Editing and mixing text prompts using design tool paradigms
Reprompt is a prototype that explores prompt editing through the lens of familiar design paradigms—resizing, layers, and boolean operations—using Anthropic's API. Rather than reverse-engineering prompts from AI-generated images, Reprompt asks: what if we could remix them?
The goal was to experiment with new interaction models that make prompt manipulation more intuitive, playful, and expressive—less like editing text, and more like editing visuals. By borrowing workflows from tools like Figma and Illustrator, this project imagines a future where prompt engineering feels more like creative direction—guided by gestures like expanding, subtracting, combining, or isolating semantic layers—with a multimodal LLM dynamically responding behind the scenes.
The main goal of this prototype was to explore ideas that might translate into real creative workflows—or help uplevel existing ones in design tools.
Resizing a prompt could unlock a new way to preserve a subject while adjusting the depth and detail of guidance—similar to how adjusting the number of anchor points in a vector shape changes its complexity. This allows for a more nuanced approach to prompt scaling, without rewriting from scratch.
Semantic layers could let users segment a prompt into meaningful parts—potentially tied to different regions of an image—so you could make targeted edits without manually drawing a mask or rewriting the entire prompt. This could lead to more intelligent, AI-assisted ways to select objects, inpaint, and create sub-prompts for specific areas of an image.
Boolean operations like union, subtract, and intersect could open up new ways to remix prompts. Rather than relying on vague style references or composition blending, users could more intentionally control how different ideas influence the final output.
Ultimately, this is about reimagining prompt engineering as something closer to creative direction—leveraging familiar design paradigms to make working with LLMs feel more intuitive, visual, and expressive.
