Contact

One person reads this inbox. Email is the only channel, because it is the only one worth maintaining for a site this size.

Email

[email protected]

What to expect

Every message gets read. Most get a reply. Being straight with you about the limits: this is a one-person project, so there is no support desk, no ticket system, and no guaranteed response time. If your message is a quick bug report it will usually get a faster answer than a long feature request, simply because one is easier to act on.

If you do not hear back, it is almost certainly because the question was already answered elsewhere on the site rather than because the message was ignored — the FAQ covers most of what people write in about.

What is worth writing in about

A format that will not load. JPG, PNG and WEBP should all work. If one of them fails, tell me the format and roughly the file size and I will look at it.
A measurement that looks wrong. The Read from your image panel shows the raw numbers. If a number is clearly wrong for the image, that is a bug and I want to know.
A prompt that reads badly for a specific model. Especially Midjourney or Flux, since those two are formatted quite differently.
Something on the site that is inaccurate. Including this page.

Known issues

Three things are already known and are written up on other pages. Each one has a condition attached, so check yours against it before writing in.

HEIC does not load outside Safari. The tool takes any format your browser can decode and draw to a canvas. JPG, JPEG, PNG, WEBP and GIF always qualify, and on Safari and iOS HEIC does too. On other browsers HEIC is not decodable, so the file never reaches the canvas. The condition is the browser rather than the file — your file size is never read and there is no size cap. The format list is on the front page.
A transparent background is measured as black. The analyser reads the red, green and blue bytes and never the alpha byte, and nothing is painted behind the image first. On a test buffer that is half transparent black (RGB 0, alpha 0) and half opaque white, meanL comes back 127.50 of 255; changing the invisible half to RGB 255 with alpha still 0 moves it to 255.00. The condition is an unfilled transparent area, so flatten the background before you drop the file in. The test is written up on where your image goes.
A ultra-wide panoramic image — wrong article. The sentence-shaped outputs come from a template that always opens with A, so the article is never chosen for a vowel-initial word. The condition is a width-to-height ratio of 2.2 or above, which is the only framing band whose phrase starts with a vowel. Change it to An and move on. Written up on fix wrong image to prompt.

If you are writing in about a bug

Three things make a report actionable, and they take about a minute to gather:

What you did — which format you uploaded, which model and prompt length you had selected.
What you expected and what actually happened — paste the generated prompt if there was one.
Your browser and device — Chrome on Windows, Safari on iPhone, and so on. A lot of this tool depends on canvas behaviour that differs between browsers.

Please do not attach the image itself. The whole point of the tool is that images never need to travel anywhere, and that includes bug reports — a description is almost always enough.

Using the site

Add an image. Drop a file on the box on the front page, paste one, or pick one from your device.
Pick the model. Midjourney, Stable Diffusion, Flux or DALL·E — the prompt is written differently for each.
Copy it out. The prompt appears beside the image, and the panel under it lists every value that was measured.

Nothing to install, no account to make, and the image never leaves your browser. The How it works and FAQ sections on the front page cover the rest.