# NovelAI
NovelAI [Diffusion](https://www.edgepedia.org/diffusion) is a family of anime-focused text-to-image models served by subscription, built by the company NovelAI (Anlatan). The family began in October 2022 as a fine-tune of [Stable Diffusion](https://www.edgepedia.org/stable-diffusion) trained on Danbooru-tagged images, moved through an SDXL-based third version, and became a line of fully original architectures with V4 and V5. This article covers the model family only; the company, its founders, and its text-generation product are separate subjects.
| Fact | Value |
|---|---|
| First release | October 2022, an anime fine-tune of Stable Diffusion[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2) |
| V1 training data | ~5.3 million images (~6TB) with detailed text tagging, on a base trained on ~2 billion LAION images[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2) |
| V1 compute | ~1.6GB model, about three months on nodes of 8x A100 80GB GPUs[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2) |
| V3 architecture | Based on Stability AI's SDXL with NovelAI additions[2](https://journal.novelai.net/introducing-novelai-diffusion-anime-v3-6d00d1c118c3/) |
| V4 | First family of completely original models, no public base model[3](https://journal.novelai.net/release-novelai-anime-diffusion-v4-curated-preview-en-ca4b0b11e671/) |
| V5 (2026) | Twice the size of V4.5, trained on 268,000 B200 GPU-hours[4](https://docs.novelai.net/en/image/models/) |
| Weight release | Anime V1 (Curated/Full), V2, and Furry Beta weights published under CreativeML Open RAIL-M plus CC BY-NC-SA 4.0[4](https://docs.novelai.net/en/image/models/) |
| October 2022 leak | V1 weights leaked within days of release and are credited by a community history with seeding the SD 1.5-era anime finetune ecosystem[5](https://aniworldai.org/en/novelai/history) |
## What NovelAI's image models are
NovelAI Diffusion is a subscription image-generation service whose models are trained specifically for anime-style output. The company describes the current flagship, V5 Full, as a completely original image generation model, the result of building its own architecture and improving it over V4.5, twice the size of V4.5, and trained in-house on 268,000 B200 GPU-hours.[4](https://docs.novelai.net/en/image/models/) The family's history splits at V4: everything from V1 through V3 was a fine-tune of a public base model, while V4 onward were trained from the ground up.[3](https://journal.novelai.net/release-novelai-anime-diffusion-v4-curated-preview-en-ca4b0b11e671/)
The models are used through NovelAI's subscription tiers rather than as a standalone download for recent versions. V5 is available to all subscribing tiers, though it is the first generation with a usage limit on free generations for Opus subscribers; all earlier models remain unlimited for Opus users.[4](https://docs.novelai.net/en/image/models/)
## Versions and architecture as published
**V1 (October 2022).** The original NovelAI Diffusion was fine-tuned from Stable Diffusion, which the company notes was trained on about two billion images from the LAION dataset (~150TB). The fine-tuning dataset consisted of about 5.3 million images (~6TB) with very detailed text tagging data.[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2) The resulting model is small, roughly 1.6GB, and produces images without referring to outside data.[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2) Training ran for about three months on compute nodes with 8x A100 80GB SXM4 cards linked via NVSwitch with 1TB of system RAM.[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2) A distinctive technical choice: NovelAI trained against the penultimate layer's hidden states rather than the final layer, finding this let the model make better use of the dense information in tag-based prompts and better disentangle concepts like colors.[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2)
**V2.** The second anime model raised the training resolution from 512x768 to 1024x1024, making 832x1216 the default portrait resolution without requiring SMEA sampling, and raised the maximum free-generation resolution for Opus subscribers to 1024x1024.[6](https://blog.novelai.net/introducing-novelai-diffusion-anime-v2-dc668ebca34b)
**V3.** The third anime model was based on [Stability AI](https://www.edgepedia.org/stability-ai)'s SDXL with what the company calls its own secret sauce on top; NovelAI reports it renders dark scenes much more easily than stock SDXL. Training ran on the company's H100 GPU cluster, named Shoggy, hosted by [CoreWeave](https://www.edgepedia.org/coreweave).[2](https://journal.novelai.net/introducing-novelai-diffusion-anime-v3-6d00d1c118c3/) The company describes V3 as offering much higher coherence and knowledge than its older models, and notes that weights placed closer to the start of the prompt carry more emphasis.[4](https://docs.novelai.net/en/image/models/)
**V4 and V4.5.** V4 is NovelAI's first family of completely original image generation models, trained from the ground up without relying on a public base model like Stable Diffusion. It added first-class support for prompting images with multiple characters, natural language understanding, and better English text rendering, and it interprets prompts differently from V3.[3](https://journal.novelai.net/release-novelai-anime-diffusion-v4-curated-preview-en-ca4b0b11e671/) V4.5 replaced the Flux VAE used in V4.0 with a custom VAE better suited to the content the models generate; the company considered calling it V5 but reserved that name for later plans.[7](https://journal.novelai.net/novelai-diffusion-v4-5-curated-is-here-9e5ba1f628f3/) The V4 Full dataset was updated by one month relative to V4 Curated.[4](https://docs.novelai.net/en/image/models/)
**V5 (2026).** V5 Full is the current flagship, described as twice the size of V4.5 and trained on 268,000 B200 GPU-hours.[4](https://docs.novelai.net/en/image/models/) Vendor-documented additions include sharper natural language understanding, multi-language prompting including Japanese, alpha transparency support, reworked character positioning with support for up to 22 distinct characters, improved English/Japanese/Chinese text rendering, and single-image fully paneled comic generation.[4](https://docs.novelai.net/en/image/models/) The product page frames V5 as a step change from combined improvements in prompt understanding, VAE, and composition rather than any single feature.[8](https://novelai.net/v5)
## The October 2022 leak and its ecosystem legacy
Within days of the October 2022 release, the V1 weights leaked. A community history of the model line states that the entire SD 1.5-era anime ecosystem grew out of the leak: [Anything V3](https://www.edgepedia.org/anything-v3), hundreds of merges, and the familiar WebUI presets trace back to the leaked weights.[5](https://aniworldai.org/en/novelai/history)
The leak's influence extended beyond derived checkpoints. NovelAI's prompting conventions, [Danbooru](https://www.edgepedia.org/danbooru) tags instead of sentences, quality tags, artist tags, and character positions, were later adopted by open anime models including Anima, Illustrious, and Pony, making the leaked model's tag syntax a de facto standard for anime image prompting.[5](https://aniworldai.org/en/novelai/history)
## By the numbers
All figures below are vendor-reported.
- **V1 dataset:** about 5.3 million images, ~6TB, with detailed text tagging; base model trained on ~2 billion LAION images, ~150TB.[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2)
- **V1 model size:** roughly 1.6GB.[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2)
- **V1 training:** about three months on 8x A100 80GB NVSwitch nodes.[1](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2)
- **V3 training hardware:** H100 cluster hosted by CoreWeave.[2](https://journal.novelai.net/introducing-novelai-diffusion-anime-v3-6d00d1c118c3/)
- **V5 training:** 268,000 B200 GPU-hours.[4](https://docs.novelai.net/en/image/models/)
## Vendor claims versus independent evaluation
NovelAI's published quality claims come from its own testing. In the company's blinded internal comparison of V4.5 Curated against V4.0, V4.5 won 92.1% of comparisons when the prompt was not visible, which the company reads as a large improvement in baseline coherence, aesthetics, and image quality, and 75.6% or higher on prompt-following tasks.[7](https://journal.novelai.net/novelai-diffusion-v4-5-curated-is-here-9e5ba1f628f3/) The company also describes V4.5 as well-behaved and easy to steer, with better backgrounds than V4.0.[4](https://docs.novelai.net/en/image/models/)
Every quality claim about the family, including the 92.1% win rate, is vendor-run; no third-party evaluation of NovelAI's models, and no measured comparison against Midjourney's niji mode or open anime finetunes, is present in the evidence. Readers should treat the win-rate figures as the company's own assessment of its successive versions, not as independent verification.
## Licensing and availability of released weights
NovelAI has published weights for some early models: the Anime V1 (Curated and Full), Anime V2, and Furry (Beta V1.3) models were released under a dual license combining the CreativeML Open RAIL-M terms and Creative Commons BY-NC-SA 4.0, a non-commercial share-alike license. The company notes these models are no longer selectable in the product.[4](https://docs.novelai.net/en/image/models/) Later models, from V3 onward, are subscription-only; the evidence contains no published weights or output-usage terms for them.
## References
1. [The Magic behind NovelAIDiffusion (October 2022, archived)](https://web.archive.org/web/20221103080716/https:/blog.novelai.net/the-magic-behind-novelaidiffusion-b4797e0d27b2)
2. [Introducing NovelAI Diffusion Anime V3](https://journal.novelai.net/introducing-novelai-diffusion-anime-v3-6d00d1c118c3/)
3. [Release — NovelAI Anime Diffusion V4 Curated Preview](https://journal.novelai.net/release-novelai-anime-diffusion-v4-curated-preview-en-ca4b0b11e671/)
4. [NovelAI Documentation — Image Models](https://docs.novelai.net/en/image/models/)
5. [NovelAI (nai-diffusion): the history of anime models from the leaked 'anime' to V5](https://aniworldai.org/en/novelai/history)
6. [Introducing NovelAI Diffusion Anime V2 (NovelAI blog)](https://blog.novelai.net/introducing-novelai-diffusion-anime-v2-dc668ebca34b)
7. [NovelAI Diffusion V4.5 Curated is Here!](https://journal.novelai.net/novelai-diffusion-v4-5-curated-is-here-9e5ba1f628f3/)
8. [NovelAI Diffusion V5 product page](https://novelai.net/v5)
---
*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › Model families and named models › Image generation models*
*Initially written Sep 17, 2026 · Edited: Oct 11, 2026; Oct 15, 2026; Oct 25, 2026*
*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*
License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license