Images and files

Vresk makes pictures inside the conversation — ask for one in plain words and it arrives in the reply, ready to refine, re-frame, or drop into a deck. This page covers making and editing pictures, attaching your own files, and how to tell which engines can actually look at an image.

Asking for a picture

There is no image mode to switch on. When a message reads as a request for a picture, Vresk paints one — and a follow-up about that picture is understood as being about the picture, not as a new topic.

What you say	What happens
“Draw me a fox in a raincoat”	A new picture, generated in the reply.
“A phone wallpaper of a misty forest”	A new picture in the shape it is for: tall for a wallpaper, story or poster, wide for a banner or header, square for a logo or avatar. Anything else is square.
“Make the sky darker”	An edit of the picture already in the conversation. The edit is a new image; the original stays.
“More like this, but wider”	A variation. Vresk re-uses the exact starting point of the earlier picture, so you get the same scene re-framed — wider, taller, or square — not an unrelated second picture.

If it answered in words instead

Short, direct phrasing reads most reliably as an image request — “a picture of…” or “a logo of…” is always understood. Long, mixed asks may be answered as text; ask again with the picture part on its own.

Pictures count toward your allowance

Making a picture draws on your allowance, the same as everything else you do: your weekly allowance, shown on Your limits, or in a guest session, the allowance for the day.

Editing part of a picture

You can point at the part of a generated picture you want changed, instead of describing where it is.

Select the area
On a computer, drag a rectangle over the picture. On a phone, tap one of nine areas — top left, centre, lower right, and so on.
Say what should change there
An edit bar appears naming the area you picked. Type the change — “make this part a sunset” — and press Enter.
Get a new picture
The edit arrives as a new image in the conversation. The original is kept, so you can compare or go back.

A guide, not a mask

The selection guides the edit rather than masking it, so areas nearby can shift a little. And an edit always comes back square, whatever the shape of the picture it started from — to change the shape, ask for a variation instead (“wider”, “taller”).

Edit controls appear on pictures Vresk generated — never on photos you upload. That is deliberate: your own images are yours, and Vresk does not offer to repaint them. What it does with uploads is described in your data.

Putting a picture in a deck

A generated picture carries a “put this picture in a deck” control. Vresk looks through the decks in the current chat and lists the ones with an open picture slot — you pick the deck and the slide, and the picture is placed there. Nothing is filed anywhere you did not choose.

If no deck in the chat has room for a picture, Vresk says so and suggests asking for a deck first. More on decks and their image slots in decks and documents.

Attaching your own files

The paperclip button in the message box attaches files, and dropping files onto the message box does the same. You can attach up to four files per message, at up to 4 MB each. Vresk accepts:

Images — JPEG, PNG and WebP.
PDF and Word (.docx) documents.
Plain text, Markdown and CSV files (a CSV is read as plain text).

A photo that is over the size limit is shrunk on your device to fit before it uploads. An oversized document cannot be shrunk without losing content, so it is refused with a message instead. And if a document holds no readable text — a scanned PDF, for instance — Vresk warns you at upload time, before you build a question on something the model cannot quote.

What happens to an uploaded image

Every image you attach is re-encoded on our servers before it is stored, which removes camera metadata such as location. Details in your data.

Which engines can see images

Some engines can read images and some cannot, and Vresk marks the difference rather than hiding it. In the engine picker, engines that can see carry an eye mark, and searching for vision or photo surfaces exactly the ones that can. The list changes as engines are added, so the picker — not this page — is the place to check.

If you pin an engine that cannot see onto a conversation that contains images, Vresk refuses by name: “That model can’t read the images in this conversation — pick a model that can read photos.” It never quietly swaps in a different engine behind the label. More on picking a model in choosing an engine.