| SYS_PROMPT = "" | |
| USER_PROMPT = """# CONTEXT # | |
| You are a powerful video captioner.I want to tag 200,000 video files for use in training a text-to-video dataset. The purpose of the video tags is to train a text-to-video model. You need to provide a structured, detailed, and accurate description of the given video. | |
| # OBJECTIVE # | |
| Video Description Task Instructions | |
| Video Content Description: | |
| Detail and Accuracy: Provide a detailed and accurate description of the video content. Include all key objects, their types, colors, actions, positions, and relative positions. Describe the overall atmosphere. | |
| Persons and Animals: If there are people, describe their appearance and actions. If there are animals, describe their behavior to give a clear understanding of the scene. | |
| Multiple Scenes: If the video has multiple scenes, describe how they transition and highlight the differences between them. | |
| Objectivity: Do not include imagined content or overly subjective feelings. Ensure all descriptions are based on what can be confidently determined from the video. | |
| Grammar and Length: Use correct English grammar. Each descriptive sentence should be at least three sentences long. | |
| Video Quality Evaluation: | |
| Aesthetic Value: Evaluate the aesthetic value, including composition, color harmony, and overall visual effect. Score this aspect from 1 to 5 and explain your reasoning. | |
| Clarity: Assess the clarity, including resolution and detail presentation. Score this aspect from 1 to 5 and explain your reasoning. | |
| Emotional Impact: Evaluate the emotional impact, including how well the video conveys emotions and resonates with the audience. Score this aspect from 1 to 5 and explain your reasoning. | |
| Summary: Provide a summary of the scores for aesthetic value, clarity, and emotional impact. | |
| Film Perspective Analysis: | |
| Shot Analysis: Analyze the type of shots used (close-up, medium, long shot, etc.). | |
| Camera Movements: Describe the camera movements (push, pull, pan, tilt, track, crane, etc.). | |
| Composition: Analyze the composition of the shots. | |
| Interpretation: Provide your interpretation and feelings about the photographic work. | |
| # STYLE # | |
| cinematic language，such as narrative techniques, visual aesthetics, editing styles, and sound design. | |
| # Output Structure # | |
| Video Content: | |
| {Detailed description of the video here, meeting the above requirements}. | |
| Video Quality: | |
| {Evaluation score and explanation of the video quality here}. | |
| Film Perspective Description: | |
| {Analysis of the video from a film perspective here}. | |
| Example: | |
| Video Content: | |
| A stylish woman strides down a Tokyo street illuminated by warm neon lights and animated city signage. She sports a black leather jacket, a long red dress, black boots, and carries a black purse. Her look is completed with sunglasses and red lipstick. Her demeanor is confident and casual. The damp street reflects the vibrant lights, creating a mirror effect. The scene is bustling with numerous pedestrians. | |
| Video Quality: | |
| Aesthetic Value: | |
| - Composition and Color: The video showcases a well-balanced composition with harmonious color schemes, achieving a visually pleasing effect. Techniques such as symmetry and dynamic composition are skillfully employed. | |
| - Camera Work: The visual experience is enhanced by smooth transitions and diverse angles. | |
| - Score: 4/5 | |
| Clarity: | |
| - Resolution: The video boasts high resolution with clear details. | |
| - Detail Presentation: It presents rich details with no noticeable blurriness or distortion. | |
| - Score: 5/5 | |
| Emotional Impact: | |
| - Emotion Conveyance: The video successfully conveys joy and excitement, striking a chord with the audience. | |
| - Resonance: The compelling emotional expression, supported by well-integrated music and visuals, creates a strong impact. | |
| - Score: 4/5 | |
| Summary: | |
| - Aesthetic Value: 4/5 | |
| - Video Clarity: 5/5 | |
| - Emotional Impact: 4/5 | |
| Film Perspective Description: | |
| Characters: | |
| - Woman: A stylish woman dressed in a black leather jacket, long red dress, black boots, and carrying a black purse. She wears sunglasses and red lipstick. | |
| Scenes: | |
| - Tokyo Street: The street is filled with warm glowing neon lights and animated city signage, with damp reflective surfaces and numerous pedestrians. | |
| Shot 1: | |
| - The woman walks confidently and casually down the Tokyo street. | |
| - She heads towards the camera in a panoramic view with central composition. The camera is at eye level and follows her with a handheld shot. | |
| - Duration: 36 seconds | |
| Shot 2: | |
| - The woman continues her walk down the Tokyo street, maintaining her confident and casual demeanor. | |
| - She approaches the camera, with a close-up of her face, transitioning to a torso mid-shot. The camera remains at eye level, following her with a handheld shot. | |
| - Duration: 24 seconds | |
| """ | |
| SKIP = 2 | |
| TEMP = 0.3 | |
| TOP = 0.75 | |
| MAX_TOKEN = 512 | |
| API_CLASSES = { | |
| 'Azure': 'AzureAPI', | |
| 'Google': 'GoogleAPI', | |
| 'Anthropic': 'AnthropicAPI', | |
| 'OpenAI': 'OpenAIAPI' | |
| } | |
| PROVIDERS_CONFIG = { | |
| 'Azure': { | |
| 'model': ['GPT-4o', 'GPT-4v'], | |
| 'key_label': 'Azure API Key', | |
| 'endpoint_label': 'Azure Endpoint' | |
| }, | |
| 'Google': { | |
| 'model': ['Gemini-1.5-Flash', 'Gemini-1.5-Pro'], | |
| 'key_label': 'Google API Key', | |
| 'endpoint_label': 'Google API Endpoint' | |
| }, | |
| 'Anthropic': { | |
| 'model': ['Claude-3-Opus', 'Claude-3-Sonnet'], | |
| 'key_label': 'Anthropic API Key', | |
| 'endpoint_label': 'Anthropic Endpoint' | |
| }, | |
| 'OpenAI': { | |
| 'model': ['GPT-4o', 'GPT-4v'], | |
| 'key_label': 'OpenAI API Key', | |
| 'endpoint_label': 'OpenAI Endpoint' | |
| } | |
| } | |
| GENERAL_CONFIG = { | |
| 'temp': { | |
| 'label': 'Temperature', | |
| 'default': 0.3, | |
| 'min': 0, | |
| 'max': 1, | |
| 'step': 0.1 | |
| }, | |
| 'top_p': { | |
| 'label': 'Top-P', | |
| 'default': 0.75, | |
| 'min': 0, | |
| 'max': 1, | |
| 'step': 0.1 | |
| }, | |
| 'max_tokens': { | |
| 'label': 'Max Tokens', | |
| 'default': 4096, | |
| 'min': 512, | |
| 'max': 4096, | |
| 'step': 1 | |
| }, | |
| 'frame_format': { | |
| 'label': 'Frame Format', | |
| 'default': 'JPEG', | |
| 'choices': ['JPEG', 'PNG'] | |
| }, | |
| 'frame_skip': { | |
| 'label': 'Frame Skip', | |
| 'default': 2, | |
| 'min': 2, | |
| 'max': 100, | |
| 'step': 1 | |
| }, | |
| 'group_size': { | |
| 'label': 'Group Size', | |
| 'default': 10, | |
| 'min': 1, | |
| 'max': 100, | |
| 'step': 1 | |
| } | |
| } | |