Google launches Nano Banana 2.1: image prices cut nearly in half, 2K/4K quality boosted

songguoxiansen · x · 2026-10-08

Google released Nano Banana 2.1 (built on Gemini 3.6 Flash) on October 6, after appearing as "beluga" in Flow a day earlier. It's live in the Gemini app, AI Studio, Flow, Google Ads and more.

Key upgrades

• Better 1K/2K/4K quality, text rendering, and infographic layout
• Mask editing: paint or circle a region to edit only that part
• Subject consistency: up to 14 reference images, 4 consistent people and 10 objects
• Fixed seam artifacts in ultra-wide ratios (1:4, 8:1) at 2K/4K
• Web-grounded generation: checks Google Search before rendering real subjects
• Three thinking levels (low/medium/high) trading speed for instruction following

Pricing: gemini-nano-banana-2.1, paid tier only. Image output roughly halved: $0.0336 (1K), $0.0504 (2K), $0.0756 (4K); input rose from $0.50 to $1.50 per million tokens, output from $3 to $7.50.

Benchmarks: Google's internal human preference eval scores 1050 for thinking 2.1 vs 990 for 2.0 and 935 for Pro; on third-party Arena it ranks 5th in text-to-image and 6th in image editing, still behind GPT Image 2.5.

Known limits: blurry small text at 1K, imperfect character consistency, partial instruction execution in mask edits, occasional left/right confusion, limited 3D reasoning and factual accuracy.

Related event: Nano Banana 2.1 Rolls Out with Cheaper Pricing and Sharper Text, Community Tests Flood Social Media(6 posts)→

Original post →

More from Multimodal

• Prompt template: psychedelic sunset serenade silhouettes with flowing outlines — LudovicCreator · 2026-10-08
• World of Warcraft looks far better in Ghibli style, says share — djcows · 2026-10-08
• Direct link to the Pixel Promo Video skill behind the VibeCafe pixel-art promo — yihui_indie · 2026-10-08
• Salt++ post-training lifts 4-step streaming audio-video generation by up to 57% — Xingtong Ge · 2026-10-08
• Grok Imagine Video 1.5 Lite hits OpenRouter at 2 cents per second, a quarter of the full model — tetsuoai · 2026-10-08
• Duotone Portrait Prompt: High-Contrast Neon-Noir With Midjourney v8.2's --raw and --stylize 150 — michaelrabone · 2026-10-08

Multimodal channel →