Summary
Creator Youri van Hofwegen walks through a complete end-to-end workflow for generating AI video content that maintains a consistent likeness across every shot — solving one of the most common failure modes in AI filmmaking. The pipeline runs entirely inside OpenArt, combining the platform’s character builder, the newly integrated GPT Image 2 model, and Seedance 2 for animation.
The process begins with uploading multiple face photos from different angles to OpenArt’s character builder to create a named, reusable character profile. Reference images are then generated using GPT Image 2 — with the @ symbol syntax used inside prompts to anchor the AI to a specific character identity. The video demonstrates that adding the single word “photorealistic” to a prompt significantly shifts output quality toward camera-realistic results, a discovery the presenter made through side-by-side testing. For location generation, Claude analyzes the existing reference images and produces a matching location prompt that inherits the established color palette, lighting, and mood automatically.
The final scene — a two-character action sequence in a Japanese dojo — is assembled in Seedance 2 using up to 15-second clips at 1080p resolution, with both character and location references passed simultaneously so the model knows exactly who appears where. Viewers come away with a reproducible framework for face-consistent AI video production, practical prompt techniques for GPT Image 2, and a clear understanding of how to chain Seedance 2 clips into a coherent edited sequence.
📺 Source: Youri van Hofwegen · Published May 07, 2026
🏷️ Format: Tutorial Demo
