NovelAI: A Comprehensive Guide to AI Image Generation

NovelAI Diffusion is the image generation component of NovelAI. Its remarkable ability to produce high-level illustrations is attributed to its extensive training data, which is believed to include a vast collection of images from platforms like Pixiv and other 'booru' sites. This rich dataset, often accompanied by detailed tags, allows NovelAI Diffusion to generate images that closely resemble professional artwork, leading to its widespread acclaim.

While NovelAI offers free text generation, image generation requires a paid subscription. There are three main plans: Tablet, Scroll, and Opus. For image generation purposes, the Scroll plan offers similar functionality to the Tablet plan, with the primary difference being in text generation capabilities. - **Tablet:** Suitable for users who want to experiment with image generation. It offers a good balance of features and cost. - **Scroll:** Primarily focused on text generation with some image generation capabilities. - **Opus:** The most comprehensive plan, offering the highest Anlas (in-service currency for image generation) and advanced features. Anlas is consumed when generating images; for instance, generating one image typically costs 5 Anlas. The Tablet plan at $10/month allows for nearly 200 images, while the Opus plan at $25/month provides around 2000 images. Users can choose a plan based on their expected usage and budget.

The image generation interface allows for detailed customization: - **Generation Mode:** Choose from NAI Diffusion Anime (Curated) for standard illustrations, NAI Diffusion Anime (Full) for adult content, or a mode for generating beast-like images. The 'Curated' mode is recommended for general use. - **Prompt:** This is where you describe the image you want to create, detailing characters, style, quality, and more. Using existing prompt examples can be a good starting point. - **Undesired Content:** Similar to negative prompts in other AI art tools, this field specifies elements you want to exclude from the generated image, such as low-resolution artifacts or poor quality. - **Image Settings:** Select the desired resolution for your image. You can input custom values or choose from presets. - **Number of Images:** Specify how many images you want to generate simultaneously.

NovelAI offers several advanced features to refine your generated images: - **Variation:** This function allows you to generate images that are similar to a selected base image. Simply click the variation button next to an image to produce four new variations. - **Enhance:** Use Enhance to upscale and add detail to an image. This is useful for improving the clarity and quality of a composition you like. Key parameters include: - **Magnitude:** Controls the level of detail and 'writing' in the image. Higher values result in finer details but consume more Anlas. - **Upscale:** Determines the magnification factor for the image resolution. - **Inpaint:** This feature enables you to selectively edit specific areas of an image. After selecting an image from your history, click the Inpaint button. You can then mark the areas you wish to modify and provide new prompts to guide the regeneration of those sections. This is effective for correcting unnatural elements or making targeted changes.

NovelAI offers advanced settings to fine-tune image generation: - **Steps:** Increasing the number of steps generally leads to higher quality images but consumes more Anlas and increases generation time. Steps between 26 and 30 offer a good balance, with 28 being the default. Lowering steps (e.g., to 26) can reduce Anlas consumption. - **Prompt Guidance:** This parameter adjusts the influence of the prompt on the image. Higher values make the prompt more dominant but can sometimes lead to less coherent images. - **Seed:** The seed value controls the randomness of image generation. Fixing the seed allows for reproducible results, which is useful for comparing the effects of different parameters or understanding how prompts are interpreted. - **Sampler:** The sampler determines the noise generation method used by the Stable Diffusion model. While there are various samplers, the default option is generally sufficient. **Comparison with Stable Diffusion:** While NovelAI excels at easily generating high-quality anime art, it offers fewer model choices compared to Stable Diffusion. With the emergence of powerful models like Anything v5, which can produce even cleaner and more aesthetically pleasing illustrations, and the availability of free services that utilize these models, some users may find Stable Diffusion more versatile and cost-effective for their needs. However, NovelAI's user-friendly interface and curated models make it an excellent starting point for many.