Introducing anima_pencil-XL-v5.0.0_clear! Harnessing Natural Language Prompts for Crisp, Usable AI Art

happy_little_blazer_girl
  • anima_pencil-XL-v5.0.0_clear has been released
  • Supports natural language prompts
  • Features a charming round face and soft expression

Introduction

Hello everyone, this is Easygoing.

Today, I’m excited to introduce the release of anima_pencil-XL-v5.0.0_clear, a model that supports natural language input and outputs clearer, crisper anime illustrations.

anipen_clear_268
anima_pencil-XL-v5.0.0_clear

What is the anima_pencil-XL Model?

The anima_pencil-XL model is an SDXL-based anime model originally published by bluepen5805.

gantt
    title pencil-XL series
    dateFormat YYYY-MM-DD
    tickInterval 6month

    section blue_pencil-XL
        blue_pencil-XL-v0.0.1 : done, 2023-01-30, 2025-11-08
        blue_pencil-XL-v7.0.0 : done, 2024-06-23, 2025-11-08	

    section anima_pencil-XL
        anima_pencil-XL-v1.0.0 : done, 2024-01-11, 2025-11-08
        anima_pencil-XL-v5.0.0 : 2024-06-25, 2025-11-08	

    section pony_pencil-XL
        pony_pencil-XL-v1.0.0 : done, 2024-05-02, 2025-11-08
        pony_pencil-XL-v5.0.0 : done, 2024-06-30, 2025-11-08	

    section illustriious_pencil-XL
        illustrious_pencil-XL-v1.0.1 : done, 2024-10-21, 2025-11-08
        illustrious_pencil-XL-v5.0.0 : done, 2025-03-14, 2025-11-08	

    section noob_v_pencil-XL
        noob_v_pencil-XL-0.65s : done, 2024-11-27, 2025-11-08
        noob_v_pencil-XL-v3.0.0 : done, 2025-05-14, 2025-11-08	

	section 4nima_pencil-XL
        4nima_pencil-XL-v1.0.0 : done, 2025-01-30, 2025-11-08	

	section true_pencil-XL
        true_pencil-XL-v3.0.0 : done, 2025-07-01, 2025-11-08	

    section mellow_pencil-XL
        mellow_pencil-XL-v1.0.0 : done, 2025-11-05, 2025-11-08
		

blue_pencil-XL model is a complex model created by merging a vast number of models, ranging from realistic to anime styles, and is known for its wide range of expressive capabilities.

  • anima_pencil-XL ≈ blue_pencil-XL + Animagine-XL

The anima_pencil-XL model builds upon this by merging the Animagine-XL series and its derivatives, incorporating the best qualities of the Animagine-XL line.

Key Features of the anima_pencil-XL Model!

Let's compare the anima_pencil-XL model with the Animagine-XL model.

Upscale_Detailer_color_correct_187 Upscale_Detailer_color_correct_152
anima_pencil-XL-v5.0.0_clear | Animagine-XL_4.0_Opt_clear

Characteristics of the anima_pencil-XL Series:

  • Rounder, more childlike face (Childlike features)
  • Usable with natural language prompts
  • Bright face and overall illustration

Most anime models, including Animagine-XL and illustrious-XL, are trained heavily on tag-based datasets like danbooru, typically resulting in characters with smaller, sharper faces and narrow eyes. In contrast, the anima_pencil-XL model produces unique round faces with a charming appeal.

As anime models undergo repeated tag-based training, advanced models often become solely responsive to tag inputs.

However, the anima_pencil-XL model is designed to maximize prompt interpretation by rolling back the CLIP functionality, allowing it to utilize natural language prompts. This is particularly effective for expressing delicate emotions and a sense of softness.

Due to the high stability of CLIP in the anima_pencil-XL model, quality prompts and negative prompts are not strictly necessary.

Furthermore, the anima_pencil-XL model consistently depicts faces brightly, making it easy to create illustrations with a cheerful and luminous atmosphere.

Improvements in anima_pencil-XL-v5.0.0_clear

Now, let's examine the features of the newly fine-tuned anima_pencil-XL-v5.0.0_clear model.

The left side shows the output from the newly tuned anima_pencil-XL-v5.0.0_clear model, and the right side shows the output from the original anima_pencil-XL-v5.0.0 model.

Purple Witch

Comparison of anime illustrations of a witch in a purple outfit. Left: anima_pencil-XL-v5.0.0_clear , Right: Original v5.0.0
anima_pencil-XL-v5.0.0_clear | anima_pencil-XL-v5.0.0 (Original)

Girl in a Blazer

Comparison of anime illustrations of a girl wearing a blazer. Left: anima_pencil-XL-v5.0.0_clear Right: Original v5.0.0

Magnified View

Magnified comparison of the anime illustrations of a girl in a blazer. Left: anima_pencil-XL-v5.0.0_clear, Right: Original v5.0.0.

The anima_pencil-XL-v5.0.0_clear model has been adjusted for higher contrast and more vivid colors.

Looking at the magnified details of the eyes and clothing, the left image from the anima_pencil-XL-v.5.0.0_clear model shows clearer expression and more accurate details.

Other Examples!

Here are some other examples of what the model can do.

Pumpkin Land

An anime illustration of a female character resembling a witch with a purple witch's hat. Levitating candy cane and pumpkin. Generated by anima_pencil-XL-v5.0.0_clear.
Cheerful Vibe
an animated female character, resembling a witch, donning a purple witch's hat with golden embellishments, stands against a darkened backdrop filled with levitating items like a candy cane and a carved pumpkin. her eyes are closed, and she wears a matching purple and black attire with a matching hat featuring a smiling face. the overall mood is mysterious and enchanting.,

dutch angle, dynamic angle, close up, upper body, happy, smile, laugh, peaceful, wind, vivid, colorful, small breasts, mature, elegant, black background, burgundy, rouge, alizarin, royal purple, indigo, deep blue

Summer Festival (Ennichi)

An anime illustration of a young female character in a kimono with red flowers standing in a night market (ennichi). Lanterns and crowds. Generated by anima_pencil-XL-v5.0.0_clear.
Charming Round Face
a young female character in a kimono with a red flower in her hair stands in a traditional japanese street at night, surrounded by lanterns, and people, with a warm and serene ambiance.,

dutch angle, dynamic angle, close up, upper body, happy, smile, laugh, peaceful, wind, vivid, colorful, small breasts, mature, elegant, black background, burgundy, rouge, alizarin, royal purple, indigo, deep blue

Red and White Roses

An anime illustration of a silver-haired, blue-eyed woman holding a bouquet of white roses in a room with a city view. Generated by anima_pencil-XL-v5.0.0_clear.
Delicate Expression
an animated female character with silver hair and blue eyes holds a bouquet of white roses in a room with a city view, wearing a red and black outfit with a black choker, surrounded by a warm and cozy atmosphere with soft lighting and a framed picture on the wall.,

dutch angle, dynamic angle, close up, upper body, small breasts, mature, elegant, black background, burgundy, rouge, alizarin, royal purple, indigo, deep blue

Leopard

A close-up anime illustration of a leopard with yellow eyes. Detailed fur, dark background. Generated by anima_pencil-XL-v5.0.0_clear.
Compositional Strength inherited from Animagine-XL
a close - up of a leopard's face, with striking yellow eyes and detailed fur, is depicted against a dark background with a hint of blue and black colors. the leopard's fur is detailed, and its eyes are focused intently on the panther's face. the overall style is realistic and detailed, with a focus on the leopard's fur and the eyes.,

dutch angle, dynamic angle, close up, upper body, happy, smile, laugh, peaceful, wind, vivid, colorful, black background, burgundy, rouge, alizarin, royal purple, indigo, deep blue

Avian Person

A vivid anime illustration of an avian person/rooster. Dark background, red, blue, and purple talons. Generated by anima_pencil-XL-v5.0.0_clear.
Soft Color Expression
a vividly colored rooster stands prominently against a dark background, its vibrant plumage contrasting with the darker tones of the surrounding environment. its striking plumage stands out prominently, with its striking plumage accentuating its striking colors. the bird's talons are a mix of red, blue, and purple, adding a touch of surrealism to the scene.,

dutch angle, dynamic angle, close up, upper body, happy, smile, laugh, peaceful, wind, vivid, colorful, small breasts, mature, elegant, black background, burgundy, rouge, alizarin, royal purple, indigo, deep blue

The Recipe for anima_pencil-XL-v5.0.0_clear

The anima_pencil-XL series is subject to the Fair AI Public License 1.0-SD, which requires the sharing of the merge recipe when publishing derived models.

The Merge Recipe for anima_pencil-XL-v5.0.0_clear

Screenshot of the ComfyUI merge workflow for anima_pencil-XL-v5.0.0_clear. Showing UNET and VAE adjustments.

The anima_pencil-XL-v5.0.0_clear model features improvements to the UNET and VAE relative to the original anima_pencil-XL-v5.0.0.

  • UNET: Suppression of overly reactive layers (e.g., time_embed., label_embed.)
  • VAE: Replacement with an anime-specific VAE (SDXL Anime VAE Dec-only B3)

In the anima_pencil-XL-v5.0.0_clear model, the activity of overly reactive initial layers within the UNET has been suppressed.

Additionally, the VAE was replaced with an anime illustration-specific one to achieve more vivid color expression.

Is anima_pencil-XL-v5.0.0_clear's File Size Large?

The anima_pencil-XL-v5.0.0_clear model is slightly larger than a typical SDXL model.

  • anima_pencil-XL-v5.0.0_clear: 6.94 GB
  • Typical SDXL model: 6.78 GB

The reason for this is that the replaced VAE was merged in FP32 format.

VAE is Processed in BF16 Format

Although SDXL initially spread in the FP16 format, there were early issues where VAE processing was unstable, sometimes generating black images.

Consequently, both ComfyUI and Stable Diffusion webUI now perform VAE processing in BF16 format on compatible GPUs.

Sign Exponent Mantissa Significant Digits Precision
FP32 1 bit 8 bits 23 bits 6–7 digits 🌟
FP16 1 bit 5 bits 10 bits 3–4 digits
BF16 1 bit 8 bits 7 bits 3 digits
FP8(e4m3) 1 bit 4 bits 3 bits 1–2 digits ⚠️
FP8(e5m2) 1 bit 5 bits 2 bits 1–2 digits ⚠️
FP4 1 bit 2 bits 1 bit Less than 1 digit
  • FP32 → BF16: No degradation
  • FP16 → BF16: Slight degradation occurs

While the FP32 → BF16 conversion introduces no degradation, a slight loss of precision can occur with FP16 → BF16 conversion.

Anime illustration of a silver-haired woman in a black dress holding a bouquet of red and white roses. Example using FP32 VAE.

In practice, the difference in image quality is not noticeable to the naked eye. Since there is no change in VRAM usage or execution time, the FP32 format VAE was adopted for the anima_pencil-XL-v5.0.0_clear model.

CLIP-L for Automatic Prompt Generation!

The natural language prompts used in these examples were created by captioning the images using the comfy-cliption custom node with CLIP-L.

Input Image

Anime illustration of a purple-haired girl holding an orange cat. Example input for CLIP-L captioning.

CLIP-L Caption

A young female character with purple hair and blue eyes is holding a fluffy orange cat (with blue eyes), wearing a pink sweater and blue overalls, against a dark background with purple and blue streaks, conveying a sense of joy and curiosity.

While CLIP-L is contextually less accurate than Florence-2 (which we used previously) for captioning, it features a broader knowledge base, giving it a wider range of interpretation.

CLIP-L is capable of verbalizing hue and atmosphere, making it well-suited for generating illustrations with an artistic flair.

Furthermore, natural language prompts inherently contain more information than tag-based prompts, making them optimal for seeking variations and delicate expressions.

Summary: Try anima_pencil-XL-v5.0.0_clear

  • anima_pencil-XL-v5.0.0_clear has been released
  • Supports natural language prompts
  • Features a charming round face and soft expression

In this post, we introduced the anima_pencil-XL_v5.0.0_clear model.

Although the anima_pencil-XL series has been around for over a year and a half and may lack the sheer detail accuracy of the latest models, it remains a unique anime model for its ability to understand natural language prompts and render soft expressions.

An anime illustration of a blonde woman holding a blue-purple bouquet. Generated by anima_pencil-XL-v5.0.0_clear.

The anima_pencil-XL model is easy to use, and with this latest revision, I believe it has gained an expressive power comparable to the newest models. I encourage everyone to give it a try!

Thank you for reading all the way through!