Grok Image Generation Guide: Prompts, Edits, Aurora Tips

Grok, the assistant from xAI, generates images from inside a conversation rather than a dedicated generator panel. You ask for a picture the way you would ask a question, Grok returns one or more images, and you keep talking to change them. That conversational loop is the main thing to learn. This guide covers how to phrase the first request, how to steer revisions, what the underlying Aurora model is good and bad at, and where the limits sit for free and paid plans.

Where image generation lives

Image generation is available on grok.com, in the Grok tab of the X app and in the standalone Grok apps. In each case you type into the normal chat box. There is usually an attachment or tools button for uploading a reference image, and in 2025 xAI added a separate Grok Imagine area for images and short videos with audio. For single images in chat, the plain conversation is enough.

Which features you see, and how many images you can make per day, depends on plan and region. Free accounts have lower limits than paid tiers. Those allowances change often, so check xAI’s current help pages rather than older posts.

Writing the first request

Grok treats the request as an instruction, so a direct sentence works better than a keyword list. A dependable pattern:

1. Ask for the image explicitly. "Generate an image of" or "Make a picture of" removes any ambiguity about whether you want a description or a picture.
2. Subject with one detail. "A lighthouse keeper in a yellow oilskin coat."
3. Setting and light. "On a rocky headland at dusk, the lamp just lit, long shadows."
4. Framing. "Wide shot, low angle."
5. Medium. "Photographic, 35 mm" or "gouache illustration, limited palette". Aurora leans photographic by default, so name the medium if you want anything else.

Avoid stacking quality words. "Highly detailed, 8k, award-winning" adds nothing a modern model needs and can push the image towards a generic glossy look. Avoid brand names and real people, both for reliability and for rights reasons.

Steering revisions in chat

The real advantage of a chat interface is iteration. Instead of rewriting, reply with the change:

• "Same scene, but make it dawn instead of dusk."
• "Keep everything, move the keeper to the right third of the frame."
• "Turn this into a flat vector illustration with three colours."
• "Closer. Portrait framing, head and shoulders."

One change per message keeps the result predictable. If Grok drifts too far after several turns, paste your best version of the full prompt again as a reset. You can also upload an image and ask Grok to make a version in a different style or setting; the model uses it as a loose reference rather than a strict template, so expect the composition to change.

What Aurora does well, and what it does not

The Aurora model is strong on photographic realism, natural lighting and everyday scenes. It is reasonable with short text, though lettering should be kept to a word or two in quotation marks. Hands, reflections and crowds can still go wrong, as with every model. Its default look is realistic and slightly cinematic, so for illustration, product renders or diagrams you need to say so clearly and sometimes repeat it in the revision.

Grok is also comparatively permissive about subject matter, which means the responsibility for what you publish sits with you. Check the usage terms before using output commercially.

Settings at a glance

Grok’s chat image generation has few exposed controls; most steering happens in words. The table shows what you can influence.

Control names and availability differ between grok.com, the X app and the mobile apps, and they change with updates.

Keeping track of what worked

Because the prompt is spread across a conversation, the final image rarely has one clean prompt attached to it. When you get a result you like, ask Grok to "write out the full prompt that describes this image" and save that text with the download. It gives you a single line to reuse next time, and something to hand a client who asks how the image was made.

Try it

Generate an image of a small red fishing boat moored in a quiet harbour at dawn. Mist over the water, a single gull on the bow, pastel sky. Wide shot, eye level, boat slightly left of centre. Photographic, soft natural light.

Then reply: "Same scene as a loose watercolour with visible paper texture." Comparing the two shows how much the medium instruction changes while the composition holds.