Explore fal's Collection of The Best Image Generation APIs

How do I get started with fal's image generation API?

Every image model on fal shares the same SDK pattern. Swapping Nano Banana 2 for FLUX.2 [dev] or GPT Image 2 is a one-line endpoint change, with the auth flow and queue behavior unchanged.

Install the client

bash
npm install --save @fal-ai/client

Set your API key

bash
export FAL_KEY="YOUR_API_KEY"

Call a model

js
import { fal } from "@fal-ai/client";

const result = await fal.subscribe("fal-ai/nano-banana-2", {
  input: { prompt: "A photorealistic Tokyo cafe at golden hour" }
});

What AI image models handle text inside images well?

GPT Image 2, Nano Banana 2, Ideogram V3, and Recraft V3 and V4 all treat typography as a primary capability.

• GPT Image 2 renders character-accurate text across Latin scripts and CJK, fitting multilingual marketing and editorial work.
• Ideogram V3 was built specifically for posters and logos, with typography as a first-class output.
• Recraft V3 and V4 output vector formats alongside raster, useful when a wordmark has to scale from favicon to billboard.

Which AI image models nail photorealism?

The FLUX family covers most photoreal production work on fal, with a few partner models filling specific niches.

• FLUX 1.1 [pro] ultra delivers up to 2K resolution tuned for print-scale work where surface detail needs to survive close inspection.
• FLUX.1 Kontext [pro] is built around prompt following with photorealistic rendering, useful when the same scene structure repeats across a series.
• GPT Image 2 reasons about lighting and material properties, which holds up in scenes with mixed reflectivity and complex shadow geometry.
• Imagen 4 is Google's flagship image model on fal, another option when scene fidelity is the priority.
• Recraft V4 was developed with designers for brand systems where lighting and material rendering need to match across hundreds of assets.

What if I need to generate a lot of images quickly?

For high-volume work on fal, the Turbo and distilled models generate in roughly 1,2 seconds.

• FLUX.2 [dev] Turbo is fal's own distilled model, with one of the lower per-image rates in the catalog.
• FLUX.1 [schnell] runs in 1 to 4 inference steps and prices per megapixel, favorable for batches of small images.
• Z-Image Turbo is a 6B parameter model purpose-built for fast generation, useful when speed and quality headroom both factor into the call.
• fast-sdxl is the right call when you already have SDXL prompts and LoRAs dialed in and don't want to redo that work.

You can generate low-resolution drafts on a Turbo model to scout the prompt space, then run only the keepers through a higher-fidelity model at full resolution.

What does image generation cost on fal?

Image generation on fal is priced per output. Some models charge per megapixel of image area, others charge per image.

As a worked example, a team generating 1,000 marketing images per month at 1K resolution pays roughly:

• $8/month on FLUX.2 [dev] Turbo
• $80/month on Nano Banana 2
• $150/month on Nano Banana Pro