Nano Banana 2.1 blind-tested against 3 rivals: text nails it, no model draws 8:17 on a clock
Hot-Wrangler-3692 · reddit · 2026-10-24
The author ran one blind test across Nano Banana 2.1, GPT Image 2.5 Sunburst, Seedream v5.0 Pro, and Grok Imagine Image 2.0 with the same cluttered-workshop prompt (first generation only, defaults, closest native 4:3 per endpoint) via atlas cloud.
Key findings
- Surprise: all four rendered the 'PARTS RETURN' sign text correctly — but none drew a clock showing 8:17
- Counting: only A (Nano Banana 2.1 Developer) and B (GPT Image 2.5 Sunburst) produced exactly three wrenches; C gave two, D (Grok Imagine) gave four despite an explicit 'no extras' instruction
- Style: A followed constraints best but looked processed; B felt most like a real photo; C nailed the lived-in clutter but lost a wrench; D looked convincing yet over-arranged
Cost twist: actual per-request costs were A $0.0175, B $0.00438, C $0.072, D $0.06 — listed starting prices didn't match real charges. The author cautions it's a single sample per model with non-identical endpoint settings.
More from Multimodal
- FLUX.3 Klein (5.6B) tested across 20 anime styles: still behind MOSS-Image, Krea 2 and Qwen-Image 2.1 — SkirtSwimming8950 · 2026-10-24
- 2-step DMD2 LoRA brings Qwen Image 2.1 to real-time generation on ZeroGPU — multimodalart · 2026-10-24
- Dreamina's $0.29 first month comes with 1,575 credits — but how far do they go? — eyishazyer · 2026-10-24
- AI-generated Tesla Roadster street footage is getting hard to tell from real video — The_Wist · 2026-10-24
- Kimi builds Singapore Changi Airport in Blender end-to-end, with native Tripo plugin support — crystalsssup · 2026-10-24
- Artefact AI Film Festival opens entries: every production phase must use a generative AI tool — lmoroney · 2026-10-24