xAI released Grok Imagine Image 2.0 on August 7, 2026, making the model generally available as the Quality Mode on grok.com/imagine and inside the company's iOS and Android apps. The update centers on editing as a first-class capability rather than an afterthought, adding region-level tools, multi-reference inputs, and a smart-resize feature that recomposes a single image across nine aspect ratios.
The company lists its models on the Arena leaderboards under the SpaceXAI name, a branding change that follows Anthropic's compute agreement with SpaceXAI on May 6 and SpaceX's Nasdaq listing in June. xAI claims the new model ranks second worldwide in both text-to-image generation and image editing, trailing only OpenAI's gpt-image-2 in the standings published with the announcement.
What's new
- Region editing: A magic wand tool modifies only the area a user points at; segmentation selects precise regions; background removal exports subjects with transparency.
- Multi-reference editing: Up to five input images can be combined in a single generation, removing the manual compositing step previously required.
- Smart resize: One image is recomposed across nine aspect ratios — from 1:2 tall banner to 2:1 wide banner — with the model filling the new frame.
- Templates: Pre-configured workflows for product shots, professional headshots, e-commerce listings, game assets, and marketing posters package common jobs into ready-made starting points.
- Text and layout fidelity: Training emphasized planning typography and layout so dense, multi-part visuals hold together and small text renders sharply.
Developers can already test the model through the Vercel AI Gateway using the identifier xai/grok-imagine-image-2.0-preview, with resolution options of 1k or 2k and batch generation supported. Image editing is performed by passing existing images in the prompt.images parameter. API access directly from xAI is listed as "coming soon" with no date provided.
Why it matters
The Arena placement puts xAI's image line in direct competition with OpenAI and Google's Gemini entries, the two labs whose models occupy the top and surrounding positions on the same leaderboards. For users, the editing-first design reduces the round-trip between generation and external tools — background removal and region changes now happen inside the same interface. The multi-reference capability is particularly relevant for e-commerce and marketing workflows where product consistency across angles and backgrounds previously required manual stitching.
Our take
The editing toolkit feels like the real product differentiator here; the Arena rank is a useful signal but the region wand, segmentation, and multi-reference workflow address the friction that keeps professionals in Photoshop after the first generation.
