Novel AI
An uncensored, privacy-first AI writing suite with fiction-trained prose models, keyword-triggered Lorebook memory, and best-in-class anime image generation, all in one subscription.
The most capable prose model and memory system for fiction, wrapped in a steep learning curve and a two-currency price tag.
Should you use NovelAI?
Fiction writers, worldbuilders and roleplayers who want an uncensored prose model with real long-story memory and do not mind a busy editor.
You want a chat-style assistant or photorealistic art. NovelAI is a writing cockpit tuned for anime, not a general tool.
Text is unlimited from Tablet at $10/mo , but heavy image work pushes you to Opus at $25 for effectively unlimited generations.
The Anlas currency confuses people , and there is no annual discount. Big models like Xialong are Opus-only.
What is NovelAI?
NovelAI will not chat with you. It is a writing surface, not an assistant: you write a line, the model continues in your voice, and you steer it with a Lorebook, an author's note, memory and a wall of sliders for randomness and repetition. Built by Anlatan as a successor to AI Dungeon, it runs its own fiction-trained models (Kayra, the Llama 3 based Erato, and the Opus-only Xialong) with no content filter on text and client-side encryption so the company cannot read your stories. That combination, freedom plus privacy, is the whole pitch.
Bundled into the same subscription is a separate product: anime image generation on NovelAI's in-house Diffusion models, which produce some of the most accurate anime art of any generator. It is web only, aimed squarely at novelists, roleplayers and anime artists, and it makes no attempt at photorealism or a beginner-friendly on-ramp. If you want a general chatbot or realistic images, this is the wrong tool.
Key features & how they perform
Each feature rated from hands-on testing and aggregated review sentiment.
Fiction-trained prose models
Erato and Kayra build a named character, setting and mood from one sentence. The output reads like prose worth editing, not deleting.
Lorebook memory
Keyword-triggered entries feed character and world facts to the model only when relevant, keeping long stories consistent without burning context.
Anime image generation
NAI Diffusion turns a plain description into clean anime art on the first try, with tag prompting, multiple character prompts and Vibe Transfer.
On-canvas editing tools
Inpainting, image-to-image, upscaling, background removal and seed reuse sit above the canvas, so a generation becomes a starting point.
No text content filter
The text side runs without moderation within platform terms, which is exactly why writers left AI Dungeon for it.
Encryption-first privacy
Stories are encrypted client-side and the company states it does not log or train on your prompts, images or text.
Feature ratings blended from G2, Reddit, Product Hunt & documentation review patterns + hands-on testing.
One writing session, from a single line to finished anime art
A story from one prompt, a Lorebook memory test, and a first image generation.
Start a story from one line of direction
A new story opens as an almost empty editor. There is no chat box and no template, just a text field and a Send button, which is the point: you are meant to write, not converse.
I kept the input deliberately open to see how much the model would invent on its own rather than how well it follows a detailed brief.
"Write a mysterious fantasy story about a young woman who discovers a magical power she was never supposed to have."
Read what the model built from that sentence
From one line, the model produced a named protagonist, a setting with its own texture and a consistent mood across several paragraphs.
Elodie Vancroft walking through the rain-soaked courtyard of the Royal Collegium reads like something a person wrote on a good day. The vocabulary shifts to match the scene and the dialogue beats land where they should.
Build a Lorebook entry and confirm it loads
Memory across a long story is where most writing tools fall apart. NovelAI answers this with the Lorebook, so I built a test around it rather than taking the claim on trust.
I created an entry for a character named Kaelen and gave him checkable traits: a former knight who lost his left arm at the Siege of Vardholm, never removes his red scarf, speaks bluntly and distrusts nobles. His name became the activation key, and the entry showed a 40 token count.
A Lorebook entry only feeds its details to the model when the trigger word appears in the story, so before writing anything I checked the entry was registered and reachable through Quick Access.
Trigger the entry and watch it turn facts into behavior
For the real test I wrote a single line, "The tavern door opened and Kaelen walked in," and let the model continue with no other prompting.
It did more than recall the facts. Kaelen stumbles in, the empty left sleeve of his tunic flaps like a wounded bird, he catches himself with his remaining hand, and moments later he shoulders past a nobleman without apology. Every planted trait appeared, and none of it was in the sentence I wrote.
Switch to image generation on the free trial
The image side is a separate product bundled into the same account. The free trial hands you 30 generations, which is enough to judge the output before paying anything.
I wrote a natural-language prompt rather than fighting with tags, to see how far plain description would get me.
"Anime-style young woman with long silver hair and violet eyes standing on a neon-lit rooftop at night, magical blue symbols glowing around her, cinematic lighting, detailed, atmospheric."
See the output and the tools attached to it
The result came back clean on the first try: sharp linework, coherent lighting and a composition that matched the description without any tag-wrangling. Anime styling is clearly where this model spends its strength.
The image arrives inside a full editing surface, with upscaling, variations, seed reuse and export sitting right above the canvas, so one generation becomes a starting point rather than a finished product.
NovelAI pricing
Figures taken from the official NovelAI pricing page.
| Plan | Price | What's included |
|---|---|---|
| Free Trial | $0 | 30 image generations up to 1024x1024 · limited text generation · no card required |
| Tablet | $10 /mo | Unlimited text & TTS · 1,000 Anlas/mo · pay-as-you-go image gen · 20% Anlas discount |
| Scroll Popular | $15 /mo | Everything in Tablet · larger memory context for storytelling · 1,000 Anlas/mo |
| Opus | $25 /mo | Unlimited standard images · 10,000 Anlas/mo · largest token context · Xialong model access |
Pros & cons
Specific conclusions from testing and real user reviews, not generic filler.
✅ Pros
- Erato and Kayra build a full scene from one sentence of input
- Lorebook wove all four planted Kaelen traits into behavior unprompted
- Unlimited text generation on every paid tier, including $10 Tablet
- NAI Diffusion produced usable anime art on the first prompt
- No content filter on text within platform terms
- Client-side encryption; company states it does not train on your work
- 30-generation free trial needs no card details
⛔ Cons
- Editor feels like a flight cockpit for the first hour
- Anlas dual-currency system regularly confuses new subscribers
- No annual discount on any tier
- Xialong, the strongest text model, is locked to Opus at $25
- No photorealism; output is anime and fantasy only
- Erato prose can read flat for literary fiction, needing rewrites
Synthesized from real reviews on G2, Reddit, Product Hunt, Steam forums & official documentation · paraphrased, not quoted
NovelAI scorecard
Rated against what an AI writing and anime-art suite is actually built to do.
| Dimension | Verdict | Score |
|---|---|---|
| Prose quality & coherence Readability and rhythm of generated text | Excellent |
8.5
|
| Lorebook & memory Consistency across a long story | Excellent |
8.5
|
| Creative control & customization Presets, sliders, author's note, modules | Good |
8.0
|
| Anime image quality Accuracy and detail of generations | Excellent |
8.5
|
| Content freedom Lack of filtering on text output | Good |
8.0
|
| Privacy & encryption Client-side encryption, no training on data | Excellent |
8.5
|
| Editing & post-generation tools Inpaint, img2img, upscale, seed reuse | Good |
7.5
|
| Ease of use Learning curve of the editor | Average |
5.5
|
| Pricing clarity & value Anlas system, tiers, annual options | Average |
6.0
|
| Versatility beyond anime Style range and photorealism | Average |
5.5
|
| Scrutool Score Equal-weight average of all 10 dimensions |
7.5
|
What users say about NovelAI
The themes reviewers raise most often, by share of analysed reviews.
% = share of analysed reviews mentioning each theme (G2, Reddit, Product Hunt, Steam forums)
The serious writer's tool, once you get past the cockpit
NovelAI is the rare AI writing product that treats you like a writer rather than a chatbot user. The prose models build a coherent scene from a single line, and the Lorebook does something most tools only claim to: it held every trait I planted about a character and turned static description into behavior three sentences later. The bundled anime generator is genuinely best-in-class and produced usable art on the first prompt. The costs are just as real. The editor overwhelms newcomers for the first hour, the Anlas currency confuses people, there is no annual discount, and the strongest text model sits behind the $25 Opus tier. Fiction writers, worldbuilders and roleplayers who want freedom and long-story consistency will find this worth the setup. Anyone after a conversational assistant, photorealistic images or a gentle on-ramp should look elsewhere.
Discussion
Join the discussion and share your perspective.