NovelAI

Where companion apps center on a relationship and roleplay platforms center on characters, NovelAI centers on creative output: it is a tool for producing prose and pictures, with the human firmly in the editor's seat. Its audience is correspondingly specific — AI-assisted novelists who want a model to draft and continue their stories, anime and illustration creators who use NovelAI Diffusion for art, and a notably large Japanese user base where the platform has deep penetration. On the text side, NovelAI runs its own in-house model, Erato, which is a deliberate strategic choice: by training its own model rather than renting an external one, the platform is largely insulated from third-party model supply-chain changes that periodically disrupt competitors. On the image side, NAI Diffusion is a heavily customized derivative of Stable Diffusion, fine-tuned for anime and illustration styles, with a custom tokenizer that handles Japanese and English prompts and a simplified interface that hides much of Stable Diffusion's parameter complexity. The current NAI Diffusion V4.5 brings better hands, more natural poses, and finer expressions over the previous V3. Commercially, NovelAI is subscription-led rather than free: a limited free trial exists for evaluation (with restricted output and queueing), but sustained use means a paid tier. Content moderation is moderate — NSFW output is available but gated behind a setting rather than on by default — which positions NovelAI as more permissive than mainstream chat apps but more controlled than the fully unfiltered roleplay tools. It runs on the web only; there is no native mobile app, though the browser client works on phones and tablets.

ck taste, while Tablet, Scroll, and Opus progressively raise your image budget (story writing is unlimited on every paid tier). Most newcomers who intend to stay land on Scroll, but you can start lower and upgrade once you understand your usage. To try story creation, open the Storyteller interface and click New Story. Type an opening — a hundred to three hundred words is a good amount to establish the scene and characters — then press Enter or click Generate, and the AI continues the next passage. As your story grows, use the Memory panel to store worldbuilding and key facts so the model keeps them in mind over the long haul. The rhythm is collaborative: you write, the AI extends, you edit and steer, and you generate again at the points where you want the model's help rather than on every sentence. To try image generation, switch to the Image tab and pick a model — NAI Diffusion V4.5 is the latest and the recommended anime-style default, while V3 remains available for users who prefer the older look. Enter a prompt describing your subject, then add unwanted elements to the Undesired Content box (terms like blurry, lowres, and bad anatomy) to steer the result away from common artifacts. Generate, inspect, and refine. Within ten minutes you will have written a few collaborative paragraphs and produced your first images.

**Generate** continues the current passage and is best used at turning points rather than line by line. **Memory** is always-on long-term storage for global setup — the world, the protagonist's background, the rules — with a ceiling of roughly two thousand tokens, so it should hold third-person, objective facts and be pruned as the story moves on. **Author's Note** injects temporary direction such as desired style, tone, or pacing — for example a bracketed instruction to keep the prose terse and the pacing fast. **Lorebook** is the more granular system: instead of being injected constantly like Memory, each entry is triggered by keywords, so a name like a side character or a place pulls in its specific lore only when relevant. The practical division is to keep durable global setup in Memory and scale detailed worldbuilding — secondary characters, locations, items, even nested entries — through the Lorebook. The image side is NAI Diffusion. A workable prompt structure is subject, then clothing and appearance, then scene and background, then mood and lighting, then style modifiers; quality tags such as masterpiece and best quality and scene tags such as detailed background and cinematic lighting are common boosters, while a standard block of negative tags in the Undesired Content box suppresses the usual failure modes. Fixing the seed makes results reproducible so you can iterate on a prompt deliberately, and testing at low resolution before committing to a high-resolution batch saves your image budget. It helps to understand how NAI Diffusion relates to plain Stable Diffusion. They share the same open-source lineage, but NovelAI is tuned and packaged for convenience: it works out of the box at a fixed monthly cost, with a simplified UI and cloud generation, where a local Stable Diffusion install is free of subscription cost but demands hardware, setup, and parameter tuning to match quality — while offering far more customization through LoRA, ControlNet, and plugins, plus full local privacy. Many creators use both: NovelAI for fast iteration and concept work, local Stable Diffusion for heavy, highly customized pieces. Each generated image consumes Anlas, NovelAI's image currency, at roughly three to five Anlas for a standard image — which is why your tier's Anlas allowance is the main thing that determines how much art you can make.

Story writing is unlimited on all three; what scales is the monthly image allowance, measured in Anlas — roughly 1,000 on Tablet, 3,000 on Scroll, and 10,000 on Opus — together with queue priority. Opus additionally unlocks custom module training, letting you train a personal style module from ten to twenty consistent reference images to reproduce a particular look or keep a series visually coherent. Choosing a tier comes down to how much you generate. For light use that is mostly writing with occasional images, Scroll is comfortable. For moderate daily image use, Scroll can get tight toward month's end, and Opus or extra Anlas purchases start to make sense. Heavy daily image generation effectively requires Opus or top-up Anlas. Several habits stretch a subscription further: offloading routine image work to a free local Stable Diffusion setup and reserving NovelAI for quick validation; lowering steps per image (dropping from around 28 to 20 steps cuts Anlas use meaningfully with little quality loss); testing prompts at low resolution first; and watching for the holiday promotions and Anlas bonuses NovelAI periodically runs. Because pricing and quotas can change, the figures above are a guide rather than a guarantee — the official novelai.net announcements are the authoritative source before you commit.

FictionLab leans toward roleplay and scene narration on its own Ophelia/Wraith models with a freemium model and looser moderation, and DreamGen targets technically-minded roleplay with its Lucid model and a famously steep, "homework-level" learning curve. NovelAI sits between them on accessibility — low technical barrier, moderate moderation, a fixed subscription — and stands apart by pairing serious writing tools with built-in image generation. Against AI Dungeon, the contrast is sharper: AI Dungeon is a game-like, turn-by-turn interactive fiction engine with no image system, whereas NovelAI is a solo writer's prose co-pilot with structured Memory, Lorebook, and Author's Note control plus Diffusion art. Pick NovelAI to sit down and draft a long chapter; pick AI Dungeon to play an uncertain, improvised adventure. The platform also carries a piece of history worth knowing: its custom image model leaked publicly in 2022, seeding a wave of community Stable Diffusion builds and, as a side effect, leaving many people with the mistaken impression that NovelAI is itself a free tool. The official platform tightened security afterward, and current models are available only through NovelAI — the frozen leak versions never receive new improvements. NovelAI is the right tool if you write long-form fiction and want an AI co-pilot with genuine narrative control, if you create anime-style illustration and value an out-of-the-box workflow, or if you work in Japanese, where the platform is especially strong. It is a weaker fit if you expect a sustained free tier (the free trial is only a taster), if you want a companion or character-chat experience rather than authoring tools, if you need a native mobile app, or if you want the maximal customization and zero marginal cost of a local Stable Diffusion rig. For its core audience — writers and anime-art creators who want power without self-hosting — NovelAI remains one of the most established and self-reliant options in the space.