Image Generation Models
NovelAI Diffusion has 6 different models you can choose from when generating images.
Each of these models will behave differently, and should be selected according to what kinds of images you want to generate.
A description of the model you are currently selecting is displayed right above the prompt box. You can click it to select another model.
Current
NovelAI Diffusion V4.5 Curated
|
|
|
V4.5 Curated is our latest image generation model. Available for all subscribing tiers.
It is a completely original image generation model, and is the result of applying all that we learned from V4.0 in practice. Compared to V4.0, we find V4.5 to be, overall, a well behaved and easy to steer model.
Just like our previous Curated models, the NovelAI Diffusion V4.5 Curated model is trained on a somewhat smaller training set of cleaner and more focused images.
Goose Tip: The Curated model is great for general use, and if you want to make extra sure you don't generate any sensitive images on accident. It's the safest choice when streaming video!
With V4.5, we made numerous changes and optimizations to the V4.0 model architecture, which improve the overall quality of generations, eliminate visual artifacts and improve the model's capability to learn various concepts. Due to this, V4.5 also comes with first class support for prompting images with multiple characters, natural language understanding and further improvements to the rendering of English text into your envisioned scenes.
V4.5 has a prompt token length limit of up to ~512 T5 tokens over the base prompt and all character prompts combined.
Note: since V4.5 uses the T5 tokenizer, be aware that most Unicode characters (e.g. colorful emoji or Japanese characters) are not supported by the model as part of prompts.
For those that are a fan of furry and kemono art styles, you can now begin your prompts with fur dataset, as a tag to then prompt away as you used to with our previous Furry models.
Last but not least, V4.5 also allows for free generations for Opus users as long as the usual conditions are met.
NovelAI Diffusion V4 Full
|
|
|
V4 Full is our latest image generation model. Available for all subscribing tiers.
It is part of our first family of completely original image generation models, trained from the ground up without relying on a public base model like Stable Diffusion. V4 Full not only includes a more comprehensive, albeit less curated, dataset, but the data was also updated by one month compared to V4 Curated.
Goose Tip: Give the V4 Full model a try if you're attempting to generate something specific and the V4 Curated model doesn't seem able to do it!
It brings many new functionalities to your toolbox, such as first class support for prompting images with multiple characters, natural language understanding and the ability to better render English text into your envisioned scenes. All that is accompanied by an enhanced level of detail, improved creativity and a vastly expanded and updated knowledge base, as well as a much expanded dataset.
V4 models also have a higher prompt token length limit of up to ~512 T5 tokens over the base prompt and all character prompts combined, giving you more space to express your ideas than ever before.
Note: since V4 uses the T5 tokenizer, be aware that most Unicode characters (e.g. colorful emoji or Japanese characters) are not supported by the model as part of prompts.
For those that are a fan of furry and kemono art styles, you can now begin your prompts with fur dataset, as a tag to then prompt away as you used to with our previous Furry models.
Last but not least, V4 also allows for free generations for Opus users as long as the usual conditions are met.
NovelAI Diffusion V4 Curated
|
|
|
V4 Curated is our other latest image generation model. Available for all subscribing tiers.
Just like V4 Full, it is part of our first family of completely original image generation models, trained from the ground up without relying on a public base model like Stable Diffusion.
Compared to V4 Full, the NovelAI Diffusion V4 Curated model is trained on a somewhat smaller training set of cleaner and more focused images.
Goose Tip: The Curated model is great for general use, and if you want to make extra sure you don't generate any sensitive images on accident. It's the safest choice when streaming video!
This model also brings many new functionalities to your toolbox, such as first class support for prompting images with multiple characters, natural language understanding and the ability to better render English text into your envisioned scenes. All that is accompanied by an enhanced level of detail, improved creativity and a vastly expanded and updated knowledge base, as well as a much expanded dataset.
V4 models also have a higher prompt token length limit of up to ~512 T5 tokens over the base prompt and all character prompts combined, giving you more space to express your ideas than ever before.
Note: since V4 uses the T5 tokenizer, be aware that most Unicode characters (e.g. colorful emoji or Japanese characters) are not supported by the model as part of prompts.
For those that are a fan of furry and kemono art styles, you can now begin your prompts with fur dataset, as a tag to then prompt away as you used to with our previous Furry models.
Last but not least, V4 also allows for free generations for Opus users as long as the usual conditions are met.
NovelAI Diffusion Anime V3
|
|
|
V3 is our previous anime image generation model. Available for all subscribing tiers.
Based on SDXL, along with some technology of our own, it offers much higher levels of coherence and knowledge compared to our older models.
Furthermore, V3 has a special feature of focusing more on the tags placed closer to the start of the prompt. Giving you more control over how your images are generated. More about that in the Tagging page
V3 also benefits from the higher base resolution of our V2 model (832x1216), and it also allows for free generations for Opus users as long as the usual conditions are met.\
NovelAI Diffusion Furry V3
|
|
|
The furry model makes a comeback with its own V3 model!
Brought up to date with our advancements in technologies, the V3 version of the Furry model sports extensive improvements to quality and accuracy.
NovelAI Diffusion Furry V3 has the same higher base resolution (832x1216) as the Anime V3 model, access to the same samplers, Tag ordering and also the Vibe Transfer tool.
Don't forget that the furry models use different Tags from the Anime models'. Make sure to use the handy tag suggestions to your advantage!
Goose Tip: Give the Furry model a try and don't overlook the possibilities of this model, it can do so much more than furry art!
NovelAI Diffusion Anime V2
|
|
|
Our V2 anime model is a step up from our first models.
It uses the same technologies as the V1 anime models, but benefits from updated training methodologies.
V2 also marks our increase in training resolution, meaning that it generates higher resolution images without the need of SMEA sampling.
When using the NovelAI Diffusion AnimeV2, and onward, the default portrait resolution will be 832x1216, increased from V1's 512x768.
Of course Opus subscribers will be able to generate 1024x1024, 832x1216, 1216x832, and similar resolution images for free, depending on the other settings.
Retired
NovelAI Diffusion Furry (Beta V1.3)
|
|
|
The NovelAI Diffusion Furry V1.3 model makes it easy to create content themed around furries and anthropomorphic animals.
This is our older, Beta furry model.
Note that this model uses different Tags from the Anime models'.
Though no longer available in the model selection list, we have released the weights of the Furry (Beta V1.3) model under a dual license, with both terms of the original CreativeML Open RAIL-M license and the Creative Commons BY-NC-SA 4.0 license in a blogpost here.
NovelAI Diffusion Anime (Curated)
|
|
|
The NovelAI Diffusion Anime (Curated) model provides good baseline quality and predictable subject matter.
It is trained on a somewhat smaller training set of cleaner and more focused images.
Goose Tip: The (Curated) model is great for general use, and if you want to make extra sure you don't generate any sensitive images on accident. It's the safest choice when streaming video!
Though no longer available in the model selection list, we have released the weights of the NovelAI Diffusion Anime (Curated) model under a dual license, with both terms of the original CreativeML Open RAIL-M license and the Creative Commons BY-NC-SA 4.0 license in a blogpost here.
NovelAI Diffusion Anime (Full)
|
|
|
The NovelAI Diffusion Anime (Full) model has an expanded training set that allows for a wider variety of generations.
Compared to the (Curated) Model, it may have an easier time working with cruder concepts.
Goose Tip: Give the (Full) model a try if you're attempting to generate something specific and the (Curated) model doesn't seem able to do it!
Though no longer available in the model selection list, we have released the weights of the NovelAI Diffusion Anime (Full) model under a dual license, with both terms of the original CreativeML Open RAIL-M license and the Creative Commons BY-NC-SA 4.0 license in a blogpost here.
Note
Please keep in mind that the NovelAI Diffusion models will not produce the same results as the standard Stable Diffusion model, even with the same prompt and seed.