Editor
Image to Map
Upload a reference image — a sketch, a screenshot, concept art — and ForjeGames will turn it into a playable 3D map.
Overview
Image to Map runs in the Image editor mode. You drop an image into the chat, add an optional text description, and the AI analyses the picture to recreate its layout, colour palette, and atmosphere as a Roblox map. It's the fastest way to turn a mood board or a quick sketch into something playable.
How it works
Under the hood, each image goes through three stages:
- Vision analysis — Claude Vision describes the scene in structured form: dominant shapes, objects, colour palette, lighting, mood.
- Depth estimation — Apple's Depth Pro model estimates a relative depth map so we know where the foreground and background sit.
- Map composition — the structured description and depth map are combined with the Build pipeline to place terrain, props, and lighting.
Uploading an image
You can attach images in three ways:
- Drag the file onto the chat panel.
- Click the + button below the chat box and pick Attach image.
- Paste directly from the clipboard (⌘ V). Works with screenshots.
Accepted formats: PNG, JPG, WebP. Max size: 10 MB per image.
What makes a good input
- Clear composition — a single, unambiguous subject beats a busy collage.
- Readable shapes — the AI doesn't need photorealism; crude geometric sketches work great.
- Distinct colours — strong colour contrast helps the model infer materials.
- One viewpoint — don't try to show all four walls of a room in one picture. Take separate shots instead.
Examples of inputs that work well: hand-drawn top-down maps on graph paper, screenshots from other games, Pinterest mood boards, concept art, architectural elevations.
Controlling the output
You can steer the result with a text description alongside the image. Mention scale, player count, gameplay loop, and any details the image doesn't show.
Example:
"Use this sketch as the top-down layout. It should be a 4-player horror map — dark lighting, fog, distant thunder audio. The red X is the spawn."
Limitations
- Text in images is ignored — we don't OCR signs or labels. Describe them in the prompt instead.
- Exact proportions drift — depth estimation is relative, so the output won't be pixel-accurate to the reference.
- Copyrighted art — don't upload images you don't have the right to use. The output is yours, but the input must be too.