CivArchive
    Preview 143587082
    Preview 143587000
    Preview 143586999
    Preview 143587003
    Preview 143587004
    Preview 143587009
    Preview 143587011
    Preview 143587037
    Preview 143587038
    Preview 143587059
    Preview 143587062
    Preview 143587068
    Preview 143587069
    Preview 143587079
    Preview 143587099
    Preview 143587100
    Preview 143587109
    Preview 143587113
    Preview 143587129
    Preview 143587128

    Ver. 2.0 adopts the "2.9B" base model, enabling more sophisticated visual expression and improving colors, background detail, and rendering quality.
    Ver2.0では、より高度な表現が可能な「2.9B」ベースモデルを採用し、色合い・背景の詳細度・塗りの品質などを向上しました。

    "HotaruBreedAnima" is an Anima-based checkpoint designed for generating 2D illustrations, with a focus on cuteness.
    HotaruBreedAnimaは二次元イラストレーション生成に特化したモデルで、可愛さを追求しています。

    Several variations are available :
    以下のバリエーションが用意されています :

    • Basic / 基本タイプ

    • Anime

      • 2D, anime-style rendering

      • 2Dタイプ、アニメ塗り

    • Soft (for v2.0)

      • Thinner linework and a soft, airy texture compared to the basic type

      • 基本タイプより線が細め、ふわっとした質感

    • SemiReal (for v1.0)

      • 2.5D, semi-realistic rendering

      • 2.5Dタイプ、若干リアル寄り

    I have also released the LoRA series which allows you to modify such as face part shapes, expressions and effects for this checkpoint.
    このデータモデルと組み合わせ、顔パーツの形や表情・エフェクトなどをいろいろ変更できるLoRAシリーズも公開しています。


    Highlights and cautions / 特徴と注意事項

    I feel that model merging is similar to crossbreeding flowers (selective breeding), so I include "Breed," meaning selective breeding, in the names of the models I create. As a feature of the Breed series, I merge many LoRAs that I trained myself to adjust image quality, colors, shading, and eye shapes, as well as to improve the quality of linework and eye textures, reduce grain noise, and improve the quality of HiRes Fix.
    This model has a very fixed art style. Please note that attempts to change the art style or aesthetic using prompts are generally ineffective. It's probably not suitable for use as a basis for learning drawing styles either.

    私はモデルのマージはお花の交配(品種改良)と近いものを感じているため、作成したデータモデルの名前には品種改良を意味する「Breed」を付けています。
    Breedシリーズの特徴として、作者が独自に学習させた多数のLoRAをマージすることで、画質・色味・陰影や眼の形などを調整したり、輪郭線および眼の質感向上や粒子ノイズ低減、HiRes Fixの高品質化などを図っています。
    このモデルは絵柄が強く固定されています。プロンプトによる絵柄・作風の変更はあまり効きませんのでご注意ください。絵柄学習のベースとして使うのにもおそらく不向きです。


    How to use / 使い方

    The basic settings follow those of the original Anima (base 1.0 or 2.9B) base model.
    基本的な設定は、オリジナルのAnima(base 1.0または2.9B)に準拠します。

    LoRAs developed for Anima Base 1.0 can also be applied to the 2.9B version. Please use a compatible system, such as Forge Neo.
    Anima base 1.0向けに開発されたLoRAも、2.9Bバージョンに適用できます。Forge Neoなど、対応したシステムを利用してください。

    If you want to customize the appearance (facial features, expressions, etc.) or expressions of the generated characters, you can use the various LoRAs I have published.
    出力キャラクターの外見(顔パーツ、表情など)や表現を細かく調整したい場合は、当方で公開中の各種LoRAが利用できます。

    This guide assumes that you are using Forge-neo.
    Forge-neoで使用する場合を想定した記述です。

    • CLIP : qwen_3_06b_base.safetensors

    • VAE : qwen_image_vae.safetensors

    • Sampling Method : ER SDE

    • Schedule Type : Beta

    • Sampling steps : 20 - 50 (recommended : 30)

    • Shift : 3

    • CFG Scale : 3-6 (recommended : 5)

    • Resolution : mainly tested at 960 × 1280, but various resolutions are supported as well as original Anima.

    masterpiece, score_7,
    best quality, amazing quality, very aesthetic, absurdres, perfect anatomy, detailed anatomy,
    perfect symmetrical face, perfect hands, detailed eyes, detailed hair, slender, shiny skin, beautiful skin,
    extremely detailed background,
    worst quality, low quality, score_1, score_2, score_3, artist name, blurry, jpeg artifacts, bad anatomy,
    shiny skin, shiny hair, white pupils, blank pupils,
    pillarbox background, letterbox background, framed background, empty background, simple background, white background

    The second and third lines are optional and can be omitted.

    2行目・3行目はオプションで、省略可能です。  

    Cautions for HiRes Fix / HiRes Fixの注意点

    When performing HiRes Fix, use the Anima HiRes Fix extension and set the Denoise Strength to 0.4 or lower.
    Not just this model, but Anima-series models in general tend to produce grainy images when using only the default HiRes Fix feature.

    HiRes Fixを行う際は、Anima Hires Fix 拡張機能を使用し、Denoise Strengthは0.4以下としてください。
    本モデルに限らず、Anima系モデルは、標準機能のHiRes Fixだけだと画質が荒れやすくなります。

    Description

    FAQ

    Comments (4)

    mint5958318Oct 2, 2026· 1 reaction
    CivitAI

    Your checkpoints are always incredible!

    pasteldonutsOct 8, 2026
    CivitAI

    I didn't know a new "breed" had been created. Love this. A minor pity is the lack of decorative tendency that the older IL checkpoints had, but the high consistency and the strong expression more than make up for it

    richerichOct 14, 2026
    CivitAI

    Nice. Possible to look into a 2.9B version?

    cayden_caileanOct 16, 2026
    CivitAI

    Please port to Anima you eyes and face expressions LORAs, they worked great!

    Checkpoint
    Anima

    Details

    Downloads
    216
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/23/2026
    Updated
    11/22/2026
    Deleted
    -

    Files

    HotaruBreedAnima_SemiReal_v10.safetensors

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.