The Cheapest AI API for Images and Video, Honestly Compared
“Cheapest AI API” isn’t one answer, and anyone who gives you a single winner is selling something. The price depends on what you’re generating — images or video — and, just as often, the lowest sticker price isn’t actually the cheapest once you count how each provider rounds, meters, and bills.
One scope note first: this guide is about media generation — image and video APIs. If you’re after the cheapest language model API, that’s a different market (DeepSeek, Llama, GPT-mini and friends, priced per token) and out of scope here. And this is a cost comparison, not a quality ranking — if you care about which model looks best, our best AI video generators roundup covers that. For images and video, here are the real per-task winners, taken from each provider’s own pricing page on July 22, 2026 — plus the four hidden costs that can make a “cheap” API the expensive choice.
The cheapest AI API, by task
There’s no single cheapest AI API — it splits by task, and the sticker price is only half the story. For images, the cheapest credible option is Flux Schnell at about $0.003 per image (on Replicate and fal). For video, fal’s LTX Video at $0.02 per clip is the low end, with Google’s Veo 3.1 Lite at $0.05/second the cheapest of the premium models. But before you pick on price alone, check how the provider meters — per megapixel, per second, with minimums and retries — because that’s where a lower headline rate quietly becomes a higher bill.
| Cheapest by task (July 2026) | Option | Price | Source |
|---|---|---|---|
| Image generation | Flux Schnell (Replicate / fal) | $0.003 / image | first-party |
| Image (Google) | Imagen 4 Fast | $0.02 / image | first-party |
| Video generation | fal LTX Video | $0.02 / video | first-party |
| Video (Google) | Veo 3.1 Lite | $0.05 / second (720p) | first-party |
All prices are July 22, 2026 snapshots from each provider’s own pricing page and change often — one model in this comparison was delisted within two days of our checks — so verify before you commit.
Cheapest image generation API
For raw price per image, one model is hard to beat: Flux Schnell. Replicate lists it at $3 per 1,000 output images — $0.003 each — and fal prices the same model at $0.003 per megapixel, which works out to roughly the same for a standard 1024×1024 image. That’s the floor for credible, API-accessible image generation right now.
Replicate’s pricing, captured July 22, 2026 — Flux Schnell at $3 per 1,000 images ($0.003 each).
Everything else is a step up from there. Google’s cheapest, Imagen 4 Fast, is $0.02 per image (Google’s pricing) — about 6.7× Flux Schnell, but from a first-party Google endpoint with the reliability that implies. Google’s Nano Banana line starts a bit higher again — the cheapest tier, Nano Banana 2 Lite, is $0.0336 per 1K image — with much stronger text and detail; we break the family down tier by tier in the Nano Banana price breakdown.
| Image API | Price per image | Notes |
|---|---|---|
| Flux Schnell (Replicate / fal) | $0.003 | Cheapest credible option; fal meters per megapixel |
| Imagen 4 Fast (Google) | $0.02 | First-party Google |
| Nano Banana 2 Lite (Google) | $0.0336 | Google’s cheapest Nano Banana tier |
A note in the interest of honesty: if you were hoping an aggregator would undercut this, it generally won’t on images. The cheapest image model we could verify on Modellix, for instance, is Qwen-Image at $0.0207 — well above Flux Schnell’s $0.003. Aggregators earn their keep on convenience, not on beating a specialist’s floor price (more on that below).
Cheapest video generation API
Video is where “cheap” gets slippery, because providers don’t price on the same axis — some charge per clip, others per second — so compare carefully.
The lowest per-clip price we found is fal’s LTX Video at $0.02 per video (note: per video, not per second — a detail several roundups get wrong). Among the premium, audio-capable models, Google’s Veo 3.1 Lite is the cheapest at $0.05/second (720p) and $0.08 (1080p); the full Veo 3.1 lineup and rates covers the higher tiers. Ultra-low-res, per-second “flash” video models can go lower still — fractions of a cent per second at 480p — but at that resolution and length you’re comparing a different product, not an apples-to-apples clip.
fal’s LTX Video page, captured July 22, 2026 — $0.02 per video (priced per clip, not per second).
| Video API | Price | Notes |
|---|---|---|
| fal LTX Video | $0.02 / video | Cheapest per clip |
| Veo 3.1 Lite (Google) | $0.05 / second (720p) | Cheapest premium tier, with audio |
| Veo 3.1 Fast (Google) | $0.10 / second (720p) | Mid-tier, faster than Standard |
Because the units differ, do the arithmetic for your actual clip. A per-second model at $0.05 costs $0.40 for an 8-second clip; fal’s $0.02-per-video flat rate can undercut that if its length and resolution fit your need. The cheapest headline number isn’t automatically the cheapest for your clip.
Why the lowest price isn’t always the cheapest
This is the part price tables leave out. Four things routinely make a “cheaper” API cost more:
- Metering unit and rounding. A per-megapixel price (like fal’s) rounds up to the next megapixel, so a 1.1 MP image can bill as 2. A per-image price doesn’t. At scale, the unit matters as much as the number.
- Minimum charges and startup fees. Some endpoints bill a floor per request regardless of output size, which erases a low per-unit rate on small jobs.
- Failed and retried generations. If a provider charges for non-successful calls, your effective cost climbs with every retry. The best ones bill only successful requests — worth confirming before you commit.
- Monthly fees and expiring credits. A low per-image rate behind a $30/month minimum, or credits that expire in 30 days, isn’t cheap if your volume is bursty. Pay-as-you-go with no floor usually wins for irregular workloads.
The practical move: convert every option to your comparable unit — cost per finished 1024px image, or per delivered 8-second clip — and fold in your acceptance rate and expected volume. The winner on that number is rarely the one with the flashiest headline price.
Normalized that way, the field is easy to read:
| Normalized cost | Cheapest | Mid | Premium |
|---|---|---|---|
| Per 1024px image | Flux Schnell $0.003 | Imagen 4 Fast $0.02 | Nano Banana 2 Lite $0.0336 |
| Per 8-second clip | fal LTX $0.02 (flat/video) | — | Veo 3.1 Lite $0.40 ($0.05/s × 8) |
Two things jump out. On images, the cheapest and the “premium-cheap” options are an order of magnitude apart, so the choice is really quality-vs-cost, not a rounding error. On video, a flat per-clip price and a per-second price cross over at a certain length — fal’s $0.02 flat beats Veo Lite’s $0.40 for a standard clip, but for a two-second snippet the per-second model can win. Always price your actual output, not the headline.
When a one-stop API is worth it — even if it’s not the cheapest
If the lowest possible price is your only goal, go direct to the specialist: Flux Schnell on Replicate or fal for images, fal or Google for video. That’s the honest recommendation, and it’s worth saying plainly — an aggregator like Modellix is not the cheapest sticker price; fal is generally the low-cost leader.
Where a single aggregated API does earn its place is when comparison and operations cost you more than the per-image delta. Routing a dozen providers through one key with one unified bill, per-call logging of cost and latency for every generation, and the ability to run the same prompt across models to see which is actually cheapest for your use case — that’s operational leverage a single-vendor account can’t give you, and it’s the reason to use one despite paying a small premium over the rock-bottom rate. (New to calling these endpoints? Our text-to-image API guide walks through the basics.) If you’re still price-shopping specific models, the Nano Banana Pro and Veo 3.1 model pages show exactly where each lands.
FAQ
What is the cheapest AI API for image generation?
Flux Schnell, at about $0.003 per image on Replicate and fal — the lowest credible per-image rate as of July 2026. Google’s cheapest is Imagen 4 Fast at $0.02.
What is the cheapest AI API for video?
fal’s LTX Video at $0.02 per clip is the lowest per-video price; Google’s Veo 3.1 Lite ($0.05/second) is the cheapest premium, audio-capable option. Convert to your clip length before deciding.
Is the cheapest API always the best value?
No. Metering unit (per image vs per megapixel), minimum charges, billing for failed generations, and monthly fees can make a lower headline rate more expensive in practice. Compare on cost per finished output at your volume.
Is an aggregator cheaper than going direct?
Usually not on sticker price — a specialist like fal is typically lower. Aggregators are worth it for unified billing, per-call cost logging, and comparing models through one key, not for beating the floor price.
What’s the cheapest AI model API overall?
There isn’t one — it depends on whether you need image or video, your resolution, and your volume. Price each option on your comparable unit rather than trusting a single “cheapest” label.
