Grok Imagine Image 2.0: Betting on Editing, Not Generation

xAI has unveiled Imagine Image 2.0, positioning it as a tool for real iterative image work rather than just another beautiful generation. The new model emphasizes precise editing with features such as Magic Wand for localized edits, segmentation for zone selection, background removal with PNG export, and multi-reference editing supporting up to five images. Smart resize supports nine aspect ratios, and there are 15 workflow templates for typical tasks like photo editing, marketing assets, and game assets. The model also aims for consistency across characters and scenes, targeting video production workflows. In benchmarks, it ranks second in the Image Edit Arena with an Elo of 1439 and in Text-to-Image Arena with 1320, trailing GPT-Image-2. Access is currently via the Vercel AI Gateway API at $0.05 per generation, cheaper than Seedream 5 Pro, Nano Banano 2, and GPT Image 2, though slower than Nano Banano 2 but faster than GPT Image 2. The author notes that while Chinese models lag behind Nano Banano 2, this affordable and high-quality model is likely to become a baseline for many, even if not all API features currently work. In Russia, the model is not yet available except through unofficial access to the preview on Vercel AI Gateway, but the release is imminent, with availability on various services including VEGA this week.