MageTrail - Microsoft's MageFlow 4B Full Finetune

ALL PREVIEW IMAGES HAVE COMFYUI METADATA.

---

I. Introduction

MageTrail is a Danbooru/E621 proof of concept Full-Finetune of Microsoft's MageFlow 4B T2I model, using a diversity maximized condensed 41k images dataset as a way to tune booru concept and tags based prompting + Illustration capabilities into the model without having to tune with the full booru dataset.

My second attempt at fine-tuning an image model on larger scale, this finetune aim to prove to the open source community on MageFlow 4B having good potential as an architecture for further finetuning.

~ While V0.1 is still very obviously undertrained and unstable (only 100 dollars spent, it's a minor miracle that it's learning this well), the model has shown great promise in quickly learning and adapting booru concept and tags to its knowledge base

~ The architecture behind MageFlow 4B shows good promise for further investment:

• Being 15-20% faster than NVIDIA Cosmos2/Anima on inference despite being 2 billion parameters larger
• Using MageVAE which perform better than QwenVAE on all usage, only behind the strongest open source VAE currently being Flux2VAE (which it was distilled from), also having the ability to slot in Flux2VAE for inference
• Having a decent Qwen 3 VL 4B Text Encoder
• 256-2048 pixels resolution native support
• Being fairly quick to learn and adapt to new knowledge without any knowledge forgetting

Future goal for the project: gather funding of 700~ dollars to finetune the model to 200 epoch for full convergence of booru concepts (V0.5) and then further small scale funding to finetune my 10k artist collection dataset into it. Any donation will help with achieving this goal, you can do so through:

Crypto

0x6a4bc748cd0bb9ced9a360eb0eb79f4f106614f8 (USDT - BEP20 Network)

12PPVYUeS1MerNp38Tpns5qXR6cmhu9tws (Bitcoin - BTC Network)

0x6a4bc748cd0bb9ced9a360eb0eb79f4f106614f8 (Ethereum - ERC20 Network)

FitfJAsxLUBuSgDJJaHgBXJpt1sMm5FzF1Tvf1SHW5Up (Solana - SOL network)

Please handle your money carefully and make sure the address you're sending to is correct.

Ko-fi

https://ko-fi.com/talanartvn

II. Model Details

Base model	Microsoft's MageFlow 4B
Method	Full Finetune
Trainer	My Diffusion-Pipe fork
Hardware	x8 H100 HBM3 80GB, courtesy of Banodoco grant
Total training time	8 hours (~64 H100 hours)
Total samples seen	~ 326656
Training resolutions	1024²

Training run

Version 0.1 (initial 20 epoch run → extended 10 epoch run)

Budget: 130~ dollars (25-30 lost due to experiments and mistakes)

Full config: Training and Dataset

• Learning rate: 7e-6
• LR scheduler: Warmup -> Constant -> REX to 0e-7
• Precision: Full BF16
• Optimizer: AdamW8bit with Kahan summation (to offset BF16 precision roundoff)
• Weight decay: 0.02
• Timestep sampling: Logit-Normal, shift 6, sigmoid scale 1.0

Additional training features

• Tag dropout: 10%
• Caption dropout: 5%
• Mixed captions at 25/25/25/25 ratio (tags only, NL only, tags-nl, nl-tags)
• Tag shuffle
• Caption shuffle
• Artist trigger attribution system

---

III. Recommended Settings

These are the settings used for the sample images above (ComfyUI):

• Shift: 5.0
• Steps: 30
• CFG: 5
• Sampler: euler_ancestral or er_sde
• Scheduler: simple or beta

These are just my usual settings and workflow — feel free to experiment.

Artist Trigger: This model use the Drawn by artistname trigger, if you want to use the model built-in artist tag, please always put one at the start of your prompt. (Currently V0.1 barely support any artist or characters though)

---

IV. Dataset

Booru-Essence-2026 41k images

Originally created by Lodestone Rock, the dataset was updated to 2026 tag standard and captioned with SOTA API captioners, see dataset repo for details. Model training, Dataset and Captioning tooling lives in the utils folder of the training repo.

---

VI. Notes from the Training Diary

Full training diary: MageTrail-diary

---

VII. License

This model is a Derivative of Microsoft's MageFlow and is distributed under the same MIT License as the base model, with no additional restrictions.

---

VIII. Acknowledgments

Beeg thanks to:

• Banodoco and their Discord — Their 88.77 dollar grant made this project possible, the biggest thanks to them
• Lodestone Rock — Creator of the original version of the dataset that this model is trained on
• Motimalu — Inspiration behind finetuning practices and configs
• Bluvoll — diffusion-pipe fork derived from to use for training, and general training advice
• Anzhc — general training advice
• Nruaif — diffusion-pipe fork derived from to use for training, and general dataset handling/training advice
• Astromahdi — jupyter workspace where I processed and store the dataset
• animetimm/DeepGHS — Danbooru tagging model
• RedRocket — E621 tagging model
• Format inspired by Motimalu's Kirazuri diary