Descriptions:
Distorted proportions and mismatched camera angles are among the most common failure modes in AI-generated character imagery and video. This tutorial from Veteran AI introduces Proportion Changer, a ComfyUI extension that solves both problems through a technique called Pose Alignment — extracting skeletal pose data as JSON from a source image or video, then using a reference transformation node to remap that pose to match the proportions and perspective of a different character or viewpoint.
The guide covers three distinct production workflows in depth. First, integration with ControlNet using ZImage Turbo, where the extension corrects mismatches between 3D reference model proportions and real human anatomy. Second, use within Steady Dancer video generation, demonstrating how Proportion Changer aligns a driving video’s full-body perspective to a half-body reference image — producing character-consistent output across shots that would otherwise drift in composition and framing. Third, multi-angle perspective switching, where a single driving video can generate outputs ranging from full-body to tight chest-level framing while maintaining pose coherence throughout.
Key technical nodes covered include the DW Pose Detector for JSON pose extraction, the Pose Render Node for visualizing body, face, and hand components, and the Proportion Changer Reference transformation node itself. All workflows demonstrated in the video are available on RunningHub, an online ComfyUI workspace that hosts up-to-date extensions without requiring local installation. The technique is particularly valuable for animators, filmmakers, and character artists working with reference-based generation who need consistent subject framing across multi-shot sequences.
📺 Source: Veteran AI · Published December 18, 2025
🏷️ Format: Tutorial Demo
