6 image models
Image generation models
Image APIs bill per generated image, but a 'unit' means different things per provider — resolution, quality tier, and whether editing counts as a new generation all move the real cost. Prices here are for a standard square generation unless noted.
Pricing checked against provider documentation on . How we verify
6 of 6 models
| Model | Provider | Price | Context | Best for | Status |
|---|---|---|---|---|---|
| Imagen 4 Fast | $0.02 / img | — | Cheapest speed-first generation | stable | |
| FLUX.2 [pro] | Black Forest Labs | $0.03 / img | — | Cheapest flagship-tier image quality | stable |
| Ideogram 4.0 Turbo | Ideogram | $0.03 / img | — | Legible text inside images | stable |
| GPT Image 1.5 | OpenAI | $0.04 / img | — | Complex multi-subject composition | stable |
| Nano Banana 2 | $0.07 / img | — | Fast generation with multilingual text | stable | |
| Nano Banana Pro | $0.13 / img | — | Multi-image compositing and iterative edits | stable |
No models match those filters.
Editing beats generation
The competitive frontier moved from "make a pretty picture" to controllable editing: in-painting, character consistency across images, and text rendering that survives a crop.
Check the licence, not just the price
Commercial usage rights, indemnification, and whether outputs can be used to train differ sharply between providers. For brand work this often matters more than cost per image.
Open weights are still viable
Self-hosting a diffusion model trades per-image fees for GPU time. It wins at high volume or when you need LoRAs and custom pipelines the hosted APIs will not run.