What this table is (and isn't)
A capability comparison for all model families, based on real testing — not vendor marketing specs. It covers the models themselves; for what Nano Banana runs in production today, see the note at the end.
Model overview
- Model | Type | Resolution | Max Duration | Released | Core Strength | Key Limitation
- NB1 | Image | HD default | — | 2024 | Fastest; good Chinese; lowest per-image cost | Unreliable text rendering; conservative creativity
- NB2 | Image | 4K native | — | 2025 | Up to 14 references; virtual try-on; 4K | Financial content restricted (since Feb 2026)
- Pro | Image | 4K+ | — | 2025 | Highest instruction accuracy; 2,000+ char prompts; web | Slow; highest cost; not suitable for bulk
- Veo 3.1 | Video | 1080p | 8s/gen | Oct 2025 | Native audio; dialogue lip sync; highest quality | 8s per generation limit; per-second pricing
- Seedance 1.0 | Video | 1080p | 12s | Jun 2025 | 7 aspect ratios; fast; lowest cost | No audio; weak narrative capability
- Seedance 1.5 | Video | 1080p | 12s | Dec 2025 | Audio-visual sync; beat matching; lip sync | Beat sync less stable with slow ambient music
- Seedance 2.0 | Video | 2K native | 15s | Feb 2026 | Multi-shot narrative; best physics; 2K | Longest generation time; ~10% failure on complex scenes
Image model key differences
- Reference images: NB1 supports 1 · NB2 up to 14 · Pro multiple
- Creative flexibility: Pro > NB2 > NB1
- Generation speed: NB1 > NB2 > Pro
- Cost: NB1 (lowest) < NB2 < Pro (highest)
Video model key differences
- Audio support: Veo 3.1 native audio · Seedance 1.5 audio-visual sync · Seedance 2.0 physics audio · Seedance 1.0 no audio
- Max single duration: Seedance 2.0/1.5 (15s) > Seedance 1.0 (12s) > Veo 3.1 (8s)
- Physics realism: Seedance 2.0 > Veo 3.1 > Seedance 1.0
- Dialogue lip sync: Veo 3.1 > Seedance 1.5
How this maps to Nano Banana today
Image generation on Nano Banana runs on Nano Banana 2 Lite, billed by resolution tier: 1 credit per 1K image, 3 for 2K, and 8 for 4K — charged only on success, and 4K requires an active subscription. The standard 2 and Pro tiers above follow public information; verify current specs before committing budget.
Video generation is currently in demo mode: the workbench runs end to end with placeholder results and never charges credits. The estimator shows 25–110 credits per 5-second clip by tier (double for 10 seconds), read from the same pricing table the rest of the site uses.
Head to the case library to see real generated examples — more useful than any spec table.