Google is Finalizing ‘Nano Banana 2 Flash’ For Expected December Release

Google is finalizing ‘Nano Banana 2 Flash’, a cost-optimized successor to its powerful image editing model Gemini 3 Pro Image. Internally codenamed ‘Mayo’, the new system targets a December release to secure enterprise adoption against rising Chinese competitors.

Leaks reveal a bifurcated strategy, pairing the efficiency-focused ‘Mayo’ with a high-end ‘Pro’ variant codenamed ‘Ketchup’. The rapid iteration aims to capitalize on the ‘Nano Banana’ brand, which has quickly become an industry standard for instruction-following precision in AI image generation.

Internal Codenames and the ‘Flash’ Pivot

Driving this rapid iteration is a market that has shifted from pure generation to precise, instruction-based editing. According to reports of the upcoming model, the new architecture bifurcates the product line into distinct tiers.

‘Mayo’ reportedly will serve as the high-volume inference model, optimized for speed and cost efficiency. A separate ‘Pro’ variant, internally dubbed ‘Ketchup’, will targetsmaximum fidelity for production-grade workflows.

Representing a significant acceleration in Google’s development cycle, the update arrives just two months after the Gemini 2.5 Flash Image release, which established the company’s dominance in the editing arena.

Early adopters found that the previous architecture solved critical issues in spatial consistency.

Google’s move to a ‘Flash’ variant specifically addresses the friction of enterprise scaling. While the current model’s pricing per image is competitive, high-volume applications require lower operational costs to remain viable.

The ‘Banana’ Standard: From Meme to Industry Benchmark

Originally a viral codename on the LMArena leaderboard, the ‘Nano Banana’ branding has transcended Google’s marketing to become a generic industry term for high-precision instruction following.

Competitors are now relying on this architecture to train their own systems. In a notable development, researchers leveraged Apple’s Pico-Banana-400K dataset to build a robust foundation for future editing models.

Spending approximately $100,000, the team generated nearly 400,000 examples using Google’s infrastructure.

Relying on a rival’s technology for data generation highlights the current gap in open-source capabilities. Google’s model has effectively become the gold standard for maintaining visual consistency during complex edits.

Such widespread recognition validates Google’s strategy of integrating these tools directly into consumer apps. By lowering the barrier to entry, the company aims to capture the casual creator market alongside professionals.

Nicole Brichtova, a Product Lead at Google DeepMind, described the impact of making these advanced workflows accessible, sating in October:

“We’re putting capabilities that used to require specialized tools into the hands of everyday creators, and it’s been inspiring to see the explosion of creativity this has sparked.”

Price Wars and Multimodal Consolidation

Google’s push for a ‘Flash’ tier is a direct response to aggressive pricing from Chinese rivals. The Seedream 4.0 launch by ByteDance introduced a model that undercuts Google by approximately 28 percent.

Market analysis shows the Chinese giant offering generation at roughly $0.028 per image, compared to Google’s $0.039. To prevent enterprise customers from defecting to these cheaper alternatives, the ‘Mayo’ model will likely need to match or beat this price point.

Beyond pricing, the competitive field is shifting toward unified “generation plus editing” workflows. ElevenLabs’ recent Studio expansion exemplifies this trend, merging video models from OpenAI and Google into a single timeline.

The ElevenLabs Team emphasized the strategic value of this aggregation in their announcement:

“Unifies the most advanced AI models with our industry-leading voice, sound, and music tools”

By centralizing these tools, competitors are challenging the fragmented workflows that Google currently dominates. However, the sheer popularity of the ‘Nano Banana’ models provides a strong defensive moat.