NovelAI
Long-memory AI fiction writer with lorebooks and no content filters
Overview
NovelAI is a subscription platform for long-form fiction and roleplay, built around its own Kayra and Xialong language models, persistent story memory via lorebooks, and a strict no-training, no-public-gallery privacy stance that appeals to writers who don't want corporate content filters.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 3 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Unlimited text gen (Kayra model)
- Image gen up to 1024x1024
- 1,000 Anlas per month
- 10,000 Anlas per month
- 2,048-token story memory
- Text-to-speech
- Xialong model (GLM-4.6 fine-tune)
- 8,192-token story memory
- Highest-priority generation
- Early access to experimental features
Use cases
What you can produce with NovelAI
- Kayra 13B and Xialong (GLM-4.6 fine-tune) in-house writing models
- Lorebook system for persistent worldbuilding memory
- Up to 8,192-token story memory (Opus tier)
- Built-in anime-style image generator (NovelAI Diffusion)
- Text-to-speech (Scroll tier and up)
- No public gallery, no default training on user text
NovelAI earns its niche by refusing to be a general-purpose chatbot: the lorebook system tracks characters, locations, and plot threads across genuinely long stories, and the house models — Kayra, now joined by the GLM-4.6-based Xialong for Opus subscribers — are tuned on fiction rather than instruction-following. No public gallery, no training on user text by default, and a permissive content policy make it the default choice for roleplay and erotica writers that mainstream tools won't touch.
The trade-offs are real: no developer API, a UI that reviewers repeatedly call clunky for newcomers, and prose that can loop or repeat itself over long sessions despite the memory system. At $25/month for Opus — the tier that actually unlocks Xialong — it's pricier than general writing assistants for a narrower use case. Fits genre fiction and roleplay writers who value privacy over polish, not productivity or brand-copy use cases.
What people say
NovelAI's core users are fiction and roleplay writers, and its 2026 update reinforces that focus: Xialong, released in March, is fine-tuned from ZhipuAI's GLM-4.6 on NovelAI's own creative dataset and is gated to Opus ($25/month) subscribers only — a deliberate way to keep the best model behind the top tier rather than spreading it across all three plans (Tablet $10, Scroll $15, Opus $25).
Reviewers converge on the same strengths: the lorebook system is the reason people stick around, letting a story remember character traits, locations, and plot threads well past what a stock chat window can hold, and the no-public-gallery, no-default-training privacy stance matters to a userbase writing content — including NSFW roleplay — that most mainstream AI providers refuse to touch.
The recurring complaints are consistent across review sites: prose repetition and looped plot points over long sessions despite the memory features, a UI that reviewers call clunky with a real prompt-crafting learning curve, and cost creeping up fast for anyone who wants both the top text model and heavy image generation — Anlas, the image-gen credit, runs out quickly on the two cheaper tiers. Image quality is also flagged as behind dedicated tools like Midjourney, which matters less for NovelAI's anime-focused niche but is a comparison reviewers keep making anyway.
There's no developer API and no meaningful free tier, so NovelAI competes purely as a paid, writer-first product rather than a platform others build on top of. That's consistent with its whole positioning: it isn't chasing the general AI-assistant market, it's chasing writers who got frustrated with mainstream chatbot guardrails and want a tool that remembers their story and doesn't moderate their fiction.
The official NovelAI GitHub organization backs up the technical claims: it hosts dozens of research and infrastructure repositories, including tooling for aspect-ratio bucketing in image-model training and a custom BPE tokenizer for English and Japanese text, evidence the team runs real in-house model development rather than wrapping a third-party API. That in-house model stack is also why NovelAI can offer a permissive content policy other providers won't touch — they aren't subject to an upstream model provider's usage terms.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including NovelAI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with NovelAI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to NovelAI directly →
Spotted something out of date? Suggest an update →
More in Productivity
