What’s new
- Editing is now first-class: a magic wand for targeted region edits, segmentation for precise selection, and background removal with transparent export are all new this generation, not bolted on top of text-to-image.
- Multi-image reference input: editing accepts up to 5 source images at once, for combining subjects, transferring styles or composing scenes.
- Noticeably stronger text and layout: xAI says it plans typography and layout the way a designer would, following instructions closely enough to keep small text legible in dense compositions.
- Smart Resize: automatically recomposes a shot across different aspect ratios instead of just cropping it.
What it’s good for
- E-commerce and product photography: recoloring, cutting out backgrounds, producing transparent assets.
- Avatar and character work: professional headshots, game icons, character sprites.
- Precise local edits where you want to change one area without disturbing the rest of the image.
Prompting tips
- Point at exactly what should change: combine the magic wand or segmentation tools with a prompt that states the target region and the desired outcome.
- Write layout requests like a design brief: it understands design language — typography, white space, composition — so describe it the way you'd brief a designer.
- Ask directly for background removal: it's one of the model's headline features, so just say "remove the background, export with transparency".
- Use the count parameter for batches: up to 10 images per request, useful when you want several options at once.
- Pick a quality tier instead of relying on "auto": the API's quality parameter (low/medium/auto) resolves "auto" to low for generation but medium for editing — set it explicitly for final output.
Remove the background from this product photo and export as a transparent PNG.
Keep the product's position, angle and color exactly as they are, with clean edges
and no leftover background color fringing.
Known limits
- Multi-image editing tops out at 5 reference images per request.
- The "auto" quality setting resolves differently depending on the task — low by default for generation, medium for editing — which is easy to trip over if you don't set it explicitly.
- Only two resolution tiers are available, 1K and 2K, with nothing higher.
Using it in Nomi
On the generation canvas, add a image card, pick “Grok Imagine Image 2.0” in the model picker and choose a mode (Text to image, Image to image). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
