MageTrail

MageTrail

Generates illustrations from booru-style tag prompts as a 2.8B-parameter fine-tune of the MageFlow 4B image model. Suits hobbyists and researchers who use ComfyUI and want tag-based illustration.

Made by RicemanT on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: MageTrail is a text-to-image model by RicemanT, a full fine-tune of the MageFlow 4B image model with a custom 2.8B-parameter architecture. It was tuned on a condensed set of 41 thousand images from booru sites so that it follows tag-style prompts and draws illustrations. The author calls it a proof of concept.

Who it suits: Hobbyists and researchers who make illustrations from tag prompts and want to test how MageFlow behaves after fine-tuning, under the MIT license.

Industry | Who uses it | How they use it | Example
Creative work | Illustrators | Write tag-style prompts to draw illustrations | Generate a character sketch from tags
Software | Developers | Load the model in diffusers or ComfyUI | Test it in a ComfyUI workflow
Research | ML researchers | Study fine-tuning a small image architecture on a small dataset | Compare versions of the model
Community | Model tuners | Build on the author's recipe for further tuning | Try your own fine-tune on the same base

Skip it if you want photo-style images or a documented, finished model, since the card calls this an early proof of concept and the training set comes from booru sites.

Where to use it: The card shows its previews with ComfyUI metadata, and the model loads through diffusers.

• ComfyUI, with a custom node that the card's generation section links
• Python with the diffusers library
• Windows, macOS or Linux, with Python for the download

Use case | Who | What they get back
Tag-based illustration | Illustrators | Images drawn from booru-style tags
Artist-style tests | Researchers | Images that show how the model learns styles
Base for fine-tuning | Model tuners | A starting point already adapted to illustration

How to set it up: You need Python, a GPU-capable setup and the custom node.

• Python with pip
• ComfyUI or the diffusers library
• The custom node named in the card's generation recommendation section

1. Install the Hugging Face hub tools with pip.
2. Download the model files with the hf command.
3. Get the custom node that the card points to.
4. Load the model in ComfyUI, or in diffusers.

pip install -U huggingface_hub
hf download RicemanT/MageTrail

How to use it: Write a prompt made of booru-style tags and run generation in ComfyUI with the custom node. The card text shown here ends before its generation recommendations, so read that section for settings.

Example: You type a short list of tags that describe a character and a scene, and run the workflow. You get an illustration drawn from those tags, and the preview images on the card show the style.

For newcomers and experts: If you are new to image models, start with one ComfyUI workflow from a preview image, since the card says its preview images carry ComfyUI metadata. Change a few tags and watch how the picture changes.

If you know the field, the card says the base architecture runs 15 to 20 percent faster at inference than NVIDIA Cosmos2 and Anima, despite being 2 billion parameters larger. It uses a Qwen 3 VL 4B text encoder and works natively from 256 to 2048 pixels. From V0.3 it is compatible with Flux2VAE.

Download the files with hf
Add the custom node in ComfyUI
Write a prompt of booru-style tags
Run generation
Review the illustration

Good to know: The card describes versions V0.1 to V0.4, where V0.4 was a small test on 100 artists for validating artist style learning. The 41 thousand image set is described as having outgrown its limits.

• License: MIT
• Cost to run: none named, apart from your own hardware
• Data it reaches: the model files on your computer, and nothing else the card names
• Not for: photo-style work, since the card describes tuning for booru concepts, tags and illustration

On Hugging Face, 13 November 2026

• Task: text to image
• Library: diffusers
• Downloads in the last 30 days: 134
• Likes: 49
• License: mit
• Based on: mage-flow-community/Mage-Flow
• Fine-tunes: 1
• Tags: diffusers, anime, furry, danbooru, e621, finetune, mage-flow, text-to-image

Links

• Model: https://huggingface.co/RicemanT/MageTrail
• Files: https://huggingface.co/RicemanT/MageTrail/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download RicemanT/MageTrail

Then follow the usage in its model card. Nothing in this kit runs by itself.