NovelAI
The subscription-based AI creation platform launched by Anlatan in the United States puts animation-style image generation and long story writing in the same subscription, focusing on creative freedom and privacy protection.
In-depth Report
-
NovelAI is a subscription-based AI creation platform launched by the American company Anlatan Inc., which puts "AI story writing" and "anime-style image generation" in the same subscription, focusing on creative freedom and privacy protection. It started as a text continuation tool in 2021. It became an instant hit after it went online for image generation in October 2022. It also caused huge waves in the AI painting circle due to the model and source code leaks that year. Nowadays, the platform uses the self-developed Kayra-XL language model (128K context) and Anime/Diffusion V4 image model. The animation quality is in the first echelon among commercial platforms. It is suitable for serious long-form creators and animation illustration enthusiasts, but the interface is English-oriented, the learning curve is steep, the free credit is very small, and the price is high, which is something ordinary novices need to weigh.
-
The developer of NovelAI is Anlatan Inc., a small studio registered in Delaware, USA. The official website is novelai.net. The product was first launched in 2021 and was positioned as an AI text continuation and story generation tool for writers. At that time, it adopted a subscription model and emphasized no content censorship and privacy protection. What really made it out of the circle was the image generation function launched on October 3, 2022. This function was based on the newly open-sourced Stable Diffusion at that time, and then used a massive animation library mainly based on Danbooru for targeted training. The quality of the generated two-dimensional images was far superior to ordinary models at the time. But then in early October 2022, NovelAI’s model weights and part of the source code were leaked by hackers and went viral on the Internet. This leak objectively exposed "high-quality animation models" directly to everyone, and spawned a large number of locally deployed animation painting models and the entire AI painting localization wave. It had a particularly profound impact on the Chinese AI painting community. It is a double-edged sword for NovelAI itself: its popularity has skyrocketed, but it has also lost its model-exclusive moat. In the following years, Anlatan continued to invest in self-developed models, gradually getting rid of its dependence on open source bases, and formed today's two self-developed model lines of text and images.
-
Text is NovelAI’s specialty. The core is AI-assisted story writing, relying on the self-developed Kayra-XL language model for plot continuation. The 128K ultra-long context window makes it less likely to "amnesia" when writing long novels, and can remember the characters and setting details in the entire book. Users can set the story style, character and world view, and the AI will write according to your preferences. In addition to regular continuations, there is also a Text Adventure interactive novel mode. The gameplay is similar to early text adventure games, except that the plot branches are generated by AI in real time, making the game more immersive. In conjunction with the Lorebook worldview knowledge base, users can save character files, locations, and world rules, and automatically inject context into appropriate paragraphs based on keyword triggers, which is very helpful for the consistency of long-form stories. In addition, it also provides emotional text-to-speech (TTS), suitable for audio books or interactive reading. The Scroll plan has 100 times per month, and Opus has more. The image aspect has been praised the most in recent years. Using Anime/Diffusion V4 diffusion model, the details and consistency of drawing two-dimensional characters are very stable. Its interface is designed around the creative experience, supporting tabbed prompts to finely control character appearance, clothing, expressions and composition. It can also use multi-character prompts to depict different characters in the same picture. There are also many practical tools: Image2Image can change details while retaining the original composition, Enhance can improve the clarity, Vibe Transfer can directly transfer the style of an image to a new one, Inpaint can use the brush to partially repair or change, and there are also post-processing tools such as background removal, line drawing, coloring, expression change, and debris removal. The most commonly mentioned point in user feedback is that "pictures are produced in almost seconds, and you don't have to spend half an hour repairing your hands." Moreover, as long as there is a browser, it can be used on any device, which is very worry-free.
-
NovelAI adopts a pure subscription system and will be divided into four levels in 2026. The free Paper plan provides 50 text generation and 30 image generation (up to 1024px) per month. You will have to wait until next month to use up. The Tablet plan is $10/month, with a high degree of freedom in text and images. The maximum image size is 1024px and comes with 1000 Anlas points. The Scroll plan is $15/month and adds 100 TTS per month to the Tablet. The Opus plan of $25/month is the flagship plan, using the best models, the highest image quality, 10,000 Anlas per month, and more TTS. Anlas are points within the platform that are used to generate high-resolution images or call special functions. Users who are serious about drawing will basically go directly to Opus. The overall monetization path is clear, and users with different strengths are covered by tiered subscriptions and point consumption.
-
Positive reviews focused on the quality and ease of image generation. Some users said that the tag system is simple and that different tags can create various painting styles, which is very interesting; others praised that it can be used on any device with a browser and is very comfortable; and some said that the biggest pleasure is that it can produce pictures almost instantly without having to retouch repeatedly. Writing users value the consistency brought by long context and Lorebook, and think it is one of the few tools that can seriously support long-form creation. Negative feedback is also relatively concentrated. Novices generally feel that the interface is not friendly to novices and requires time to explore; the server is occasionally disconnected when writing long articles, which affects the experience; the free version has too little credit and is used up quickly; there is no work sharing and community exploration mechanism, and it is more "standalone"; Opus's $25/month is too expensive for ordinary users; the interface is mainly in English and is not friendly to Chinese users. Although the model can understand and generate Chinese, the quality is not as stable as English.
-
The industry generally regards NovelAI as an important node in animation AI painting. The model leak in 2022 has been discussed repeatedly. Many analysts believe that it directly promoted AI painting from the cloud to local deployment, and also indirectly spawned a number of derivative models, which has a profound impact on the entire two-dimensional AI painting ecology. Evaluation media usually affirm its unique positioning of "story + image integration" when reviewing Hengping: Character.AI is strong in conversational role-playing but does not do image generation, and the review is strict; Janitor AI focuses on free role-playing chat but does not have images and perfect writing tools; CrushOn AI and Candy AI take the AI companion route and also do not have the ability to write or produce animations. In contrast, NovelAI has almost no direct rivals at the intersection of "writing stories, illustrating pictures, and creating creative freedom", which is also its core competitiveness.
-
The biggest historical controversy comes from the source code and model leakage incident in 2022. Although it is an "accidental open source" for the industry, it is a major event at the security and asset levels for the company itself. Another long-term controversy is the training data: the image model used a large number of animation libraries represented by Danbooru in the early days, involving copyrighted works without the authorization of the artist. This is also a legal and ethical gray area faced by the entire AI painting industry. In addition, the platform's "high degree of freedom and lack of strict content censorship" is a double-edged sword, which is not only a selling point but also brings compliance and content risks. When using it, you must also pay attention to the official domain name novelai.net. There are sites with the same name or counterfeit websites on the Internet (such as novelai.fun for Chinese novels, etc., which are different products), and beware of fraud when paying.
-
NovelAI is most suitable for serious long-form story creators, enthusiasts who need high-quality controllable animation illustrations, people who value privacy and do not want their content to be scanned by the platform or used for training, writers who like to carefully build a worldview, and mature creators who need creative freedom. If you happen to be stuck on these needs, a $15 to $25/month subscription is worth it. Conversely, users who are new to AI tools (the learning curve is steep), have limited budgets (the free version is too small), and mainly want AI companion chat or light role-playing dialogue, may find it not cost-effective. There are specialized platforms that are more suitable for such needs. If you want to test the waters at a low cost first, you can use the free Paper plan to run a few pictures and write a few paragraphs to make sure the painting style and writing feel are right before deciding whether to upgrade.
-
NovelAI is a veteran player with almost no direct rivals in the cross-track field of "AI writing stories + producing animations". It has stable image quality, creative freedom, and adequate privacy. The price is a high threshold, an English-oriented interface, a small free quota, and a high price. For serious creators with clear goals, it will still be an unavoidable choice in 2026; for casual users, it is more suitable to try it with the free quota first.
User Reviews
-
KarenDavis_88—The quality of the pictures is really top-notch, and the consistency of the animation style puts it ahead of a bunch of similar ones. -
Sophia.Alvarez—I have been using it for almost a year, and the biggest feeling is that it can produce pictures in seconds. It used to take half an hour just to repair my hands with other tools. With NAI, I basically don’t need to worry about my hands. Once the labels are in place, the composition is very stable. The shortcomings are also obvious. The interface is all in English. My mother gave up immediately after reading it. Moreover, Opus is not cheap at $25 a month. -
Gabriel.PhillipsZ—The free Paper version only contains 50 texts and 30 pictures a month, and it’s gone with a shake of the hand. -
Heather.Allen_202226—Lorebook is really a savior for long novels. I wrote more than 300,000 words of settings and stuffed them all into it. The characters and place names are not broken at all. The mechanism of automatically injecting context by triggering keywords is so smart. The 128K context is really not an exaggeration. The only drawback is that the threshold for beginners to get started is a bit high. I was confused by the various samplers and parameters at first, and it took me several days to get started. -
JMendozaJr237—The Chinese generation is still not that interesting. The quality of the English prompt words is obviously more stable. When writing Chinese novels, you have to make more adjustments. -
杨彤—Vibe Transfer is a great function. You can directly transfer your favorite painting style to a new picture. Unifying the style is so trouble-free. -
AmyFisher168—For the sake of privacy, the official said that user content will not be used to train the model and will be encrypted and stored. It feels much more secure to write something personal. -
NoahBarnes_Max—It’s really expensive, but it’s really unmatched when it comes to drawing two-dimensional objects. -
Elizabeth535—I would like to ask, is there a big difference between NAI4 and NAI3 pictures? I am wondering whether to go directly to Opus. -
Jose_AdamsX4—It’s really uncomfortable to be disconnected from time to time in the middle of writing a long article. I’m stuck when my ideas are going well. I hope the server stability can be improved. -
姚丹建—The model leak that year directly made the local deployment of AI painting popular. Looking back now, NAI is a node that cannot be bypassed. It has lost its exclusive advantage, but its self-developed models have caught up very quickly in the past two years. -
月光_2—Without a community, you can't read other people's works. It's a bit of a stand-alone game, and communication depends entirely on outside groups. -
5flvtw511—The tag system is really addictive after you get started. Different tag combinations can create completely different painting styles, from fresh to thick paint to cel. Rolling the dice at random can also give you some new inspiration. If you are not careful, you can fish for an entire afternoon, and the drawing speed is so fast that you can't stop at all. -
ffffypfeb—Anlas points are a pain to use, as the high resolution disappears as soon as I run out, and I am basically forced to go to Opus for heavy rendering. -
AGkra—Writing + illustrating can be done with one subscription. Character.AI does not provide pictures and is very strict in censorship. Janitor does not have decent writing tools. After all, NAI is the most comprehensive. It’s just that the prompt word skills are really impressive. Those who can write and those who can’t write are completely different worlds. It is recommended that novices first look through other people’s tag combinations before getting started.