Grok Imagine Image 2.0 for posters: write a brief, not a mood board

Grok Imagine Image 2.0 is the xAI image model built for work with words in it—posters, packaging, album covers, app screens, menus, book jackets. LUVI's prompting guide for the family records its central behaviour: it follows instructions closely and plans typography and layout the way a designer would. That single trait decides how to write for it. On LUVI it runs as text-to-image and edit; on September 22, 2026 one image quoted 80 credits at the low tier, 120 at the default and 160 at 2k.

One thing to settle first: this is a separate family from the older Grok image models, not a newer version of them. Two of its rules invert, and carrying the old habits across is the most common way to waste a render here.

What does "plans the layout" mean for a prompt?

It means the artefact type does a lot of work for free. Name it—"concert poster", "cereal box", "app onboarding screen", "airline safety card"—and the model brings that format's whole layout convention with it. Then give it a hierarchy rather than a layout: "a large headline, a subhead half its size, and a four-line credit block at the base" is actionable. "Well-balanced typography" is not.

A concert poster for a late-night jazz quartet, printed on heavy uncoated stock. A large headline reads "BLUE ROOM", a subhead half its size reads "FRIDAYS AT ELEVEN", and a four-line credit block sits along the base. Mid-century modernist typography in warm cream on deep navy, generous negative space, three elements only, sharp small print, ink slightly absorbed into the paper.

The clause order that suits it runs subject, then composition, then typography, then style or medium, then lighting, then camera, then material, then palette. Style or medium is the single highest-leverage phrase in the whole prompt: one noun there overrides a paragraph of adjectives.

How do I get the words themselves right?

Put the copy in quotation marks as a literal—the headline reads "BLUE ROOM"—rather than describing it. Capitalizing the literal is the most-cited way to make the model treat it as content to render rather than description, with one side effect worth knowing: it tends to set the type in caps. When mixed case matters, quote it without capitals.

Every distinct string is another thing that has to be spelled, positioned and kept legible at once. There is no published density threshold, so write the text the piece actually needs rather than pre-emptively simplifying—but know that each addition competes for the same attention.

How long should the prompt be?

Don't target a length. This model publishes no length guidance, and its own example prompts run from four to twenty-four words. Stop when every element you would reject the image for is specified, and not one clause later; anything you would accept either way, leave out.

There is a related discipline that matters more than length. Cut competing constraints rather than adding adjectives: a "wide establishing shot" of a "macro texture detail" averages into mud. If two constraints cannot both be true, keep the one you would reject the image over.

And specify the mechanism instead of the adjective wherever the result actually matters. "Single hard key light from camera-left at 45 degrees, deep falloff, no fill" is an instruction; "dramatic lighting" is a judgement the model will answer with its own taste.

Which habits from the older Grok image models must I drop?

Two, and they are the ones that look most transferable.

• References are positional, in prose. Write "the product from image 1 onto the surface in image 2, matching the lighting and colour grade of image 2". The angle-bracket tokens the earlier Grok image models use are not the syntax here.
• The 30-to-80-word sweet spot does not apply. That figure belongs to the older family. Here, length follows the specification.

What does carry across is the negation rule: there is no negative prompt and no seed, so every exclusion becomes the presence of its opposite. "No blemishes" becomes "smooth, clear skin". "Not cluttered" becomes "generous negative space, three elements only". "No text" becomes "a clean unlettered surface". A negation clause is still tokens describing the forbidden thing.

What do the parameters do, and what do they cost?

Four, and one of them has a trap in it.

• quality is low or medium. The low tier is cheaper and roughly eight times faster; medium is the default and the model's best output. Measured: 80 credits against 120 at 1k.
• resolution is 1k (1024×1024) or 2k (2048×2048), and 2k added 40 credits—160 against 120 at the medium tier.
• num_images runs 1 to 4, and each image is billed separately.
• aspect_ratio offers thirteen values, including phone-screen shapes like 9:19.5 and 20:9 alongside the usual 16:9, 9:16 and 2:3.

The trap is casing. These enum values are exact: 2k, not 2K. A value in the wrong case is dropped rather than corrected, and the model quietly falls back to its default—so you ask for 2k, pay for 1k, and only notice when you open the file.

How we checked

• Models: Grok Imagine Image 2.0 Text-to-Image, with the edit model read from the catalog.
• Settings: one image at 2:3, quoted at low and medium quality and at 1k and 2k.
• Date and what was read: September 22, 2026—the model's parameter contract and live credit quotes in LUVI, LUVI's own prompting guide for the family, and xAI's Imagine API documentation. The example prompt above was checked against the guide before printing.
• Results: 80 credits at low quality and 1k, 120 at medium and 1k, 160 at medium and 2k.
• Drawbacks: the behaviour described here comes from LUVI's prompting guide rather than a published ablation by the vendor, and no images were generated for this post, so it makes no claim about output quality.

In LUVI

Open Image 2.0 in the Workspace, name the artefact in the first clause, and let the feather button check the prompt against LUVI's guide for this family before you spend anything—it is what catches a stray negation and an angle-bracket token that belongs to the other Grok models. Set resolution and aspect_ratio in the controls, in the exact case the list shows.

New accounts start with 200 credits, and verifying a phone and a card adds up to 1,500 more. You can create a LUVI account and see what a written brief gets you.

Sources

• Imagine API overview — xAI, official developer documentation. Accessed September 22, 2026.