ChatGPT has no /xray

The correction first

ChatGPT does not have image slash commands. There is no command parser reading the line you typed, no registry of supported commands, and no official list to be found. Type /xray and nothing intercepts it.

Every catalog you will find, including this one, is a collection of shorthand that people discovered works. That includes ours. We say it on every page and we say it here, because it is the single fact that changes how you use the rest of the site.

So why does it work

Because the model reads your whole message as text, and the word after the slash is an unusually dense description.

"X-ray" carries a specific look: layered translucency, internal structure showing through an outer shell, a cold blue-grey palette, high contrast at the edges. You could write that sentence yourself. The word carries it for you, and the model has seen enough of it to know what you mean.

The slash contributes nothing except a habit borrowed from Discord and Slack, where slashes really are commands. It survives because it reads as an instruction, and because it is easy to copy and paste.

What follows from that

• Results vary between models. A word that is dense for one model is vague for another.
• Results vary between accounts, and between the same account on two different days. Nothing here is a stable API.
• The command is not a switch. It is a request, weighted by everything else in your message.
• You can combine them, because they are just words. Nothing is stopping you.
• You can ignore them entirely and describe the look in a sentence. It is the same mechanism, spelled out.

The part most lists leave out

A command needs something to act on. This is the difference between a result you keep and a result you delete, and it has nothing to do with the command itself.

Point /xray at a face and you get a skull. Point it at a hiking boot and you get leather over foam over a rubber sole, laces and eyelets reading as structure through the shell — which is the whole point of the effect. The command did not change. The subject did.

Every example image in this catalog was cast, not just generated: the subject was chosen so the command would have something to reveal, take apart, or rebuild. That is the part a scraped word list cannot give you.

How to use one

Write the command, then name a subject. That is the whole technique.

Add constraints in plain words if the first result is not what you meant: a material, a light, an angle, a mood. You are not fighting a parser, so nothing you add is invalid syntax.

• /xray a hiking boot
• /blueprint a small house, cutaway, with dimensions
• /miniature a coastal town inside an open suitcase

When it does nothing

If the model replies with text instead of an image, it read your line as a question rather than a request for a picture. Say what you want made.

If you get an image that ignores the style, the word was not dense enough for that model. Describe the look in a short sentence instead and keep the subject the same.

If you are editing an existing image rather than generating a new one, say so explicitly. The same word behaves differently when there is already a picture in the conversation.