Intro

Another real open weight (Apache 2.0) model released (2026-06-16). That I believe worth some attention. Model family got Base Turbo and Edit.

There are also demo sites to generate some images for test. for free.

Also Available on CivitAi for Generation

2714686 boogu-image-turbo ⚡8 BUZZ + tips (for 4 STEPS 1024px)
2714299 boogu-image-base ⚡50 BUZZ + tips (for 50 STEPS 1024px)
2714541 boogu-image-edit ⚡120 BUZZ + tips (for 50 STEPS 1024px)

Since model is not yet merged to diffusers and therefore it still not added to SD.Next I am bit limited with testing it. But some generation done using CivitAi

Image Quality

Do not know what to start from. I definitely not the GOAT. It makes mistakes. Images not always perfect but it is not bad either.

Let's see the Ai Art 30 set

Ai Art Series

So do we got some extra fingers? Yes. Sometimes. But not like everytime.

hands through the glass. yes. also happens

Relatively good with emotions

different type of smiles

let's compare

A photorealistic cinematic scene captured from a slightly low three-quarter angle inside a softly lit apartment hallway. A woman gently pulls her husband's tie toward her with a playful yet tender smile, stopping him just before he leaves for work. His expression mixes surprise, affection, and quiet admiration. Warm morning sunlight streams through a nearby window, creating soft rim lighting and subtle lens flare across polished wooden floors. Their faces remain close while the background falls into creamy bokeh. Natural body language, realistic fabric textures, delicate skin details, elegant color grading, intimate storytelling, authentic emotions, high dynamic range, shallow depth of field, aesthetically pleasing composition, refined editorial photography style.

might be a bit over exaggerated, but some emotions captured)

and some other models

Illustrative at first glance looks good

let's check also test v3

it is some mistakes also spotted

and also, some defaulting to illustrative stile

but in general, it is ok

like 1970 N.Y. Street

old gas station

pixel art

yes, it is important) check if model know Oslo Opera House. It knows

foreshortening is also mostly ok

and that complex surreal art is not pushing model into the darkness mess

Base is not in scope of this article, but it is usable. Some models have base only for training that is generate some trash content, In Boogu we can generate with it.

Oh.. wait.. an ASCII art (it is in test 1 and 2, but let's execute it now:

Turbo

Base

more like extreme pixel art, but at least it trying to.

Art Styles

from test v1 results . yes you can add a line like "cubism art" at the end of the prompt an get it.

Empty prompts

From test v0 results. That is kind of test that I believe should give overview of "what is model should be good at". Boogu results are - vegetables.

Test edit

middle painting replaced with scissors abstract art

Resources needed

Boogu 10B (Total Parameters is 18.5B). It consumes about 48 GB of "unified" (v)RAM in bfloat16. Similar models Z Image Turbo 6B (Total parameters: 10.2B), FLUX.2 Klein 9B (Total parameters: 17.3B), ERNIE Image Turbo 8B (Total parameters: 15.3B), KREA 2 Turbo 13B (Total parameters: 17.1B). So it a bit bigger. But what about time to generate?