Descriptions:
Youri van Hofwegen demonstrates a custom AI video prompt generator that converts rough scene ideas into structured, model-specific prompts for AI video tools including Seedance 2.0, VO3, and Cling — all hosted within the Higsfield platform. The core insight is that each AI video generator has a distinct prompt grammar, and generating output in JSON format from the start (rather than natural language) meaningfully improves result quality by matching how the underlying models actually parse instructions.
The tutorial walks through three modes in sequence. The first (text-only) takes a plain description — a rally car drifting through a desert hairpin — and generates a full JSON prompt including subject, action, shot type, lens specification (35mm wide), lighting, color palette, and a negative prompt flagging unwanted elements. The second mode introduces a reference image created via Higsfield’s Cinema Studio, which was trained exclusively on real film footage for cinematic output. The third mode handles stylized non-realistic content like animation and claymation, demonstrated with a Pixar-style character.
Practical details covered include switching between Seedance 2.0 and other models, using the regenerate button when output misses the mark, and copying prompts directly into Higsfield’s video generator. The finished clips show smooth physics-accurate water motion and natural character movement — results the creator attributes directly to the structured prompting approach.
📺 Source: Youri van Hofwegen · Published June 02, 2026
🏷️ Format: Tutorial Demo
