Skip to content

Attach an image in chat

User

You can attach not only documents to a chat, but images as well — a screenshot of an error message, a photo of a sketch, a section of a diagram. The model sees the image and answers based on it.

Three things behave differently from what you would expect of file attachments. They are further down and are the actual reason for this page: HEIC is not offered, not every model can see images, and an image only counts for the same turn.

  1. Attach — in the input area, click the attach icon (paper clip). There is a quicker way via the clipboard: click into the input area and press Ctrl+V (Cmd+V on a Mac) — a copied screenshot is attached straight away, without going through the file dialog.
  2. Pick an image — the file dialog shows the supported image formats. The image appears as a preview in the input area. When pasting, this step falls away; the preview appears right there.
  3. Write your question — say what the AI should do (“What does this error message say?”, “Turn this table into Markdown”). Without a question it stays unclear what you are after.
  4. Send — with Enter or the send button.
ItemValue
Formats.png, .jpg / .jpeg, .gif, .webp, .bmp
Not supported.heic
Maximum size10 MB per file, measured on the original before downscaling
Downscalingautomatic, in the browser: longest edge 1568 px, JPEG quality 0.85

.heic is the default format of the iPhone camera — and CompanyWizard deliberately does not offer it. The reason: the browser cannot decode HEIC. Offering a format that fails after the upload is worse than one the file dialog never lists in the first place.

How to get there anyway:

  • Switch permanently on the iPhone: Settings → Camera → Formats → “Most Compatible”. New photos are then JPEG.
  • Convert a single image: pick JPEG or PNG when sharing or exporting.

Images are scaled down, never up. A small screenshot stays exactly as sharp as it was — nothing is upsampled. Only images whose longest edge exceeds 1568 px are reduced.

Two side effects that are not bugs:

  • Transparent areas turn white, because the conversion to JPEG cannot carry transparency.
  • Very fine text in a large screenshot can become harder to read after downscaling. In that case crop the relevant section instead of attaching the whole screen — it improves the answer anyway.

Image capability is a property of the model, not of CompanyWizard. Models that can process images are marked with the “Vision” badge in the model picker.

If an image is attached and the selected model cannot read images, a notice appears right away:

The selected model cannot read images. Pick a model with image support, or remove the image.

The notice also appears if you attach the image first and switch to a model without image support afterwards. If you send anyway, the turn is refused — and the image is removed with it. So after switching models you have to attach it again.

CompanyWizard stopping here instead of carrying on is intentional: an answer that quietly skips the image looks like a model that looked and found nothing worth mentioning. A visible refusal is the more honest answer.

Managed AI: nothing to configure here. Image turns never end up on a tier that cannot see.

This is the point that surprises people most often:

The image data is not stored in the conversation history. All the history holds is a line such as [Bild: errormessage.png].

For you this means: a follow-up question in the next turn no longer refers to the image. All the model sees there is the file name, not the content.

Practical consequence — ask everything about an image in one message:

  • ❌ First “What do you see in the image?”, then in the next turn “And what does the third line say?”
  • ✅ “Describe the error message and give me the full text of the third line.”

If you do need a follow-up later: simply attach the image again.

Image attachments are an administration decision and are off by default. If the switch is off in your installation, an image is discarded — even if you were able to attach it. This applies to both routes alike: Ctrl+V does not bypass the switch, pasted images go through the same check as ones picked via the paper clip.

The reason is data protection, not thrift: the PII protection inspects text and cannot inspect an image. Talk to your administrators if you need image attachments. → PII switch for image attachments