).""", | |
| "flux_optimization": """Analyze this image for FLUX prompt generation with professional photography principles: | |
| 1. DESCRIPTION: Detailed scene description focusing on: | |
| - Visual composition and subject placement | |
| - Lighting quality, direction, and mood | |
| - Color relationships and tonal values | |
| - Artistic style and photographic approach | |
| - Technical qualities that enhance realism | |
| 2. CAMERA_SETUP: Professional equipment recommendation: | |
| - Camera body suitable for the scene type | |
| - Lens choice with specific focal length and aperture | |
| - Lighting setup considerations | |
| - Technical settings for optimal results | |
| Generate content optimized for photorealistic output.""", | |
| "cinematic_analysis": """Analyze this image for cinematic prompt creation: | |
| 1. DESCRIPTION: Focus on cinematic elements: | |
| - Scene composition and framing | |
| - Lighting mood and dramatic elements | |
| - Color grading and visual atmosphere | |
| - Character positioning and environmental context | |
| - Storytelling visual cues | |
| 2. CAMERA_SETUP: Cinema-grade equipment: | |
| - Professional cinema camera recommendation | |
| - Lens choice for cinematic effect | |
| - Lighting setup for mood and story | |
| - Technical considerations for film-quality output | |
| Optimize for high-end generative engines with cinematic capabilities.""" | |
| } | |
| # Professional Photography Integration Settings | |
| PROFESSIONAL_PHOTOGRAPHY_CONFIG = { | |
| "enable_expert_analysis": True, | |
| "knowledge_base_integration": True, | |
| "technical_enhancement": True, | |
| "composition_guidance": True, | |
| "prompt_condensation": True, | |
| "scene_detection_keywords": { | |
| "portrait": ["person", "face", "portrait", "headshot", "model"], | |
| "landscape": ["landscape", "nature", "mountain", "sky", "horizon"], | |
| "street": ["street", "urban", "city", "documentary", "candid"], | |
| "architectural": ["building", "architecture", "structure", "interior"], | |
| "cinematic": ["film", "movie", "cinematic", "dramatic", "story"], | |
| "commercial": ["product", "commercial", "advertising", "brand"] | |
| }, | |
| "enhancement_priorities": [ | |
| "technical_accuracy", | |
| "professional_terminology", | |
| "equipment_specifications", | |
| "lighting_description", | |
| "composition_analysis", | |
| "prompt_optimization" | |
| ], | |
| # Prompt optimization settings | |
| "prompt_optimization": { | |
| "max_length": 150, | |
| "remove_redundancy": True, | |
| "add_technical_specs": True, | |
| "include_composition_terms": True, | |
| "mandatory_iso_settings": True, | |
| "enforce_aperture_format": True | |
| } | |
| } | |
| # Flash Attention Installation Command | |
| FLASH_ATTN_INSTALL = { | |
| "command": "pip install flash-attn --no-build-isolation", | |
| "env": {"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"}, | |
| "shell": True | |
| } | |
| # Export main configurations | |
| __all__ = [ | |
| "APP_CONFIG", | |
| "BAGEL_CONFIG", | |
| "get_device_config", | |
| "get_bagel_device_map", | |
| "PROCESSING_CONFIG", | |
| "FLUX_RULES", | |
| "SCORING_CONFIG", | |
| "ENVIRONMENT", | |
| "BAGEL_PROMPTS", | |
| "PROFESSIONAL_PHOTOGRAPHY_CONFIG", | |
| "FLASH_ATTN_INSTALL" | |
| ] |