Other PixAI Models

Tsubaki.3 is PixAI's current flagship. This page puts the other models that remain fully available side by side — a comparison table to answer "Which model fits my needs?", followed by an archive of each model's details. New to model architectures? Start with How AI Image Generation Works for the fundamentals; for parameter and tool support, see the comparison tables at the bottom of this page, or the Model Parameters page.

Common Models at a Glance

PixAI provides multiple models trained for anime, and also lets you upload your own local models to the community. Here's an overview of six commonly used official models.

In most cases, we recommend the three models below. Whether you're new or experienced, these three cover most of what you'll want to create:

Tsubaki.2

Tsubaki.2 is a DiT model released by PixAI in March 2026.

Why Tsubaki.2?

Compared to traditional SDXL models, Tsubaki.2 is a genuine leap forward:

Exclusive Features

One-Click Style Presets

A Tsubaki.2-only feature — pick a style preset on the right side of the generation panel and switch art styles instantly.

The four images below use exactly the same prompt (Standard mode); only the style preset changed:

Four Generation Modes

Tsubaki.2 offers four generation modes. They're essentially four different models, each with its own texture and feel — not just a speed-vs-quality slider. The price difference reflects how much compute each mode uses. The four images below use exactly the same prompt; the differences come from the modes themselves.

Haruka V2

Haruka v2 is PixAI's classic anime SDXL model. Its clean anime art style paired with refined hands and eyes has made it a long-time favorite. If you prefer traditional tag-based prompts or want to layer styles freely with the rich SDXL LoRA ecosystem, Haruka v2 is a great fit.

Key Strengths

• Major improvements to hand rendering — handles common AI hand issues much better, with accurate gestures and correctly held objects
• More expressive eyes — refined pupil highlights, eyelash layering, and eye lighting make characters come alive
• Optimized specifically for anime style — better hair sheen transitions and smoother fabric folds for cleaner, more polished characters
• SDXL LoRA compatible — full support for the SDXL LoRA ecosystem, so you can freely stack artist styles and character LoRAs

Reference Pro

PixAI Reference Pro is PixAI's multi-image editing model. Upload one or more reference images, then describe the changes you want in plain text — the AI parses the characters, poses, outfits, style, and composition in your images and transforms them based on your instructions.

Core Features

• Multi-image input — upload up to 10 images as references
• High-resolution output — supports 2K and 4K
• Flexible aspect ratios — 16:9, 9:16, 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4
• Visual understanding — automatically parses characters, poses, outfits, style, and composition

Editing Example

Upload a reference image and tell Reference Pro what to change in a single sentence. The example below uses just one sentence to turn a standing pose into a sleeping pose, while keeping the character's looks, outfit, and overall mood intact. The right image below is the result of applying "Change it to a sleeping Mio" to the left image.

Parameter Differences Between Models

Different model architectures expose slightly different settings on the generation panel. The table below compares the four main PixAI models — Tsubaki.2, Tsubaki.1, Haruka v2, and Reference Pro.

Tool Differences Between Models

Different model types also expose slightly different tools. PixAI tools generally fall into two groups: pre-generation tools (which add effects during generation) and post-generation tools (which create variations from already-generated images).

For details on each tool, see the Tools section.

Pre-generation Tools

Post-generation Tools