Avoiding AI Design Slop with Claude Design 2.0
DesignCoursego watch the original →
the gist
Avoid generic AI output by building a custom design system, feeding it specific visual references, and using iterative prompting to refine components rather than accepting one-shot results.
Establishing a Design Foundation
To avoid generic AI output, treat the design process as a collaborative effort where the model acts as an executor for your specific creative direction. Start by gathering high-quality inspiration from sites like Mobbin to identify a specific aesthetic, such as "fun" or "brutalist," rather than relying on default AI styles. Export these references as JPEGs or PNGs to feed into Claude Design 2.0 as assets. When creating the design system, define a clear brand identity and upload these reference images to guide the model's color palette, typography, and component motifs.
Iterative Refinement
Avoid accepting the initial output as final. Use the feedback and edit features in the design system interface to force the model to adhere to specific constraints. For example, instead of accepting default sticker motifs, provide precise instructions: "Add a black icon of a person running, encased in a white circle, center-aligned at the top of each motif." Use the edit interface to manually override color values for primary, hover, and background states to ensure they align with your brand vision rather than the model's default choices.
Directing the Hero Section
When generating layouts, provide detailed prompts that specify component placement and interaction. If the initial result is unsatisfactory, use follow-up prompts to refine specific elements. For instance, if a character needs to overlap text or requires specific animation, provide a screenshot of the current state alongside a reference image of the desired style. Continue to iterate by adjusting individual properties, such as pupil rotation, character scale, or scroll behavior, until the design matches your intent. The goal is to move beyond the "one-shot" workflow, which produces generic results, toward a process of continuous refinement.