What can Boogu-Image-0.1 do?

Boogu-Image-0.1 generates text-to-image outputs at 1K, 1.5K, and 2K resolution and edits existing images from text instructions. The release ships four checkpoints — Base, Turbo (a 4-step distilled variant), Edit, and Edit-Turbo — plus FP8 quantized versions for lower-VRAM inference.

Where can I download Boogu-Image-0.1?

Boogu-Image-0.1 weights live on the Boogu organization on HuggingFace (huggingface.co/Boogu). Inference code, install scripts, and demo scripts sit at github.com/Boogu-Project/Boogu-Image under Apache-2.0. The arXiv paper is at 2607.13125.

What hardware runs Boogu-Image-0.1?

Boogu-Image-0.1 runs on a single 12 GB CUDA GPU with sequential CPU offload; 24 GB or more is recommended for full performance. The team tested on CUDA 12.6 and PyTorch 2.7.1; FP8 quantized weights are provided for tighter memory budgets.

Is Boogu-Image-0.1 commercial-safe?

Yes — Boogu-Image-0.1 ships weights, code, and training recipes under the Apache-2.0 license, which permits commercial use and modification with standard attribution. Users are still responsible for content licensing of any images they generate.

Boogu Project · 2026-07-14 · major

Boogu-Image-0.1 — 10B open image model trained on 208M images for ~$400K

Boogu-Image-0.1 is an Apache-2.0 open 10B image generation model. Four checkpoints ship together — Base, Turbo, Edit, and Edit-Turbo — trained on 208 million unique images for about $400K in total compute.

Boogu-Image-0.1 GitHub repository header for the Boogu Project image generation model

Apache-2.0 10B image generation model reports near closed-source quality after training on 208M images for about $400K in compute.

Quick facts

Maker	Boogu Project
License	Apache-2.0
Parameters	10B
Variants	Base, Turbo, Edit, Edit-Turbo
Resolutions	1K, 1.5K, 2K
Training data	208M unique images
Training cost	~$400K

What is it?

Boogu-Image-0.1 is a 10B open-weights image generation and editing model released July 14 under Apache-2.0. Four checkpoints ship together — Base for text-to-image, Turbo as a 4-step distilled variant, Edit for instruction-based image editing, and Edit-Turbo. FP8 quantized versions run on a single 12 GB GPU with CPU offload.

How does it work?

The Boogu Project trained the model on 208 million unique images for roughly $400K in total compute — an order of magnitude less data than typical frontier image models. Supported outputs cover 1K, 1.5K, and 2K resolution across nine aspect ratios, with bilingual text rendering as a headline feature. Text guidance runs at scale 4.0 with 25–50 inference steps in the recommended profile.

Why does it matter?

Frontier-quality image generation with weights, code, and training recipes under Apache-2.0 lets teams fine-tune or self-host without a closed-source license or high per-call price. Boogu-Image-0.1 also documents a cheap training recipe: the paper walks through data curation and pipeline tricks that got the team to competitive quality on a $400K budget.

Who is it for?

open-source image-gen researchers and teams building editing pipelines

Frequently asked questions

What can Boogu-Image-0.1 do?: Boogu-Image-0.1 generates text-to-image outputs at 1K, 1.5K, and 2K resolution and edits existing images from text instructions. The release ships four checkpoints — Base, Turbo (a 4-step distilled variant), Edit, and Edit-Turbo — plus FP8 quantized versions for lower-VRAM inference.
Where can I download Boogu-Image-0.1?: Boogu-Image-0.1 weights live on the Boogu organization on HuggingFace (huggingface.co/Boogu). Inference code, install scripts, and demo scripts sit at github.com/Boogu-Project/Boogu-Image under Apache-2.0. The arXiv paper is at 2607.13125.
How much did it cost to train Boogu-Image-0.1?: The Boogu Project reports about $400K in total training compute on 208 million unique images — roughly an order of magnitude less data than typical frontier image models use. The paper walks through the data-curation and training-pipeline choices that made that budget work.
What hardware runs Boogu-Image-0.1?: Boogu-Image-0.1 runs on a single 12 GB CUDA GPU with sequential CPU offload; 24 GB or more is recommended for full performance. The team tested on CUDA 12.6 and PyTorch 2.7.1; FP8 quantized weights are provided for tighter memory budgets.
Is Boogu-Image-0.1 commercial-safe?: Yes — Boogu-Image-0.1 ships weights, code, and training recipes under the Apache-2.0 license, which permits commercial use and modification with standard attribution. Users are still responsible for content licensing of any images they generate.

Try it

git clone https://github.com/Boogu-Project/Boogu-Image