fal.ai

Generative media platform with one API for the best image and video models (Flux 2, Nano Banana Pro, Ideogram), wired into NicAI to make sales-deck visuals on demand.

Table of Contents

fal.ai is a generative media platform: one API to run the best image, video and audio models (Flux 2, Nano Banana Pro, Ideogram 4, Recraft, and over a thousand more) on serverless GPUs. For NicAI it is the missing image backend. Instead of waiting on a designer or trawling stock libraries, a prompt becomes a slide visual in seconds: hero backgrounds, section dividers, abstract concept art, icons and illustrations for sales presentations. You pay per generation, with no infrastructure to run.

fal.ai
Generative media platform for developers - fal.ai
https://fal.ai

What it offers

  • Over 1,000 production models for image, video, audio and 3D behind one API. The live gallery is at fal.ai/models.
  • Fast inference on serverless GPUs (H100, H200, B200), so there is no infra to manage.
  • Pay per use, with Python and JavaScript SDKs and a queue API for batches.
  • LoRA fine-tuning to lock a house style across a deck.
  • Used by 1.5M+ developers.

For business people

The point is simple: type a prompt, get an on-brand slide visual in seconds. Use it for atmosphere and illustration in decks: backgrounds, dividers, concept imagery, simple icons. It removes the stock-image subscription and the designer bottleneck for everything that is not a real product shot or a chart.

  • Cost: pennies per image on pay-per-use, versus a stock plan or design time.
  • Which model for what: Flux 2 for general high-quality images, Nano Banana Pro for edits and tight prompt adherence, Ideogram 4 when you need readable text inside the image (a title on a hero graphic), Recraft for clean vector and brand-style icons.
  • Judgement and limits: do not generate fake logos, fake charts, or anything that misrepresents data in front of a client. Keep one consistent style so a deck looks like a deck, not a collage. AI images still produce artifacts (hands, text, edges), so review every image before it ships.

For technical people

Install the client, set the key, call an endpoint. Store the key in ~/ai/_shared/.env, never in a deck or repo.

import fal_client  # pip install fal-client ; export FAL_KEY=...

result = fal_client.subscribe(
    "fal-ai/flux-2-flex",
    arguments={
        "prompt": "clean abstract data-network background, deep navy, lots of negative space",
        "image_size": "landscape_16_9",
    },
)
image_url = result["images"][0]["url"]   # hosted image URL
  • Endpoints follow fal-ai/<model>. The exact slug and its parameters live on each model's page at fal.ai/models. subscribe() blocks until the job is done and returns hosted image URLs; for fan-out use the queue API.
  • Auth is a single FAL_KEY from fal.ai/dashboard/keys.
  • Hosting for decks: the result is a temporary fal URL. For a reveal.js deck, download it and push it through the existing ~/ai/_shared/imagekit.py helper so the image sits on the ImageKit CDN, then request the exact size with transform_url(path, width=1920, format="webp", quality=80).
  • Style lock: LoRA fast training fine-tunes a model on a handful of brand images so a whole deck shares one look.

Wire it into NicAI

A small skill closes the loop: input a prompt, a slug, and an aspect ratio; generate with fal; host on ImageKit; return the CDN URL to drop into the slide. Keep the deck-visual prompts in a style reference file (palette, mood, "no text unless asked") so output stays on-brand across slides. Batch a section's images through the queue API in one run.

Image models worth knowing

Model Endpoint Best for
Flux 2 fal-ai/flux-2-flex General high-quality images and backgrounds
Nano Banana Pro fal-ai/nano-banana-pro Edits, compositing, strong prompt adherence (Google)
Ideogram 4 fal-ai/ideogram-4 Readable text inside the image
Krea 2 fal-ai/krea-2/turbo Aesthetic, photoreal looks

Recraft is the pick for vector and brand-style icons but the gallery moves fast, so treat fal.ai/models as the live source of truth.

Pricing

Pay per generation, priced per model. Distilled and turbo models cost a fraction of a cent to a few cents each; premium image models cost more, and video far more. Check fal.ai/pricing for current rates before wiring a loop that fans out hundreds of images.

My take

This is the image backend NicAI was missing. For sales decks it removes the stock-image tax and the designer wait for anything that is atmosphere or illustration. The discipline is taste: one consistent style, no fake data, and a human review before a client sees it. Pair Ideogram or Recraft for on-brand graphics with text, and Flux 2 for hero imagery.

Further reading