About CagliostroLab Projects Philosophy Connect
Core Member @ Cagliostro Research Lab

Engineering AI Models, Empowering Creators

Indonesian AI developer bridging high-fidelity generalist models with stylized, aesthetic-focused checkpoints.

Explore My Work Get in Touch
KayfaHaarukku Profile
About Me

Hi, I'm Kayfa

AI Developer & Open-Source Contributor

I'm an Indonesian AI developer known for creating distinctive anime-style Stable Diffusion models. My work is characterized by a strong open-source philosophy and detailed community documentation.

I specialize in crafting models with distinct artistic "flavors"—specifically soft pastel palettes and bold, thick line art that stand apart from standard AI generations.

Beyond AI development, I'm a passionate VTuber enthusiast since 2020. I'm currently working as a project staff at PT Jati Karya Bahagia for BloomsWork, a co-working, virtual office, meeting room, and office room provider based in Bogor, Indonesia.

Previously, I worked as a video editor for Fanfare Ch. until March 2024, and was an intern at PT Tera Multi Wahana until February 2025. I'm also part of the Civitai Creators Program since February 2024.

5+ Model Series
2023 Started Dev
100% Open Source
Professional Role

Cagliostro Research Lab

🔬

Main Model Trainer

Animagine XL Team

I'm a key figure at Cagliostro Research Lab, progressing from dataset curator in version 3.1 to the main trainer for their flagship 4.0 release. The lab is known for creating state-of-the-art anime generation models used by the global AI art community.

Animagine XL 4.0 — Main Trainer & Co-Developer

Serving as the main model trainer for Animagine XL 4.0, working alongside Linaqruf and Raelina who assist with other aspects. This version represents a major technical leap, retrained from SDXL root to dramatically improve anatomy and generation stability. Released early 2025.

Animagine XL 3.1 — Dataset Curator

Served as a dataset curator for version 3.1, helping to refine the model's ability to handle complex anime tags and aesthetic evaluation, laying important groundwork for the 4.0 release.

Team Projects

CagliostroLab Collaborations

🔥

Animagine XL 4.0

Main Trainer

The flagship release from Cagliostro Research Lab. As the main model trainer, I led the training process working alongside Linaqruf and Raelina. This state-of-the-art generalist anime model features dramatically improved anatomy and stability.

CagliostroLab SDXL Retrain Improved Anatomy
View on Civitai
🎭

AnimaTensor Series

Main Trainer

A series of high-quality anime-style image generation models developed with CagliostroLab. As main trainer, I focused on creating models optimized for Tensor.Art's platform with exceptional aesthetic quality.

CagliostroLab Tensor.Art High Quality
View on Tensor.Art
🌅

Moescape Diffusion 2.0

Assistant Trainer

A collaborative project with CagliostroLab focused on landscape and environment generation with anime aesthetics. Serving as assistant trainer, I helped refine the model's ability to generate beautiful scenic compositions.

CagliostroLab Landscape Focus Scenic Generation
View on Moescape
Personal Projects

Model Series & Creations

⚡

AkashicPulse EQVAE

Current Flagship Model

The latest evolution of AkashicPulse, utilizing Anzhc's EQVAE in place of the standard VAE for improved color accuracy and generation fidelity. Built on Illustrious via RouWei with the same careful training approach the series is known for.

Illustrious Base EQVAE Enhanced Colors
View on Civitai
💫

AkashicPulse

Previous Flagship Model

A finetune based on RouWei, an Illustrious-based model. Designed to deliver stunning results superior to the competition through careful training and special treatment to fix model quirks.

Illustrious Base RouWei Finetune Superior Quality
View on Civitai
🌸

UrangDiffusion

SDXL Community Model

The successor to AingDiffusion XL. "Urang" means "We" in Sundanese, signaling a shift toward a more community-focused model while maintaining high-fidelity anime generation.

SDXL Architecture Character-Focused High Fidelity
View on Civitai
🎨

AingDiffusion

The Model That Started It All

The SD 1.5 project that established my reputation in Stable Diffusion. A merge of anime models capable of generating high-quality images with my personal taste. The name "Aing" is informal Sundanese for "I" or "My."

SD 1.5 Personal Taste Legacy Model
View on Civitai
🌟

AingDiffusion XL

Animagine on Steroids

The SDXL evolution of the original AingDiffusion. Known for its distinct pastel color palette, soft lighting, and gentle gradients. I describe it as "Animagine on steroids" for specific tastes.

SDXL Pastel Aesthetic Soft Lighting
View on Civitai
✨

KayWaii XL

Bold & Cute Focus

A finetune based on PonyDiffusion v6, focusing on "thick coating" with bold, distinct line art. Popular among users who prefer sharper outlines and more defined character features.

Pony v6 Base Thick Line Art Bold Outlines
View on Civitai
Software & Tools

Developer Tools

🤖 AiChanBot

An intelligent Discord bot featuring intelligent model selection and advanced capabilities.

  • Dynamic LLM switching based on query complexity
  • Image processing and analysis capabilities
  • "Thinking mode" for complex reasoning tasks
  • Seamless Discord integration
View on GitHub
🤖
CogNotez

CogNotez

An offline-first, privacy-focused note-taking application that leverages local Large Language Models for intelligent features.

  • Offline AI-powered note-taking with local LLMs
  • Privacy-focused with end-to-end encryption
  • Google Drive sync support
  • Multi-language support (English, Spanish, Indonesian, Japanese, Javanese)
  • Rich media support with images and videos
View on GitHub
Philosophy

Open Source, Always Free

💝

"Always Free"

Despite the high GPU costs of training SDXL models, I've consistently stated that my personal checkpoints will remain free for the community. No paywalls—just passion for accessible AI art.

📚

Community Documentation

Every model release comes with extensive User Guides, educating users on specific settings like Clip Skip, VAEs, and Samplers to ensure they can replicate the developer's intended results.

Fun Fact

The Names Behind the Models

"Aing"
Sundanese (Informal)

Means "I" or "My"—reflecting that AingDiffusion is tuned to my personal taste and artistic vision.

"Urang"
Sundanese (Polite)

Means "We" or "Us"—signaling a shift toward community-centric development. A model for all of us.

Connect

Let's Connect

Find my work, follow updates, or just say hello.

Support My Work

If you enjoy my models and want to help cover GPU training costs, consider buying me a coffee!

☕ Support on Ko-fi