Kimi-VL Multi-Modal Integration Protocol

This skill provides the standardized workflow for implementing and utilizing multi-modal (visual) reasoning capabilities via the Kimi-VL client within the Forge NPS intelligence stack.

Overview

The KimiVLClient is a specialized extension of the KimiBridge. It is designed to handle visual assets (images/video frames) by encoding them into Base64 and injecting them into an OpenAI-compatible multi-modal message payload. This enables high-precision visual audits, consistency checks, and cinematic quality verification.

Implementation Architecture

1. Class Structure: KimiVLClient(KimiBridge)
2. Primary Method: analyze_visuals(image_path, system_prompt, user_prompt, schema, ...)
3. Routing Logic: Automatically forces the ModelTier.VISUAL tier via the KimiModelRouter.
4. Parameter Optimization: Uses a hardcoded low temperature ($0.1$) to minimize hallucination and maximize precision during visual audits.

Standard Workflow: Visual Auditing

When an agent (e.g., QA Agent) needs to perform a visual check: