Summary
Getting a cinematic result from an AI video generator often means burning credits on repeated takes. This tutorial from Youri van Hofwegen shows how to plan prompts more carefully and build a video from scratch in one place.
The workflow connects Claude to Higgsfield through its MCP connector, so prompts can be written, adjusted and generated in a single chat. Setup involves copying the connector link from Higgsfield and adding it as a custom connector in Claude’s settings. The video stresses spelling out the model, resolution and aspect ratio, locking texture with a reference image and naming the camera move explicitly rather than leaving defaults.
The walkthrough covers several prompt styles. Image prompting is used to build a three-panel character sheet with GPT Image 2 for consistency across shots. The environment, a rain-slick rooftop at night, is built separately as an image and refined with Nano Banana Pro texture and lighting passes. Cinematic and negative prompts then control camera movement and suppress unwanted shake or text, with clips generated on a newer video model.
Viewers will come away with a repeatable process for consistent characters and scenes, and fewer wasted generations. The video includes a promotion for a Higgsfield bonus package.
📺 Source: Youri van Hofwegen · Published September 25, 2026
🏷️ Format: Tutorial Demo
