More than four images
What it does
The assistant can make several pictures for you, for example one per tip of a presentation, or a set of icons in one style. One reply shows at most four images. If you ask for more, the reply makes four, tells you which ones are still missing and makes them when you answer continue. It is meant to name the missing pictures, not to claim that they exist; check the reply against the pictures you see.
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
Say how many pictures you want and what each one shows. Below are two prompts you can copy and adapt.
Six pictures, the rest on "continue"
Tested on: qwen3.8-27b, gemma-4-31b-it
Not recommended for: mistral-small-4
Send the first prompt. The reply makes four pictures and says which two are still missing. Then send the second prompt as a new message to get the other two.
A set of four matching pictures
Make 4 separate pictures in one matching flat style for my website, one icon per picture: a phone, an envelope, a map pin and a clock.
Tested on: qwen3.8-27b, gemma-4-31b-it, mistral-small-4
Asks for exactly four separate pictures in one style, so all of them fit into one reply.
Each prompt was run on the three Myra-hosted models, ten runs each. A run counts as passed when the assistant really made the number of pictures that was asked for (for the six pictures: four in the first reply and two more after "continue"). Whether the pictures look the way you wanted, and the wording of the reply that names the missing pictures, are not part of the check; read the reply and look at the pictures yourself. 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 |
|---|---|---|---|---|
| Six pictures, the rest on "continue" | 4 pictures in the first reply, 2 more after "continue" | 9/10 | 10/10 | 5/10 |
| A set of four matching pictures | 4 separate icons in one flat style | 10/10 | 10/10 | 9/10 |
Tips
- Ask for "separate pictures" and say "one picture per tip" or "one icon per picture". Otherwise the assistant may draw several icons into one image, so that you get fewer pictures than you asked for.
- Name the style once ("in one matching flat style") so that the pictures belong together.
- When you want more than four, ask for them in one message and answer continue when the reply says that some are missing. You do not need to repeat the description.
- If you need only up to four pictures, ask for exactly that number; all of them arrive in one reply.
Limits
- At most four images per reply. The limit is shared by everything that shows media in a reply: generated and edited images, cut-outs, video clips, charts from the code interpreter, audio clips and connector media. If a chart takes one place, three pictures are left.
- Once the limit is reached, the image tools are not offered again in that reply. The next message you send starts a new reply with room for four more.
- The assistant tells you how many pictures it made and which ones are missing. It does not say that missing pictures exist or are shown.
The full rules are in Image generation and in the per-reply media limit.
When something goes wrong
- You asked for six pictures and got four. This is the limit of four per reply. Answer continue to get the missing ones.
- You got fewer pictures than you asked for, although the reply is not full. The assistant may have drawn several motifs into one image. Ask again for "separate pictures, one per motif".
- The reply says a picture is not made because the limit is reached. The reply already shows four images or clips. Send a new message, such as continue, to start a new reply.
- A chart or a connector image took a place. Fewer pictures fit into that reply. The assistant says how many are missing; answer continue.
- The assistant says the image service is busy. Nothing was generated for that picture. Try again in a minute.
Related
- Image generation — the limit of four images per reply and the other image rules
- Inference API reference — the per-reply media limit
- Create and edit images
- Use cases