ByteDance unveils new AI image model to rival Google DeepMind's 'Nano Banana' (www.scmp.com)

ByteDance’s AI arm Seed has launched Seedream 4.0, a new image-generation and editing model that the company says outperforms Google DeepMind’s recently viral “Nano Banana” (Gemini 2.5 Flash Image) on its internal MagicBench tests. ByteDance claims stronger results on prompt adherence, alignment and aesthetics, and positions Seedream 4.0 as a hybrid that merges the text-to-image strengths of Seedream 3.0 with the targeted editing features of SeedEdit 3.0. The company also highlights commercial competitiveness: the tool is offered at the same rate as Seedream 3.0—about $30 per 1,000 generations—despite the added editing capabilities. For the AI/ML community this signals intensified competition in the image-editing frontier, where consistency and accurate adherence to prompts have been persistent challenges. However, ByteDance’s performance claims are based on proprietary benchmarks and were not accompanied by a technical report or third-party validation, so independent evaluation is still needed to confirm parity with Gemini 2.5 Flash Image. If validated, combining robust text-to-image synthesis with fine-grained editing in a single model could accelerate creative workflows, push rapid productization of editing tools, and raise the bar for benchmark design and transparency across major AI vendors.