How it works

This tool converts official LEGO instruction PDFs into accessible, text-based building instructions that a blind or visually impaired person can follow without seeing any images. It works as a four-phase pipeline — some phases are automated by AI, others need a sighted human to verify the output.

The pipeline

▶ Automated — no human input needed
Phase 1 — PDF processing

The app downloads the official instruction PDF and extracts a crop image for every building step — the assembly diagram and the parts inset. This phase is fully automated and free (no AI is used).

✎ Sighted reviewer needed
Phase 2 — Piece identification

An AI model analyses the parts inset image for each step and identifies which pieces are needed — quantity, colour, and part name. Because the AI can make mistakes (wrong colour, wrong count, wrong part), a sighted reviewer must check each identification and correct anything that is wrong before the next phase runs.

Your job: Look at the inset image. Does each piece in the table match what you see? If a colour is wrong, a quantity is off, or a part is misidentified — edit it and save. Accuracy here is critical.
✎ Sighted reviewer needed
Phase 3 — Spatial placement

The AI compares two images — the build before and after each step — and writes a description of where each piece was placed, using directions like "3 studs from the left edge of the grey plate". A sighted reviewer must check that every description is accurate, specific, and uses the correct directions before approving it.

Your job: Read each placement description carefully against the before and after images. If a direction is wrong, a stud count is off, or the description is too vague — click Edit and correct it. Once every placement for a step is accurate, click Approve. This is the most important phase.
🔒 Owner step — no reviewer input needed
Phase 4 — Accessible prose

Using the approved placement descriptions from Phase 3, the app generates a single plain-English paragraph for each step in the style of BricksForTheBlind instructions. This prose is reviewed by the builder themselves — no sighted reviewer input is needed at this stage.

Frequently asked questions

Do I need to review every step?

Yes — every step needs Phase 2 and Phase 3 reviewed before the final prose can be generated. Steps can be done in any order, but all must be completed for the full booklet to be usable.

What if I'm not sure about a placement description?

Look at both the before and after images carefully. If you genuinely cannot tell where a piece went from the images alone, note that in the description — the builder and their support network can then refer to the original PDF for that step. An honest "placement unclear from image" is more useful than a confident wrong answer.

How precise do I need to be?

As precise as the images allow. Stud counts are the gold standard — "place 2 studs from the left" is far more useful than "place near the left side". If you can count studs, count them. If the image is too small or unclear, say so.

Why this matters

Every piece you identify and every placement description you verify becomes part of the instructions a blind person will use to build that LEGO set — independently, without needing a sighted helper in the room with them. The quality of your review directly determines whether they can succeed.

Thank you for being part of making LEGO building accessible.

Start reviewing →