Skip to content

Create and edit images

What it does

The assistant can create a new picture from your description and edit a picture you attached. Everything runs on Myra's own image service in the EU. It cannot edit several pictures in one go: an edit always starts from exactly one source image.

Image generation has to be switched on for your gateway, and the model must be able to use tools. Ask your administrator if the assistant only describes pictures in words.

How to ask

Describe what you want in plain words; you do not need a special format. Below are three prompts you can copy and adapt.

A new picture

Create a watercolour picture of a lighthouse on a cliff at sunset.

Tested on: qwen3.8-27b, gemma-4-31b-it, mistral-small-4

Describes what is in the picture and the style. The assistant creates a new image from the description alone.

Edit a photo you attach

Make the background plain white like a passport photo. Keep the face and the clothes as they are.

Tested on: qwen3.8-27b, gemma-4-31b-it, mistral-small-4

Attach the photo to the same message as this prompt. It names the change and what must stay the same.

Edit a photo from earlier in the chat

In the photo above, paint the little cabin blue.

Tested on: qwen3.8-27b, gemma-4-31b-it, mistral-small-4

First send your photo in a message of its own. Then send this prompt as a new message, without an attachment. The assistant shows the most recent picture of the chat, here your photo, with the buttons Use this image, Pick another and Attach new. Nothing is edited until you press Use this image (or pick another picture); then it edits that picture.

Each prompt was run on the three Myra-hosted models, ten runs each. A run counts as passed when the assistant really made the picture, or really edited the photo you attached or confirmed. Whether the result looks the way you wanted is not part of the check; that is up to you. Under each prompt you see the models it was tested on; the table shows how many of the ten runs passed:

Prompt Test case qwen3.8-27b gemma-4-31b-it mistral-small-4
A new picture a watercolour lighthouse from a description 10/10 10/10 10/10
Edit a photo you attach a portrait photo, white background, face and clothes kept 10/10 10/10 9/10
Edit a photo from earlier in the chat a photo sent in an earlier message, confirmed with Use this image, the cabin painted blue 10/10 10/10 10/10

To repeat or vary an image, say so in words. Ask for example "again with seed 1234" to get the same picture, or "the same picture, but with seed 1235" for a variation of the same idea. The seed is shown under every generated or edited image.

To control the rendering, name the settings in words: "use 40 steps" for more refinement (1 to 50), or "guidance 5.5" for how closely the image follows your description (0 to 20, up to three decimals). Both work for a new picture and for an edit. Leave them out and the image service uses its own defaults.

To edit a picture from earlier in the chat, send the photo, then ask for the change in a new message without an attachment. The assistant shows the most recent picture with Use this image; press it, or Pick another for a different one. It never edits an older picture on its own guess, and naming a picture in the prompt does not choose it.

Tips

  • Describe the change you want for an edit, and what must stay the same ("leave the rest as it is").
  • Ask for a repeat in the same message as the description. In a later message the assistant no longer sees the exact description it sent, so the image is close but not guaranteed identical.
  • Change one thing at a time (only the seed, or only the steps) to see what it does.
  • A seed also works on an edit: the same source image, seed, description and settings should give the same edit.

Limits

  • One source image per edit. With two or more pictures in your message, attach the one you want on its own.
  • An edit keeps the aspect ratio of the source image and produces at most about one megapixel.
  • You can edit PNG, JPEG and WebP pictures of at most 12 MB. A GIF cannot be edited.
  • A seed reproduces an image only together with the same description, size and settings. The steps and the guidance are stored with the image and shown beside the seed, when you asked for them; otherwise the defaults applied.
  • Steps must be a whole number from 1 to 50, guidance a number from 0 to 20 with at most three decimals. Other values are refused; they are never rounded or replaced.
  • At most four images are shown per reply. For more, see More than four images; the full rules are in Image generation.
  • If the assistant asks for exactly the same image twice within one reply, the second request is not run again and is not charged. To get a further variant, ask for something different, for example another description or seed.

When something goes wrong

  • The assistant says a value must be a whole number from 1 to 50 (or a number from 0 to 20). The steps or guidance you named are outside the allowed range. Nothing was generated. Ask again with a valid value, or leave it out.
  • The assistant says the image service is busy. Nothing was generated. Try again in a minute.
  • It says it cannot edit because there are several images. Attach the one image you want to edit in a new message.
  • The repeat looks different. The description, size or settings differ from the first time, or the image service was updated in between. Repeat the steps and guidance you used, in the same message.