I created a series of UI explorations and bento grids in Figma to experiment with different product categories, audiences, and visual directions. Each concept uses free typefaces and custom visual assets generated through a Claude Code + Higgsfield MCP workflow, selecting and optimizing AI models based on the specific product, aesthetic, and design needs.

While the experiment was highly creative and enjoyable, it also revealed some clear limitations in the workflow. Bento grids are particularly demanding motion systems, as multiple visual elements, interactions, and text animations need to be interpreted consistently at the same time. Although the motion design skill used in this process had performed well in previous explorations, this level of visual density proved more challenging for Seedance 2.0. The prompts followed established video-generation best practices, but text animation remained the main constraint and required several iterations to achieve acceptable results. In practice, the model proved significantly more reliable when animating imagery and visual assets than when handling typography and precise text behavior.

​

One especially successful part of the process was the music direction suggested by the AI for each concept. The selected tracks complemented the visual language of each bento grid particularly well, helping reinforce the personality of each digital product and creating a stronger connection with its intended audience.