Hello Armada is a new checkpoint designed to generate and your Sexy Waifus in a visually appealing nsfw way with a more toned aged up adult female/girl.Thanks and huge shout out to to Urusan for training and creating this checkpoint on his machine using my Armada style. This checkpoint would not be possible without him so do show appreciation and thank him for his incredible contribution.<a target="_blank" rel="ugc" href="https://civarchive.com/user/urusan">https://civarchive.com/user/urusan</a>This is an Anima based checkpoint so the following Anima files are needed as well The model files go in their respective folders inside your model directory:<ul><li>anima-base-v1.0.safetensors goes in ComfyUI/models/diffusion_models</li><li>qwen_3_06b_base.safetensors goes in ComfyUI/models/text_encoders</li><li>qwen_image_vae.safetensors goes in ComfyUI/models/vae (this is the Qwen-Image VAE, you might already have it</li><li>Generation settings</li></ul><h2 id="run-settings:">Run settings: </h2>There is a lot of experimenting you can do as I didn't do all variations and test but these are the settings I use daily: Resolution: 1248 x 1856 or 1536 x 1536 Steps: 30-40CFG: 4samplers: Euler A Running the resolutions above means no need for face detailers or face fix or anything at least in Anima just text and prompt away. <h2 id="prompting">Prompting</h2>This is an anima checkpoint so just use anima You only need one positive prompt: @armadaThe other positive prompts like masterpiece, best quality, etc... you can add but not needed. Positive prompt:<pre><code>@armada, masterpiece, best quality, amazing quality, aged up, adult female, </code></pre><pre><code>optional: masterpiece, best quality, amazing quality, aged up, adult female, but it helps a lot</code></pre>Negative tags: <pre><code>(bad quality:1.15), (worst quality:1.3), worst quality, low quality, score_1, score_2, score_3, blurry, jpeg artifacts, watermark, artist tag, artist logo, sepia, body horror, bad anatomy, too many fingers, deformed hand, deformed hands, too many arms, ugly, uncanny valley, horror, decay, injury, twisted fingers, twisted arms, creepy, mutation, hideous, burned, gore, guro, trypophobia, vomit, watermark, focal blur, blurry background, sketch, white theme, pubic hair, pointy chin, pointed chin, sharp chin, chibi, loli, underage, child, hair over one eye, audience,</code></pre>Simple example <pre><code>This positive prompt will give you the first image below
@armada, masterpiece, best quality, amazing quality, aged up, adult female</code></pre><pre><code>red hair, tanned skin, tanlines, high ponytail, spiked hair, green eyes, large breasts, eyepatch bikini, string panties, beach, scenery</code></pre><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f80bec09-8ee1-4afe-a5bb-eb6163e95250/original=true/f80bec09-8ee1-4afe-a5bb-eb6163e95250.jpeg" />Enjoy the Waifus and post and enjoy !

Armada Checkpoint

armadaCheckpoint_armadaCheckpointV10.safetensors

Armada Checkpoint v1.0

触发词 Cantarella

坎特蕾拉

anima.safetensors

anima_baseV10.safetensors

v1.0

Anima is a 2 billion parameter text-to-image model created via a collaboration between CircleStone Labs and Comfy Org. It is focused mainly on anime concepts, characters, and styles, but is also capable of generating a wide variety of other non-photorealistic content. The model is designed for making illustrations and artistic images, and will not work well at realism.It is trained on several million anime images and about 800k non-anime artistic images. No synthetic data was used for training. The knowledge cut-off date for the anime training data is September 2025.<h1 id="versions">Versions</h1><ul><li>Anima-Base<ul><li>The pretrained, unrefined base model. Maximum flexibility, diversity, and style adherence.</li><li>LoRAs should be trained using this version.</li></ul></li><li>Anima-Aesthetic<ul><li>Fine-tuned for better consistency and a higher quality default art style.</li></ul></li><li>Anima-Turbo<ul><li>Distilled version for fast generations.</li><li>Use at CFG 1 and 8-12 steps.</li><li>The distillation process also increases stability and gives the model a strong default style, but reduces diversity.</li></ul></li></ul>I recommend starting with Anima-Turbo. On average, it is only slightly worse than Anima-Aesthetic, while being very fast to generate (and much cheaper if you use it on an online platform that scales the cost with step count). This makes it very convenient for quickly iterating on prompts. The increased stability can even make it better than Aesthetic in some cases.<h1 id="installing-and-running">Installing and running</h1>Get the text encoder and VAE from the <a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima">HuggingFage page</a>.The model is natively supported in ComfyUI. The model files go in their respective folders inside your model directory:<ul><li>anima-base-v1.0.safetensors goes in ComfyUI/models/diffusion_models</li><li>qwen_3_06b_base.safetensors goes in ComfyUI/models/text_encoders</li><li>qwen_image_vae.safetensors goes in ComfyUI/models/vae (this is the Qwen-Image VAE, you might already have it)</li></ul><h2 id="generation-settings">Generation settings</h2><ul><li>Works at resolutions between 512^2 and 1536^2 pixels.</li><li>30-50 steps, CFG 4-6.<ul><li>The Aesthetic version can tolerate lower CFGs such as 3, and often looks better with them.</li></ul></li><li>A variety of samplers work. Some of my favorites:<ul><li>er_sde: neutral style, flat colors, sharp lines. I use this as a reasonable default.</li><li>euler_a: Softer, thinner lines. Can sometimes tend towards a 2.5D look. CFG can be pushed a bit higher than other samplers without burning the image.</li><li>dpmpp_2m_sde_gpu: similar in style to er_sde but can produce more variety and be more "creative". Depending on the prompt it can get too wild sometimes.</li><li>euler: a basic sampler that is a bit more creative than er_sde. Good with the Turbo and Aesthetic versions, since those are naturally more stable.</li></ul></li><li>If going for a more realistic / painterly look, the beta57 scheduler (ComfyUI RES4LYF custom node pack) can help make better textures, since it puts more emphasis on low-noise timesteps.</li></ul><h1 id="prompting">Prompting</h1>The model is trained on Danbooru-style tags, natural language captions, and combinations of tags and captions.<ul><li>Use lowercase for tags, and spaces instead of underscores. Score tags are the only tags that use underscores.</li><li>Recommended positive prefix: "masterpiece, best quality, score_7, safe, "</li><li>Recommended negative: "worst quality, low quality, score_1, score_2, score_3, artist name, blurry, jpeg artifacts, chromatic aberration"</li><li>When using a tag that is different between Danbooru and Gelbooru, prefer the Gelbooru version.</li><li>Prompt weighting works, but needs a weight higher than typically used for SDXL. Example: "(chibi:2)"</li></ul><h2 id="aesthetic-version-prompting">Aesthetic Version Prompting</h2>Anima-Aesthetic is fine-tuned only on high quality images, with all of the quality tags stripped out from the captions. You don't need to use quality tags in the positive at all, but "masterpiece, best quality, " is safe to leave in. I recommend not using score_* tags in both the positive and negative prompt. It is already high quality enough and the score tags can push it too hard into slop territory.<h2 id="tag-order">Tag order</h2>[quality/meta/year/safety tags] [1girl/1boy/1other etc] [character] [series] [artist] [general tags]Within each tag section, the tags can be in arbitrary order.<h2 id="quality-tags">Quality tags</h2>Human score based: masterpiece, best quality, good quality, normal quality, low quality, worst qualityPonyV7 aesthetic model based: score_9, score_8, ..., score_1You can use either the human score quality tags, the aesthetic model tags, both together, or neither. All combinations work.<h2 id="time-period-tags">Time period tags</h2>Specific year: year 2025, year 2024, ...Period: newest, recent, mid, early, old<h2 id="meta-tags">Meta tags</h2>highres, absurdres, anime screenshot, jpeg artifacts, official art, etc<h2 id="safety-tags">Safety tags</h2>safe, sensitive, nsfw, explicit<h2 id="artist-tags">Artist tags</h2>Prefix artist with @. E.g. "@big chungus". You must put @ in front of the artist. The effect will be very weak if you don't.<h2 id="full-tag-example">Full tag example</h2>year 2025, newest, normal quality, score_5, highres, safe, 1girl, oomuro sakurako, yuru yuri, @nnn yryr, smile, brown hair, hat, solo, fur-trimmed gloves, open mouth, long hair, gift box, fang, skirt, red gloves, blunt bangs, gloves, one eye closed, shirt, brown eyes, santa costume, red hat, skin fang, twitter username, white background, holding bag, fur trim, simple background, brown skirt, bag, gift bag, looking at viewer, santa hat, ;d, red shirt, box, gift, fur-trimmed headwear, holding, red capelet, holding box, capelet<h2 id="tag-dropout">Tag dropout</h2>The model was trained with random tag dropout. You don't need to include every single relevant tag for the image.<h2 id="dataset-tags">Dataset tags</h2>To improve style and content diversity, the model was additionally trained on two non-anime datasets: LAION-POP (specifically the ye-pop version) and DeviantArt. Both were filtered to exclude photos. Because these datasets are qualitatively different from anime datasets, captions from them have been labeled with a "dataset tag". This occurs at the very beginning of a prompt followed by a newline. Optionally, the second line can contain either the image alt-text (ye-pop) or the title of the work (DeviantArt). Examples:<pre><code>ye-pop
For Sale: Others by Arun Prem
Abstract, oil painting of three faceless, blue-skinned figures. Left: white, draped figure; center: yellow-shirted, dark-haired figure; right: red-veiled, dark-haired figure carrying another. Bold, textured colors, minimalist style.</code></pre><pre><code>deviantart
Flame
Digital painting of a fiery dragon with glowing yellow eyes, black horns, and a long, sinuous tail, perched on a glowing, molten rock formation. The background is a gradient of dark purple to orange.</code></pre><h2 id="natural-language-prompting-tips">Natural language prompting tips</h2><ul><li>Follow standard English capitalization rules for character and series names.</li><li>If using pure natural language, more descriptive is better. Aim for at least 2 sentences. Extremely short prompts can give unexpected results.</li><li>You can mix tags and natural language in arbitrary order.</li><li>You can put quality / artist tags at the beginning of a natural language prompt.<ul><li>"masterpiece, best quality, @big chungus. An anime girl with medium-length blonde hair is..."</li></ul></li><li>Name a character, then describe their basic appearance.<ul><li>"Digital artwork of Fern from Sousou no Frieren, with long purple hair and purple eyes, wearing a black coat over a white dress with puffy sleeves..."</li><li>This is extra important when prompting for multiple characters. If you just list off character names with no description of appearance, the model can get confused.</li></ul></li></ul><h1 id="limitations">Limitations</h1><ul><li>The model doesn't do realism well. This is intended. It is an anime / illustration / art focused model.</li><li>The model may generate undesired content, especially if the prompt is short or lacking details.<ul><li>Avoid this by using the appropriate safety tags in the positive and negative prompts, and by writing sufficiently detailed prompts.</li></ul></li><li>The model isn't great at text rendering. It can generally do single words and sometimes short phrases, but lengthy text rendering won't work well.</li><li>The base version is a true base model. It hasn't been aesthetic tuned on a curated dataset. The default style is very plain and neutral, which is especially apparent if you don't use artist or quality tags.</li></ul><h1 id="finetuning-tips">Finetuning tips</h1><ul><li>Don't train the LLM adapter. My own training script, diffusion-pipe, lets you set llm_adapter_lr=0 to completely disable training it, and the example config has this as a default.<ul><li>Other trainers like sd-scripts have similar options that should be used.</li><li>The LLM adapter processes the text embeddings before they get to the diffusion model, and therefore has an outsized influence on the generated images. The adapter itself contains a surprising amount of knowledge and is easy to degrade by training it.</li></ul></li><li>Use a low learning rate. For a rank 32 LoRA, start with 2e-5 and adjust up or down from there.<ul><li>As a base model, there is no aggressive aesthetic tuning or RLHF you need to overcome when finetuning.</li><li>The model has an extremely large and diverse amount of visual concepts baked in already. A light touch is all you need.</li></ul></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2536147">Example</a> of a style LoRA, with dataset and configs shared.</li></ul><h1 id="online-platforms">Online platforms</h1>In addition to CivitAI, the following platforms also officially support Anima for hosted image generation.<ul><li><a target="_blank" rel="ugc" href="https://tensor.art/models/998619401697644790/Anima-Official-circlestone_labs-base-v1.0">TensorArt</a></li><li><a rel="ugc" href="https://kusart.com">KusArt</a></li><li><a target="_blank" rel="ugc" href="https://imgnai.com/">IMGNAI</a></li><li><a target="_blank" rel="ugc" href="https://www.mage.space/">mage</a></li><li><a target="_blank" rel="ugc" href="https://aliveai.app/">AliveAI</a></li><li><a target="_blank" rel="ugc" href="https://www.dreamerland.ai/">DreamerLand</a></li></ul><h1 id="license">License</h1>This model is licensed under the CircleStone Labs Non-Commercial License. The model and derivatives are only usable for non-commercial purposes. Additionally, this model constitutes a "Derivative Model" of Cosmos-Predict2-2B-Text2Image, and therefore is subject to the <a target="_blank" rel="ugc" href="https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/">NVIDIA Open Model License Agreement</a> insofar as it applies to Derivative Models.If you would like a commercial license, please email tdrussell@circlestone.aiBuilt on NVIDIA Cosmos.

Anima

anima_baseV10_txt.safetensors

qwen_image_vae.safetensors

base-v1.0

aesthetic-v1.1

aesthetic-v1.0b

turbo-v1.0

aesthetic-v1.0

preview3-base

preview2

Anima is a 2 billion parameter text-to-image model created via a collaboration between CircleStone Labs and Comfy Org. It is focused mainly on anime concepts, characters, and styles, but is also capable of generating a wide variety of other non-photorealistic content. The model is designed for making illustrations and artistic images, and will not work well at realism.It is trained on several million anime images and about 800k non-anime artistic images. No synthetic data was used for training. The knowledge cut-off date for the anime training data is September 2025.NEW: Try the <a target="_blank" rel="ugc" href="https://civitai.com/models/2560840/anima-turbo-lora">Turbo LoRA</a> for better stability and much faster generations.Versions<ul><li>Anima-Base: The pretrained, unrefined base model. Maximum flexibility, diversity, and style adherence.</li><li>Anima-Turbo: Coming soon.</li></ul>Installing and runningGet the text encoder and VAE from the <a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima">HuggingFage page</a>.The model is natively supported in ComfyUI. The model files go in their respective folders inside your model directory:<ul><li>anima-base-v1.0.safetensors goes in ComfyUI/models/diffusion_models</li><li>qwen_3_06b_base.safetensors goes in ComfyUI/models/text_encoders</li><li>qwen_image_vae.safetensors goes in ComfyUI/models/vae (this is the Qwen-Image VAE, you might already have it)</li></ul>Generation settingsWorks at resolutions between 512^2 and 1536^2 pixels.30-50 steps, CFG 4-6.A variety of samplers work. Some of my favorites:<ul><li>er_sde: neutral style, flat colors, sharp lines. I use this as a reasonable default.<ul><li>euler_a: Softer, thinner lines. Can sometimes tend towards a 2.5D look. CFG can be pushed a bit higher than other samplers without burning the image.</li><li>dpmpp_2m_sde_gpu: similar in style to er_sde but can produce more variety and be more "creative". Depending on the prompt it can get too wild sometimes.</li></ul>If going for a more realistic / painterly look, the beta57 scheduler (ComfyUI RES4LYF custom node pack) can help make better textures, since it puts more emphasis on low-noise timesteps.</li></ul>PromptingThe model is trained on Danbooru-style tags, natural language captions, and combinations of tags and captions.<ul><li>Use lowercase for tags, and spaces instead of underscores. Score tags are the only tags that use underscores.</li><li>Recommended positive prefix: "masterpiece, best quality, score_7, safe, "</li><li>Recommended negative: "worst quality, low quality, score_1, score_2, score_3, artist name"</li><li>When using a tag that is different between Danbooru and Gelbooru, prefer the Gelbooru version.</li><li>Prompt weighting works, but needs a weight higher than typically used for SDXL. Example: "(chibi:2)"</li></ul>Tag order[quality/meta/year/safety tags] [1girl/1boy/1other etc] [character] [series] [artist] [general tags]Within each tag section, the tags can be in arbitrary order.Quality tagsHuman score based: masterpiece, best quality, good quality, normal quality, low quality, worst qualityPonyV7 aesthetic model based: score_9, score_8, ..., score_1You can use either the human score quality tags, the aesthetic model tags, both together, or neither. All combinations work.Time period tagsSpecific year: year 2025, year 2024, ...Period: newest, recent, mid, early, oldMeta tagshighres, absurdres, anime screenshot, jpeg artifacts, official art, etcSafety tagssafe, sensitive, nsfw, explicitArtist tagsPrefix artist with @. E.g. "@big chungus". You must put @ in front of the artist. The effect will be very weak if you don't.Full tag exampleyear 2025, newest, normal quality, score_5, highres, safe, 1girl, oomuro sakurako, yuru yuri, @nnn yryr, smile, brown hair, hat, solo, fur-trimmed gloves, open mouth, long hair, gift box, fang, skirt, red gloves, blunt bangs, gloves, one eye closed, shirt, brown eyes, santa costume, red hat, skin fang, twitter username, white background, holding bag, fur trim, simple background, brown skirt, bag, gift bag, looking at viewer, santa hat, ;d, red shirt, box, gift, fur-trimmed headwear, holding, red capelet, holding box, capeletTag dropoutThe model was trained with random tag dropout. You don't need to include every single relevant tag for the image.Dataset tagsTo improve style and content diversity, the model was additionally trained on two non-anime datasets: LAION-POP (specifically the ye-pop version) and DeviantArt. Both were filtered to exclude photos. Because these datasets are qualitatively different from anime datasets, captions from them have been labeled with a "dataset tag". This occurs at the very beginning of a prompt followed by a newline. Optionally, the second line can contain either the image alt-text (ye-pop) or the title of the work (DeviantArt). Examples:ye-pop For Sale: Others by Arun Prem Abstract, oil painting of three faceless, blue-skinned figures. Left: white, draped figure; center: yellow-shirted, dark-haired figure; right: red-veiled, dark-haired figure carrying another. Bold, textured colors, minimalist style.deviantart Flame Digital painting of a fiery dragon with glowing yellow eyes, black horns, and a long, sinuous tail, perched on a glowing, molten rock formation. The background is a gradient of dark purple to orange.Natural language prompting tipsFollow standard English capitalization rules for character and series names.If using pure natural language, more descriptive is better. Aim for at least 2 sentences. Extremely short prompts can give unexpected results.You can mix tags and natural language in arbitrary order.You can put quality / artist tags at the beginning of a natural language prompt.<ul><li>"masterpiece, best quality, @big chungus. An anime girl with medium-length blonde hair is..."</li></ul>Name a character, then describe their basic appearance.<ul><li>"Digital artwork of Fern from Sousou no Frieren, with long purple hair and purple eyes, wearing a black coat over a white dress with puffy sleeves..."</li><li>This is extra important when prompting for multiple characters. If you just list off character names with no description of appearance, the model can get confused.</li></ul>LimitationsThe model doesn't do realism well. This is intended. It is an anime / illustration / art focused model.The model may generate undesired content, especially if the prompt is short or lacking details.<ul><li>Avoid this by using the appropriate safety tags in the positive and negative prompts, and by writing sufficiently detailed prompts.</li></ul>The model isn't great at text rendering. It can generally do single words and sometimes short phrases, but lengthy text rendering won't work well.The base version is a true base model. It hasn't been aesthetic tuned on a curated dataset. The default style is very plain and neutral, which is especially apparent if you don't use artist or quality tags.Finetuning tipsDon't train the LLM adapter. My own training script, diffusion-pipe, lets you set llm_adapter_lr=0 to completely disable training it, and the example config has this as a default.<ul><li>Other trainers like sd-scripts have similar options that should be used.</li><li>The LLM adapter processes the text embeddings before they get to the diffusion model, and therefore has an outsized influence on the generated images. The adapter itself contains a surprising amount of knowledge and is easy to degrade by training it.</li></ul>Use a low learning rate. For a rank 32 LoRA, start with 2e-5 and adjust up or down from there.<ul><li>As a base model, there is no aggressive aesthetic tuning or RLHF you need to overcome when finetuning.</li><li>The model has an extremely large and diverse amount of visual concepts baked in already. A light touch is all you need.</li></ul><a target="_blank" rel="ugc" href="https://civitai.com/models/2536147">Example</a> of a style LoRA, with dataset and configs shared.LicenseThis model is licensed under the CircleStone Labs Non-Commercial License. The model and derivatives are only usable for non-commercial purposes. Additionally, this model constitutes a "Derivative Model" of Cosmos-Predict2-2B-Text2Image, and therefore is subject to the <a target="_blank" rel="ugc" href="https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/">NVIDIA Open Model License Agreement</a> insofar as it applies to Derivative Models.If you would like a commercial license, please email <a target="_blank" rel="noopener noreferrer nofollow" href="mailto:tdrussell@circlestone.ai">tdrussell@circlestone.ai</a>Built on NVIDIA Cosmos.

Files with SHA256: bd43b7cffe1ed1153d9c41e7beb2f18cb1273eafbaa3af3edd6a173dc90a006e