NovelAI: what it costs, what it makes, and what it never states

What it generates

One subscription covers three outputs: still images, text — storytelling and a Text Adventure mode with a documented chat format, not a companion chatbot — and text-to-speech voices. No video, no public developer API.

The image models are anime with a furry mode; the one photographic dataset it documents draws landscapes without people, so we credit no realistic style. The top tier's unlimited-images promise is footnoted to single images of up to 1024×1024 at up to 28 steps, with extra limits on its Diffusion V5 model. Clip-making generators are on our video list, its anime peers on the anime and hentai list.

Real people, minors and what the policy actually says

On the pages we read on August 29, 2026 — home, terms, sign-up and the docs — there is no statement about generating identifiable real people, and none about minors. A full-text sweep of the terms, effective December 8, 2025, returned nothing for minor, child, underage, sexual, celebrity, deepfake or likeness.

An empty box here means we found no such statement, never that the tool permits the thing. The product takes uploaded images and, by its own FAQ, keeps no record of what was sent; the only moderated surface is the opt-in Explore feed.

It ticks one policy box: no vendor-run likeness program, no licensed clones. It stays because its upload feature is a generic character reference, not an undress or face-swap product; those sit on our not-listed page. The two gaps cost it two checks in the comparison grid.

Uploads, face-swap and undress

Three documented routes feed an image in: Image2Image, Vibe Transfer, and Precise Character Reference, which the docs describe as adding a picture of the character you want reproduced. That costs NovelAI the no-upload check.

We found no face-swap feature, no undress or nudify pitch and no marketing built around real people's photos. What is missing is any rule about what you may upload, and by the FAQ nothing sent in is logged or stored. Our guide explains what a reference image does.

What you get for free

The home page advertises 30 free generations at up to 1024×1024; the docs are more specific — the Paper tier gives 30 image, 50 text and 100 text-to-speech generations once, with no refreshes — while the FAQ says there are no plans for a free version.

Both are true at once: a one-off trial exists, a recurring allowance does not — enough for the free-tier check and no more. Quotas that refresh are on our free generators list.

Paid plans in practice

Three tiers, monthly, billed at 30-day intervals rather than calendar months: Tablet $10, Scroll $15, Opus $25 in US dollars. No annual plan is printed.

Inside the product the unit is Anlas. Every tier takes 20 percent off Anlas purchases; Tablet and Scroll list image generation as pay-as-you-go with 1,000 bonus Anlas a month, Opus unlimited images with 10,000, under that footnote about size and steps.

All of it is printed publicly — a check three better-ranked tools fail. At $10 it is not cheap for stills, in a list where Promptchan AI, Secrets AI and Swipey AI start at zero.

Age gate, moderation and privacy

There is no age gate worth the name: no interstitial, no document check, the 18-and-over requirement buried in the terms. Moderation covers one surface — Explore posts are reviewed manually and may be rejected at discretion — while generation is unmoderated by design.

Privacy is the strongest part of this record: the FAQ says generations are not stored or saved and that requests, uploads, edits and generations are not logged; Explore sharing is opt-in, so nothing is public by default. The terms agree — clause 6.5 forbids logging request content without explicit consent, and 6.10 says cached input images are encrypted with a client-provided key the operator cannot decrypt, then auto-deleted.

One incident is on file: on October 6, 2022 Anlatan announced an unauthorized breach of its GitHub and secondary repositories; proprietary software, source code and image-model weights leaked and were widely redistributed — company property rather than customer accounts. We found no lawsuit against or investigation of the service.

Who it suits — and who should look elsewhere

It suits an anime or furry artist who wants stills, a story engine and speech on one bill and values a service that says it keeps nothing.

Look elsewhere if you need video, a companion chat, a free quota that refreshes, or the rules on real people and minors written down. PixAI is the nearest anime-first neighbor reviewed here and also takes uploads, but it states a minors ban and makes video; Promptchan AI ticks all seven checks at a zero-dollar entry.