FAQ

FAQ

Can I use this at work?

The plugin is MIT licensed. What decides it is your company's policy on screenshots of its own systems, and the licence of the base model. Z-Image, Qwen-Image-2512 and FLUX.2 klein 4B are Apache-2.0. Everything runs locally unless you choose a hosted GPU or a cloud vision model.

Does it need my password?

No. login opens a browser window and you sign in yourself. --cdp attaches to the browser you already have open. Nothing stores or types a password. The auth/ folder does hold a live session, so keep it private.

Will it click things on our site?

No. Capture opens pages with GET and scrolls them. It clicks only what you name with --click, and the crawl skips sign-out, delete and cancel links.

What about customer data on screen?

Capture replaces e-mail addresses, digit runs and typed field values, and hides pop-ups. It cannot spot names, so use a test account or staging, or --hide those regions.

How many screenshots?

20-40 good pictures for a site, and 40-100 across sites for the brand. Variety of page types matters more than volume.

How long does training take?

On a 16 GB RTX 5060 Ti, Z-Image takes about 7 s a step at 1024 px, rank 32. So 1,500 steps is about three hours. FLUX.2 klein 4B is several times faster. A 24 GB card or a cloud GPU is faster still.

The text in the pictures is wrong

That is normal for small text. Use short labels in double quotes and bigger component sizes, or build the final with the code route.

The header shows a made-up name or the trigger word

Name the header text in the prompt, in quotes: the header shows the name "Fishing Service". A trigger that is not a real word does not help. It was tested, and the model printed it in the page more often.

Can I share the trained LoRA?

Not outside the company. It reproduces the brand's trademarks.

Why does ref say "up to 4"?

FLUX.2 klein takes four reference images (BFL's spec). More are dropped.