AI image with a working QR code: Muse Image uses code, others guess

Meta says Muse Image 'learns to write and execute code that produces accurate plots and QR codes' (Meta AI, read 2026-10-01). The image models in Sume's catalog take a prompt and return pixels; none is documented as running code. If you ask one for a QR code, treat what it draws as decoration that may not scan.

The dependable method is to split the job: have the image model make the poster with an empty, plain area, then generate the QR code with a library and paste it in. You control the data, the quiet zone and the contrast, and you can test the result with a phone.

Which approach fits which need?

The fix is a split, not a better prompt.

How do I make the poster and add the QR?

First prompt for the art: 'leave the bottom-right corner plain white and empty'. Generate through the Image API, keep the URL, then composite locally:

import io
import os
import qrcode
import requests
from PIL import Image

art_url = os.environ["ART_URL"]
art = Image.open(io.BytesIO(requests.get(art_url, timeout=60).content))
art = art.convert("RGB")

qr = qrcode.make("https://example.com/menu").convert("RGB")
qr = qr.resize((300, 300))
art.paste(qr, (art.width - 340, art.height - 340))
art.save("poster.png")
print("saved poster.png", art.size)

What should I check before printing?

Scan the saved file with two phones at the final print size. Keep a quiet margin of white around the code and use dark modules on a light ground. Do not let the image model edit the page after you paste the QR: a later edit call may redraw it. Put the QR step last.

Limits

Meta's claim covers Muse Image inside Meta's products, and its developer API availability is not stated in the launch posts I read, so I cannot say it is callable. Image models also misspell short URLs, so never ask one to render the link as text; print it from your own code. This pattern needs the Pillow and qrcode packages and a plain region in the art.