AI prompting training tool
Built internally at Artefact, Make a Scene is a prompt engineering training tool applied to image generation. Rather than reading guidelines, participants build a prompt word by word through a guided funnel: 3 steps, 14 sub-steps. Each option has a preview image on hover. At the end of the funnel, 4 variants are generated in a single request. Available in English and French.
3 steps, 14 sub-steps. Each chosen option is added in real time to the prompt bar at the bottom of the screen. At the end of the funnel, 4 image variants are generated in a single request.
A tool to learn how to talk to an AI.
Rather than reading guidelines, participants build a prompt word by word: image type, subject, action, location, style, lighting, ratio. Each choice is added in real time to the prompt bar visible at the bottom of the screen.
A visible result at every step.
Hovering over each option shows an example image to help choose. You only advance when all fields are filled, which forces prompt precision.
Four images generated in seconds.
Once the funnel is complete, the tool generates four visual variants from the built prompt. The full prompt is displayed and copyable. You can go back to any step and start over.
Available in French and English.
Bilingual interface, built for international teams.
Each screen guides the user toward a keyword to select — image type, location, render style, ratio. The prompt assembles at the bottom of the screen as choices are made. The forward arrow only appears when all fields are filled.
Once the funnel is complete, 4 variants are generated. A progress bar displays during generation, then the 4 images appear side by side, with the full prompt below, copyable with one click.
For developers and recruiters: why a Zustand store for 3×14 sub-steps, why a single request for 4 variants, and why generate previews offline rather than on the fly.
The funnel is divided into 3 thematic steps, themselves split into 14 sub-steps. Each block of the assembled prompt is clickable: the user can navigate back directly to any sub-step without losing their choices. The entire funnel state is centralized in a Zustand store, which eliminates prop drilling and makes back navigation trivial to implement.
Rather than chaining multiple calls, a single request with the `n: 4` parameter generates all 4 variants server-side at once. No Promise.all, no polling: the Server Action waits for the complete response and returns it to the client. Result: simple code, minimal delay, no intermediate state management.
Each funnel option has an example image visible on hover. Rather than generating these images on the fly, they were produced once in an offline script and are served statically. No API call at runtime, no latency for the user.
A training tool to build?
I design and build guided funnels and internal tools integrating AI image generation.
Multi-step funnel, Zustand store, 1-request image generation, Storybook design system: this project laid the groundwork I can bring to your next build.