# Image Captioning Default Configuration # This file contains default settings for the Image Captioning tab [image_captioning] # Model Configuration model_path = "" # Path to Qwen2.5-VL model (qwen_2.5_vl_7b.safetensors) fp8_vl = true # Use FP8 precision for memory savings (recommended for <24GB VRAM) max_size = 1280 # Maximum image size for processing # Caption Generation Settings max_new_tokens = 1024 # Maximum tokens to generate per caption prefix = "" # Text to add before each caption suffix = "" # Text to add after each caption custom_prompt = "" # Custom prompt (empty = use default) # Generation Parameters (Optimized for diverse, coherent captions) do_sample = true # Enable sampling for text generation temperature = 0.7 # Controls randomness (0.1=focused, 1.0=creative, 0.7=balanced) top_k = 50 # Number of top tokens to consider (higher=more diverse vocabulary) top_p = 0.95 # Cumulative probability threshold (0.95 recommended for coherence) repetition_penalty = 1.05 # Penalty for repeating tokens (1.0=none, 1.05=slight penalty) # Word Replacement Settings replace_words = "" # Word replacement pairs (format: word1:replacement1;word2:replacement2) replace_case_insensitive = true # Replace words regardless of case replace_whole_words_only = true # Only replace complete words # Batch Processing Settings batch_image_dir = "" # Directory containing images for batch processing batch_output_folder = "" # Output folder for caption files (optional, empty = save alongside images) extension = ".txt" # Caption file suffix used in text output mode output_format = "text" # Output format: "text" or "jsonl" jsonl_output_file = "" # Output file path for JSONL format scan_subfolders = false # Include images from all subfolders recursively copy_images = false # Copy images to output folder overwrite_existing_captions = false # Replace existing caption files append_existing_captions = false # Append new captions to existing caption files when not overwriting # Advanced Settings enable_progress_tracking = true # Show progress during batch processing validate_images = true # Validate image files before processing auto_resize_images = true # Automatically resize images to optimal size