[Feature] the new flux/bonsai model

[dfgdrfgyhfgfvtg]

Feature Summary

bonsai addition to te reppo

Detailed Description

is there any chance you can add the new light weight flux/bonsai model because its good quality to for the computation and time of generation

Alternatives you considered

No response

Additional context

No response

[Green-Sky]

bonsai-image is just flux2 klein 4b.

I made compatible ggufs for the binary release here https://huggingface.co/Green-Sky/bonsai-image-binary-4B-GGUF

For ternay there is missing quant support in ggml/llama.cpp. See #331

[dfgdrfgyhfgfvtg]

so i just have to use that and it will work?

[dfgdrfgyhfgfvtg]

im really sorry i just want to use it so please any help is appreciated

[Green-Sky]

Just check the sd.cpp docs on how to use flux2 klein 4b, and sub in this gguf as the diffusion model.
https://huggingface.co/Green-Sky/bonsai-image-binary-4B-GGUF/blob/main/bonsai_image_4b-mod_q8_0-q1_0.gguf

[dfgdrfgyhfgfvtg]

im really sorry for the bother but im new to tis so even if i read it i cant really do anytthing

[Disonantemus]

I was able to generate images using the binary model from @Green-Sky and the last vulkan from releases, like this:

$ ./sd/sd-cli --diffusion-model bonsai_image_4b-q1_0.gguf --llm Qwen3-4B-UD-IQ2_XXS.gguf --steps 6 --vae full_encoder_small_decoder.safetensors --cfg-scale 1.0 -p "a lovely cat" -H 256 -W 256 -o out.png

Working as expected!

---

Then, I tried a simple edit, with this:

$ magick logo: -crop 480x480+80+0 -resize x512 in.png
$ ./sd/sd-cli --diffusion-model bonsai_image_4b-q1_0.gguf --llm Qwen3-4B-UD-IQ2_XXS.gguf --steps 6 --vae full_encoder_small_decoder.safetensors --cfg-scale 1.0 -r in.png -p "Replace the background into a forest" -H 512 -W 512 -o out.png && imv out.png

in.png	out.png

• Is using the reference (EDIT mode)
• Is generating the background
• But, is changing the whole reference, should change only the background

Questions

• I'm doing something wrong?
• Is possible to use other smaller Qwen3 models, like: 0.6B, 1.7B; or maybe Qwen3.5 models?

System

distro: Arch Linux x86_64
kernel: 6.18.25-1-lts
cpu: Intel i7-4790 (8) @ 3.600GHz
gpu: NVIDIA GeForce GTX 1660 SUPER (6 GB VRAM)