Log message with prefix.

Convert PIL Image to tensor.

Create an empty black image tensor with proper format for ComfyUI

Extract frames and audio using PyAV

Convert ComfyUI tensor directly to genai Image format

Convert tensor to PIL Image with RGBA support

Clear CUDA memory.

Quick check if network is available - returns within 1 second

Process audio data for API submission with robust error handling

Randomly select styles from different categories

Convert tensor to PIL Image, handling various shapes.

Load counters from persistent file

Set counter for a path

Show helpful error message for SAM3 download.

Clear all cached models.

Create a very simple preview image for SVG

Generate a random base prompt idea

Refine mask using Guided Filter.

Sample frames evenly from video tensor

Convert tensor to base64 encoded PNG

Combine RGB image with alpha mask.

Check if ben2 package is installed

Check transformers version for compatibility

Get current counter for a path

Helper to get configuration based on model name

Get sorted list of image files in directory

Save counters to persistent file

Combine multiple style elements into a cohesive prompt

Convert image to SVG using vtracer with optimized parameters for best quality

Ensure we use a new random seed that changes each run

Generate trimap from binary mask.

Load BiRefNet model - supports multiple variants.

Dynamically fetches a list of Gemini models that support image generation. Models (verified against the deprecation schedule on 2026-05-26):

Load SAM3 using HuggingFace transformers (official implementation). This uses the official facebook/sam3 model from transformers libr

Return a list of available style categories from the data directory

Load ViTMatte model.

Return True if the given model name supports thinking/reasoning config.

Load BEN2 model from HuggingFace

Optimize SVG content to reduce file size using Python-only methods

Convert PIL Image to google.genai types.Image format

Refine mask using BiRefNet models - best for hair and fine edges.

Attempt to render the actual SVG content

Attempt to render the actual SVG content

Convert RGBA image to RGB with white background

Convert RGBA image to RGB with white background

Smart search for model file in standard ComfyUI model directories. Prioritizes specific folders (e.g., models/sams) but falls back to root/ul

Convert tensor to PIL Image with RGBA support

Convert ComfyUI tensor to PIL Image