Custom Models
Adding and configuring custom AI models for video analysis.
VPA comes with pre-configured support for popular AI models, but you can also add custom models for specialized use cases or to access the latest releases before we add official support.
Pre-configured Models
VPA includes optimized configurations for:
Vision/Analysis Models
- GPT-4o (OpenAI)
- GPT-4 Vision (OpenAI)
- Claude 3.5 Sonnet (Anthropic)
- Claude 3 Opus (Anthropic)
- Gemini Pro Vision (Google)
Text Generation Models
- GPT-4 Turbo (OpenAI)
- Claude 3 Sonnet (Anthropic)
- Gemini Pro (Google)
💡
Model selection
Different models have different strengths. GPT-4o excels at detailed visual analysis, while Claude 3.5 Sonnet offers excellent prompt generation capabilities.
Adding a Custom Model
- Navigate to Settings > Models
- Click "Add Custom Model"
- Enter the model configuration (see below)
- Test the model with a sample analysis
- Save if the test succeeds
Model Configuration
Each custom model requires the following configuration:
Required Fields
- Model ID: The exact model identifier used by the provider (e.g., "gpt-4o-2024-05-13")
- Provider: Which API provider to use (OpenAI, Anthropic, Google)
- Display Name: A friendly name shown in the UI
- Type: Vision (for analysis) or Text (for generation)
Optional Fields
- Max Tokens: Maximum output tokens (default varies by provider)
- Temperature: Creativity level 0-1 (default: 0.7)
- System Prompt Override: Custom instructions for this model
- Cost per Token: For usage estimation
Custom OpenAI-Compatible Endpoints
VPA supports OpenAI-compatible API endpoints, allowing you to use:
- Self-hosted models (LLaMA, Mistral via vLLM)
- Azure OpenAI Service
- OpenRouter and other aggregators
- Any OpenAI-compatible endpoint
Configuration for Custom Endpoints
- Base URL: Your endpoint URL (e.g., https://your-deployment.azure.com)
- API Version: Required for Azure (e.g., "2024-02-15-preview")
- Deployment Name: Your Azure deployment name (if applicable)
⚠️
Vision capability required
For video analysis, your custom model must support image inputs. Text-only models cannot be used for frame analysis.
Model Presets
Save your custom model configurations as presets for quick switching between different setups:
- Quick Analysis: Faster, lower-cost model
- High Quality: Best model for important projects
- Budget: Most cost-effective option
Troubleshooting Custom Models
- Model not found: Verify the exact model ID with your provider
- Unauthorized: Check API key permissions for the model
- No image support: Ensure the model has vision capabilities
- Rate limits: Some models have lower limits; reduce frame count