<h3 id="vintage-woodland-camouflage-bdu-as-usual-this-time-for-boogu-image-(edit).-seriously-underrated-model-especially-for-multi-peoplecomplicated-prompts-it-is-possibly-the-best-local-(and-fast)-model-at-complicated-scenes-and-anatomy-in-it.">Vintage Woodland Camouflage BDU, as usual, This time for Boogu Image (Edit). Seriously underrated model, especially for multi people/complicated prompts; it is possibly THE best local (and fast) model at complicated scenes and anatomy in it.</h3>It's not perfect but there almost never arms or legs or hands wrong, even if it is a REALLY complex scene*General notes: I had much better experiences in complex scenes, using the base model (fp8 mostly) + turbo lora, than with the actual Boogu turbo model. But all work with the lora. Make sure to load the hotfix versions of model or lora, if you downloaded it very shortly after release. It fixes some issues. The image quality in terms of blur is not always perfect. Adding a 0.2 denoise ZiT upscaler run is a good idea (it is in the comfyui templates for example)..Use the lora at 0.8 as a starter. It creates the Bdu correctly and it is flexible enough to apply it even to crazyer things. Also too high and a lora can compromise the models good anatomy handling. You can prompt the nametags, "left nametag reads 'U.S Army' right nametag reads... ". It works fairly reliably for most visible uniforms.I've used intentionally very complicated prompts and images as example. to show what i mean. Some may venture a bit into kinky territory, and while Boogu 0.1 does NOT do NSWF, it does that quite well.. And - again - it rarely messes it up.*why Boogu It beats in that aspect al of them including Flux.2 Klein4b and 9b, even Flux.2 dev has more anatomy issues. Ideogram is too complicated, Ernie Image is crap at anatomy, HidreamO1 too blurry. Z-Image is better at anatomy but BAD with complex multi people prompts. It freaks out very quickly. So we land at Qwen Image(edit) as competittion. And that is a compliment.

US Woodland BDU (Vintage) Boogu Image

likely final. It does it correct but it remains flexible. Avoid setting it to 1.0. 0.8 seems to be a good sweetspot. It looks correct, it fits correctly to any anatomical challenge, you can prompt reliably the namtags text.

Files with SHA256: 7d4082d9915f02b514efc1b19fca69386cf33a6c1a0f0b30adda32ee58a5abcd