anima nsfw
anima nsfw — The Anima text-to-image model by CircleStone Labs and Comfy Org includes built-in safety tag support for managing NSFW content. Users can control content generation through specific tags in prompts, with options ranging from safe to explicit. This page explains how safety tags work, how to structure prompts for desired content, and what limitations apply to NSFW generation.
Read source
anima nsfw: Model Overview
Anima is a 2 billion parameter text-to-image model designed for anime and non-photorealistic illustration. It supports safety tags that allow users to specify content boundaries, including NSFW generation. Anima nsfw is trained on Danbooru-style tags and natural language captions, with specific tags for content safety levels.
Safety Tag System and Content Levels
Anima uses a four-level safety tag system: safe, sensitive, nsfw, and explicit. These tags are placed at the beginning of prompts in the quality/meta/year/safety tags section. The recommended positive prefix includes "safe" as default, while the negative prompt typically includes quality tags like "worst quality, low quality" but not safety tags by default. Users can adjust safety tags to match their intended content. Anima nsfw was trained on datasets that include these safety classifications, allowing it to understand and respect content boundaries. When using NSFW tags, the model generates content accordingly, though the base version may produce unexpected results with short prompts. The safety tag system gives users granular control over content generation, from completely safe illustrations to explicit material. This system is particularly important for artists and creators who need to generate content for different audiences or purposes.
Prompt Structure for NSFW Content
To generate NSFW content with Anima, users must include the appropriate safety tag in the prompt structure. The recommended tag order is: [quality/meta/year/safety tags] [1girl/1boy/1other etc] [character] [series] [artist] [general tags]. Safety tags like "nsfw" or "explicit" should be placed in the first section alongside quality and meta tags. Anima nsfw was trained with random tag dropout, meaning not every relevant tag needs to be included. However, for reliable NSFW generation, including the safety tag is essential. Users should also write sufficiently detailed prompts to avoid undesired content, as short prompts can give unexpected results. The model's understanding of safety tags comes from its training on Danbooru-style datasets, which include these classifications. For best results with NSFW content, combine safety tags with detailed descriptions of the desired scene, characters, and actions.
anima nsfw: Capabilities and Access
Anima's NSFW capabilities are built into its core architecture, with support for safety tags that control content generation. Anima nsfw can generate a wide range of content from safe illustrations to explicit material, depending on user input. Access to NSFW features depends on the platform hosting the model.
NSFW Generation Capabilities
Anima can generate NSFW content when prompted with appropriate safety tags. Anima nsfw supports tags like "nsfw" and "explicit" in the safety tag section of prompts. It was trained on several million anime images, which include NSFW content from datasets like Danbooru. The model's understanding of NSFW concepts comes from this training data, with a knowledge cut-off date of September 2025 for anime training data. Users can generate explicit content by including the relevant safety tags and providing detailed prompts. However, the model may generate undesired content if prompts are short or lack details. The base version (Anima-Base) offers maximum flexibility and diversity, making it suitable for NSFW generation. The aesthetic version (Anima-Aesthetic) is fine-tuned on high-quality images and may handle NSFW content differently, with the recommendation to avoid score tags in both positive and negative prompts for this version.
Platform Access and Restrictions
Anima is officially supported on several online platforms for hosted image generation, including CivitAI, TensorArt, KusArt, IMGNAI, mage, AliveAI, and DreamerLand. Each platform may have its own content policies regarding NSFW generation. Anima nsfw itself does not have built-in content filters beyond the safety tag system; it relies on user-provided tags to determine content boundaries. The model is licensed under the CircleStone Labs Non-Commercial License, which restricts commercial use of the model itself but allows commercial use of generated images. Users should check individual platform policies for NSFW content restrictions. The model is not deployed by any inference provider by default, meaning users must either run it locally or use one of the supported platforms. Local users can download the model from Hugging Face and run it in ComfyUI, giving them full control over content generation without platform restrictions.
Model Versions and Download Access
Anima is available in multiple versions, each with different characteristics for NSFW generation. Anima nsfw files can be downloaded from Hugging Face and used locally or through supported online platforms. Understanding the differences between versions helps users choose the right one for their needs.
Anima Model Versions for NSFW Use
Anima comes in three main versions: Anima-Base, Anima-Aesthetic, and Anima-Turbo. Anima-Base is the pretrained, unrefined base model offering maximum flexibility and diversity, making it ideal for NSFW generation where users want full control over style and content. Anima-Aesthetic is fine-tuned for better consistency and higher quality default art style, with v1.0b being an alternate version without additional style adjustments. Anima-Turbo is a distilled version for fast generations, using CFG 1 and 8-12 steps, with increased stability and a strong default style but reduced diversity. For NSFW content, the base version is recommended because it has no aggressive aesthetic tuning that might interfere with content generation. The aesthetic version may handle NSFW content differently due to its fine-tuning on high-quality images only. Users should consider their specific needs when choosing a version for NSFW generation.
Download and Installation Process
To use Anima locally, users need to download three model files from Hugging Face: anima-base-v1.0.safetensors (diffusion model), qwen_3_06b_base.safetensors (text encoder), and qwen_image_vae.safetensors (VAE). These files go in specific folders inside the ComfyUI model directory: diffusion_models, text_encoders, and vae respectively. Anima nsfw is natively supported in ComfyUI, and workflows can be loaded from the template library or downloaded as JSON files. For NSFW generation, users should ensure they have the latest ComfyUI version. The model works at resolutions between 512^2 and 1536^2 pixels, with recommended settings of 30-50 steps and CFG 4-5. A variety of samplers work, with er_sde being a reasonable default for neutral style. The model runs comfortably on consumer GPUs due to its compact 2B parameter size.
Configuration and Prompting for NSFW Content
Proper configuration and prompting are essential for reliable NSFW generation with Anima. Anima nsfw uses specific tag formats, recommended settings, and prompt structures that differ from other text-to-image models. Understanding these configurations helps users achieve desired results.
Recommended Prompt Settings for NSFW
For NSFW generation, users should follow the recommended prompt structure: start with quality/meta/year/safety tags, then character count, character name, series, artist, and general tags. The recommended positive prefix is "masterpiece, best quality, score_7, safe, " but for NSFW content, replace "safe" with "nsfw" or "explicit". The recommended negative prompt includes "worst quality, low quality, score_1, score_2, score_3, artist name, blurry, jpeg artifacts, chromatic aberration". Tags should be lowercase with spaces instead of underscores, except for score tags which use underscores. When using tags that differ between Danbooru and Gelbooru, prefer the Gelbooru version. Prompt weighting works but needs higher weights than typically used for SDXL, for example "(chibi:2)". For the aesthetic version, avoid using score tags in both positive and negative prompts as the model is already high quality enough.
Sampler and Step Configurations
Anima works with a variety of samplers, each producing different styles. er_sde provides neutral style with flat colors and sharp lines, making it a reasonable default. euler_a gives softer, thinner lines and can sometimes tend towards a 2.5D look, with CFG that can be pushed higher than other samplers without burning the image. dpmpp_2m_sde_gpu is similar to er_sde but can produce more variety and be more creative, though it can get too wild depending on the prompt. euler is a basic sampler that is more creative than er_sde and works well with Turbo and Aesthetic versions. For a more realistic or painterly look, the beta57 scheduler from the ComfyUI RES4LYF custom node pack can help make better textures by putting more emphasis on low-noise timesteps. Recommended step count is 30-50 steps with CFG 4-5. The Turbo version uses CFG 1 and 8-12 steps for fast generation.
Practical Notes and Limitations
While Anima supports NSFW generation, there are important practical considerations and limitations users should understand. These include the model's design focus, potential issues with short prompts, and legal/licensing considerations for generated content.
Content Generation Limitations
Anima has several limitations that affect NSFW generation. The model is designed for anime and illustration styles and does not do realism well, which is intentional. It may generate undesired content, especially if prompts are short or lacking details. Users should avoid this by using appropriate safety tags in positive and negative prompts and by writing sufficiently detailed prompts. The model is not great at text rendering, handling single words and sometimes short phrases but not lengthy text. The base version has a very plain and neutral default style, which is especially apparent without artist or quality tags. For NSFW content, users should be aware that the model's training data includes a wide range of content, but results may vary depending on prompt specificity. The aesthetic version may handle NSFW content differently due to its fine-tuning on high-quality images only.
Legal and Licensing Considerations
Anima is licensed under the CircleStone Labs Non-Commercial License, which restricts commercial use of the model itself but allows commercial use of generated images. This means users can sell images created with Anima, take paid commissions, or use generated images as concept art for paid products like video games or visual novels. However, hosting the model behind an API for paid access, using it on paid online platforms, embedding model weights in monetized products, or using it to power features in larger monetized products requires a separate commercial license. The model is also subject to the NVIDIA Open Model License Agreement as a Derivative Model of Cosmos-Predict2-2B-Text2Image. Users interested in commercial licensing should contact tdrussell@circlestone.ai. These licensing terms apply regardless of whether the generated content is NSFW or safe.
FAQ
Common questions about anima nsfw content handling, safety tags, and model capabilities.
What should I know about anima nsfw?
Anima supports NSFW content generation through its safety tag system, which includes tags like safe, sensitive, nsfw, and explicit. Users must include the appropriate safety tag in their prompt to generate NSFW content. The model was trained on datasets that include NSFW content, with a knowledge cut-off date of September 2025 for anime training data. For reliable NSFW generation, use detailed prompts with appropriate safety tags and follow the recommended prompt structure. The base version (Anima-Base) offers maximum flexibility for NSFW content, while the aesthetic version may handle it differently. Users should be aware that short prompts can lead to unexpected results, and the model may generate undesired content without proper prompting.
How do I use safety tags for NSFW content?
Safety tags are placed at the beginning of prompts in the quality/meta/year/safety tags section. The recommended tag order is: [quality/meta/year/safety tags] [1girl/1boy/1other etc] [character] [series] [artist] [general tags]. For NSFW content, replace the default "safe" tag with "nsfw" or "explicit". The model was trained with random tag dropout, so not every tag needs to be included. However, including the safety tag is essential for reliable NSFW generation. Users should also write sufficiently detailed prompts to avoid undesired content. The model understands safety tags from its training on Danbooru-style datasets, which include these classifications.
What platforms support Anima for NSFW generation?
Anima is officially supported on several online platforms including CivitAI, TensorArt, KusArt, IMGNAI, mage, AliveAI, and DreamerLand. Each platform may have its own content policies regarding NSFW generation, so users should check individual platform policies. The model itself does not have built-in content filters beyond the safety tag system. For full control over NSFW generation without platform restrictions, users can run the model locally using ComfyUI. The model files are available for download from Hugging Face, and local installation requires placing model files in specific folders within the ComfyUI model directory.