Sign In

Milkiss By Hexora

Updated: Aug 16, 2026

base model

Download

1 variant available

bf16 SafeTensor

Milkiss 01.safetensors

BF16, good balance • 4.13 GB

Verified:

Type
Checkpoint Merge
Stats

527

117

Generation License Fee

1 / 1 image

Reviews
Published

Aug 1, 2026

Base Model

Anima

Hash
AutoV2
A00C158D32
default creator card background decoration
Followers - 34

34

Likes - 123

123

Downloads - 1349

1.3K

License:

Anima

The Anima Model is licensed by CircleStone Labs LLC. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

Built on NVIDIA Cosmos

00002-2146117230.png

Moescape.ai Pixai Tensor


Please don’t use my thumbnail for your own uploads or posts. Thank you! ♡


Milkiss is a versatile anime illustration model designed for colorful, polished, and highly detailed artwork. It produces expressive characters, clean linework, soft yet vivid colors, beautiful lighting, and carefully rendered backgrounds. The model works especially well with cute fashion, flowers, desserts, summer scenes, fantasy themes, elegant portraits, modern city settings, and dramatic action compositions. Its style can shift naturally between soft pastel illustrations, bright pop-art visuals, refined semi-realistic anime, and darker cinematic scenes while keeping a consistent, attractive finish.


Milkissは、鮮やかで美しく、細部まで丁寧に描かれたアニメイラストを制作できる多彩なモデルです。表情豊かなキャラクター、繊細な線画、柔らかく鮮やかな色彩、美しい光の表現、そして完成度の高い背景描写を得意としています。可愛いファッション、花、スイーツ、夏の風景、ファンタジー、上品なポートレート、都会的なシーン、迫力のあるアクション構図など、幅広いテーマに対応できます。淡いパステル調からポップで明るい作品、セミリアルなアニメ表現、ダークで映画的な雰囲気まで、魅力的で統一感のある仕上がりを保ちながら柔軟に表現できます。


Recommended settings:

CFG: 4–6

Steps: 30–50

Resolution range: 512–1536


Recommended samplers:

  • er_sde

  • euler_a

  • dpmpp_2m_sde_gpu


er_sde tends to produce cleaner anime-style linework and flatter color control.

euler_a often gives softer lines and a more painterly/2.5D feel.

dpmpp_2m_sde_gpu can create more experimental and creative outputs depending on prompting.


The model responds well to both tag-based prompting and natural language prompting, including mixed prompt structures.

Recommended prompt structure:

[quality/meta tags] → [subject] → [character/series] → [artist/style] → [general details] → [lighting/composition]

Example:

masterpiece, best quality, newest, safe, 1girl, original character, @artistname, white hair, black dress, glowing eyes, cinematic lighting, red moon, wind, night, detailed background


Natural language prompting also works very well:

A pale anime girl with long white hair standing under a crimson moon at night. Cinematic lighting, dark atmosphere, glowing red eyes, wind blowing through her hair, highly detailed illustration.


Mixed prompting:

masterpiece, best quality, newest, 1girl, white hair, black dress, glowing eyes.

A dark fantasy anime scene with cinematic lighting and red atmospheric fog.


Recommended quality/meta tags:

* masterpiece

* best quality

* newest

* highres

* absurdres

* safe / sensitive / nsfw / explicit


Recommended composition and atmosphere tags:

* cinematic lighting

* rim lighting

* depth of field

* dramatic angle

* dutch angle

* off center

* dynamic pose

* wind

* glowing

* blurry background

* bokeh

* atmospheric

* dramatic shadows


Artist prompting:

Artist tags generally work better using:

@artistname


Prompt weighting is supported, though stronger weights may work better than standard SDXL-style prompting.

Example:

(chibi:2)

(red eyes:1.8)


Time period tags

Specific year

year 2025, year 2024, ...


Period

newest, recent, mid, early, old


Meta tags

highres, absurdres, anime screenshot, jpeg artifacts, official art, etc


Safety tags

safe, sensitive, nsfw, explicit


The model generally performs better with:

* clean prompts

* focused styles

* clear character descriptions

* strong lighting descriptions

* balanced atmosphere tags


Overloading prompts with excessive conflicting tags can reduce consistency. Simpler prompts often produce cleaner results.