Google DeepMind Launches Nano Banana 2.1 Image Model
Nano Banana 2.1, the latest image generation model from Google DeepMind, is now available on Replicate. It is specifically optimized to provide an excellent balance between cost and performance.
The model incorporates significant advancements in visual design, making the resulting images look more natural and realistic. Key improvements include enhanced mask-based editing and better subject consistency.
These features make high-quality, reliable image creation more accessible and affordable for developers and users.
The most important thing about Nano Banana 2.1 isn’t simply another improvement in image quality — it’s the quality-to-price ratio. Google has roughly halved the image generation cost compared with Nano Banana 2. On Replicate, a generated image currently costs $0.0336 at 1K, $0.0504 at 2K, and $0.1134 at 4K. That’s roughly 29 1K images or 19 2K images for $1. This isn’t merely a cheaper model, either. Google reports significant improvements in prompt adherence, visual realism, text rendering, infographic design, and — particularly useful for production workflows — character and object consistency across edits. Nano Banana 2.1 accepts up to 14 reference images, with documented consistency for up to four characters and fidelity for up to ten objects. It supports 1K, 2K and 4K generation, image and mask-based editing, multi-turn editing, and grounding with Google Search and Image Search. Perhaps the most interesting claim is that Google’s own evaluations put Nano Banana 2.1 ahead of not only Nano Banana 2, but even the more expensive Nano Banana Pro in several tasks. With Thinking enabled, multi-character consistency scores 1106 Elo versus 1011 for Pro, while overall text-to-image preference is 1050 versus 935. Infographic factuality shows an even larger improvement: 0.521 versus 0.179 for Nano Banana 2 and 0.265 for Nano Banana Pro. These are Google’s own evaluations, however, so they shouldn’t be treated as an independent definitive ranking. For developers, Nano Banana 2.1 is already available through the Gemini API and through Replicate. Replicate offers particularly straightforward per-image pricing. Its early published examples show roughly 10 seconds for a 1K generation and 18–19 seconds for 2K, although these examples shouldn’t be interpreted as guaranteed API latency. If Google’s claimed improvements hold up in real-world use, Nano Banana 2.1 looks less like an incremental update and more like Google’s new default workhorse for image generation and editing: near-Pro or, in some tasks, better-than-Pro performance at mass-production pricing.
Why it matters
- —It improves the cost-to-performance ratio for image generation, making it more accessible.
- —The inclusion of mask-based editing and subject consistency adds professional-grade control to the output.
- —It represents a new, optimized tool for developers building visual AI applications.
Key facts
- Model Name: Nano Banana 2.1
- Developer: Google DeepMind
- Platform: Available on Replicate
- Optimization: Balances price and performance
- Key Features: Visual design, mask-based editing, subject consistency
The full text is in the original source. Here we provide a brief summary and key facts.