nano-banana and the batch economics
Google's current Flash Image model is Gemini 3.1 Flash Image — Nano Banana 2, ID gemini-3.1-flash-image. Same family: Lite (gemini-3.1-flash-lite-image) and Nano Banana Pro (gemini-3-pro-image). The whole family lives behind the Gemini API. The call is generate_content with an IMAGE modality, one image per call.
The reason it matters: it changed the math.
The pricing
As of September 2026, official Gemini API:
- Gemini 3.1 Flash Image (Nano Banana 2,
gemini-3.1-flash-image): $0.045 at 0.5K, $0.067 at 1K, $0.101 at 2K, $0.151 at 4K. - Gemini 3.1 Flash Lite Image (
gemini-3.1-flash-lite-image): $0.0336 at 1K only. - Gemini 3 Pro Image (Nano Banana Pro,
gemini-3-pro-image): $0.134/image at 1K/2K, $0.24 at 4K. - Gemini 2.5 Flash Image (original nano-banana): $0.039/image, retires October 2026. The preview ID is already gone.
Compare against the rest:
| Model | Per-image cost |
|---|---|
| gpt-image-2 quality=low (square) | $0.006 |
| flux-2-klein | from $0.014 |
| gemini-3.1-flash-lite-image (1K) | $0.0336 |
| recraftv4_1 raster | $0.035 |
| flux-2-pro | from $0.03/MP |
| flux-pro-1.1 (previous-gen, still live) | $0.04 |
| gemini-3.1-flash-image 0.5K / 1K | $0.045 / $0.067 |
| gpt-image-2 quality=medium (square) | $0.053 |
| Ideogram 4.0 Default | $0.06 |
| recraftv4_1 vector | $0.08 |
| Ideogram 4.0 Quality | $0.10 |
| gemini-3.1-flash-image 2K / 4K | $0.101 / $0.151 |
| gemini-3-pro-image 1K/2K / 4K | $0.134 / $0.24 |
| gpt-image-2 quality=high (square) | $0.211 |
The economic threshold
Here's the math that actually matters. Suppose your product needs one shipping-ready image per user request. Three strategies:
-
One-shot premium: send one prompt to gpt-image-2 quality=high ($0.211) or Nano Banana Pro at 4K ($0.24). Hit rate (image is usable as-is): maybe 30-40%. So real cost per shipping-ready image: $0.50-0.80, and you spent a minute regenerating.
-
Batch-and-filter: send the same prompt to nano banana 2 at 1K 10 times. Cost: $0.067 × 10 = $0.67. Hit rate per individual image is lower (~20%), but you have 10 candidates. Probability that AT LEAST one is shipping-ready is 1 - (0.8 ^ 10) ≈ 89%. Real cost per shipping-ready image: $0.67 with 89% confidence, $1.34 with ~99% confidence on the second batch. Lite at $0.0336 (1K) is the same-family cheaper row when the quality ceiling can move.
-
Self-hosted Flux: rent a GPU, run FLUX.2. Marginal cost per image drops hard at volume. Worth it above ~50,000 images/month. klein (from $0.014) is the API version of that floor.
The batch-and-filter strategy is the 2026 default for most consumer products. nano banana 2 made it economically obvious. At $0.067/image for 1K, you can generate 25 candidates for about $1.68, then use a filtering step (cheap LLM call with vision, or a CLIP-similarity score against a reference) to pick the best.
Multi-turn editing
The other reason nano-banana is interesting: it's built for multi-turn conversational editing in the Gemini API. You generate an image, then say "make the background blue and add a coffee cup on the left," and it edits in-place using world knowledge from Gemini's text base. Most other image models don't do this — they regenerate from scratch, losing identity.
This makes nano banana 2 the right pick for any flow where the user is iterating: "show me an image of X" → "now change Y" → "now zoom in on Z." Stay on the same family for the follow-up turns (nano banana 2 or Lite) so identity holds.
When to NOT use nano-banana
- Highest-end photorealism for hero shots.
flux-2-prostill beats it. nano banana 2 is good, Flux is best. - Crisp text rendering at scale. Use Ideogram 4.0. nano banana 2 got better, but Ideogram is still the specialist.
- Instruction-heavy composition. gpt-image-2 or
gemini-3-pro-image. - Native vector. Recraft V4.1 (
recraftv4_1). - Subjective magazine-shot taste. Midjourney is still ahead here for now, and it's UI-only.
For everything else — batch generation, multi-turn editing, "I need 50 variants under $5" — nano banana 2 is the default.
What this changes about your harness
Before nano-banana, the default image-gen pattern was "spend a lot, generate one, hope it's good." After nano banana 2, the default is "spend a little, generate many, filter." That changes the shape of the pipeline. Lesson 03 of this chapter walks the full pattern (brief → prompt → batch → filter → upscale → format). For now, just internalize: the cheap banana is a row in a table. Read the vendor row, then pick the count.