fal.ai pricing in 2026: image, video and GPU costs

How does fal.ai pricing work?

fal sells prepaid credits that pay for its Model APIs. Each model has its own unit: per image, per megapixel, per second of video, per 1,000 characters or per request. Models without a fixed output price fall back to per-second billing on the GPU that ran them. Figures come from fal’s pricing page and docs. Prices checked November 10, 2026.

• You pay only for successful outputs. Server errors (HTTP 500 and above) and time waiting in the queue are free.
• Purchased credits expire 365 days after purchase. Free credits and coupons expire after 1 week to 1 year, depending on the grant.
• New accounts can run 2 requests at once. The limit rises with your purchases, up to 40.
• When your balance falls below the account lock threshold, fal rejects API requests until you add credits.
• A pricing API returns the unit and unit price for any endpoint, so you can quote costs before you run them.

How much do fal.ai image models cost?

Image models bill per image or per megapixel. fal normalizes its table to 1 megapixel, and higher resolutions cost proportionally more. Edit endpoints for the Nano Banana models cost the same as generation.

How much do fal.ai video and audio models cost?

Video bills per second of output, or per token on Seedance. fal assumes a 5-second 720p clip for its comparisons. Audio models bill per 1,000 characters or per second.

What do fal.ai GPUs cost for your own models?

If you deploy your own app on fal Serverless, fal bills GPUs per second while your runners are alive. The pricing page shows a list price and a lower "as low as" price for each GPU.

What does a real fal.ai job cost?

Model choice moves the bill by 50 times or more for the same count of outputs. These examples use list prices at 1 megapixel per image.

How is fal.ai billed through Locus Pro?

Locus Pro offers fal as one endpoint, fal/generate, for image, video and audio models, paid from one prepaid credit balance that also covers other providers, with no fal account or expiring credits. You pick a fal model ID, and Locus can estimate the cost before you generate, though some estimates are unavailable. Catalog checked November 10, 2026.

A job needs enough balance for a verified upper bound. Locus then charges actual usage and returns the rest, and confirmed failures release the hold. A COMPLETED status can still hold a failed job, so check the result. Set max_credits to cap any job.

• AI image generation on Locus Pro
• Browse generation endpoints

Which fal option should you choose?

Use your own fal account and API key if you generate media every day. Jobs at that volume can cost less, and you get higher concurrency as you buy credits, the pricing API and custom GPU deployments. Remember that purchased credits expire after a year.

Locus Pro fits an agent that makes images or clips now and then, next to search, models and data APIs, where a credit pack might expire unused. Locus credits you buy do not expire, each job can carry a price ceiling, and on Enterprise you can bill your own customers for what their agents generate.

• Drafts and thumbnails: FLUX.1 [schnell] at $0.003 per megapixel.
• Edits and product shots: Nano Banana at $0.0398 per image.
• Short video on a budget: MiniMax H3 Max Turbo at $0.015 a second.
• Text to speech: compare fal’s ElevenLabs prices with ElevenLabs direct, which is lower for Multilingual v2.

• ElevenLabs API pricing