Kimi Work can show images in conversations: keep posters and feedback together

On October 15, 2026, the Kimi Work 3.2.17 release notes added image display in conversation content. The same release also corrected PDF citation matching and highlight positioning.

In everyday use, this means that when a task displays an image in the conversation, you can look at the visual material alongside its explanation. A comment such as “the heading at the top left is too small” previously required opening the image again to find it. Putting the picture next to the explanation makes that relationship easier to check.

This is a desktop presentation update. It does not announce a new image-generation model or guarantee an image in every response. Kimi's capability guide treats image generation as a separate capability requiring the relevant creative plugin. Checking an existing poster does not mean generating a replacement.

Turn a vague opinion into feedback you can locate

Suppose you need to choose a poster for a book club. Version A has a prominent heading but small location text. Version B has more whitespace but makes registration easy to miss. Simply asking AI to pick the “best” one may produce polished reasons that are hard to verify.

A more useful task is to name the images A and B, then ask for separate checks of the event name, time, location and registration method. Each comment should identify its image and location, such as whether the location in the lower half of A is legible, rather than offering a vague suggestion to improve the layout.

Seeing the image and explanation together can help reveal a mistaken version, overlooked small text or decorative words treated as essential information. The saving is in finding what to check. A person still decides which version suits the audience.

Keep the full original image

An inline thumbnail helps with quick comparison. Small type, QR codes and cropped edges still need inspection in the original. A clear preview does not prove a poster remains readable when printed or forwarded on a phone. The outcome can be a concrete list of changes for the designer, rather than a saved sentence saying “B is better.”

First check that the computer's Kimi Work client is on 3.2.17 or later. The Work overview describes Work mode, new tasks and the @ context entry. The release notes also document dragging files into the window as attachments. This small review uses your own images and does not require a drawing plugin.

1. Open Work and start a task. Prepare two posters you are allowed to process, name their files A and B, and keep the originals. Remove unrelated personal information before using an attachment.
2. Add identifiable image attachments. Drag both pictures into the window and check the names and previews. For files already in the current workspace, you can also use @ to add context. “The image from before” is not a reliable version label.
3. Ask for specific checks. For example: “Check the event name, time, location and registration method in A and B separately. Show the corresponding images in your reply and organize the comments into version, location in the image, actual visible text and suggestion. Write ‘cannot confirm’ where something is unclear; do not fill gaps. Give feedback first, without editing the originals.” This is a task request; whether images appear inline depends on the actual client result.
4. Check each comment against the image. Confirm it refers to the right version and location. Open the original to enlarge small text and QR codes; scan a code with your own device when needed. AI saying a QR code is fine does not verify its destination.
5. Decide what to change. Turn confirmed issues into a change list, then decide whether to authorize editing a copy separately. If the reply contains only text, open the image through its file link to check it. Missing inline previews are no reason to accept unlocatable conclusions.

A useful result tells you which part of which image needs a change, why, and what remains unconfirmed. Inline images make that checking process easier.