Hi! I made a proof of concept LoRA for FLUX.3 Klein to manipulate text with an intuitive UI.
retained text
Hello everyone! I made a new LoRA for FLUX.3 Image Klein to manipulate text with bboxes. I think we all know that FLUX.3 Image Klein is not a very good model, and yes it is disappointing, but these bbox tools are unprecedented currently for local image editing models (at least until Ideogram V4.5, which i am excited for!). It really allows you to train brand new concepts. Here's the rundown of what the FontBox LoRA can do/improves upon FLUX.3 Image Klein's native abilities: Adjust the size of text; change what text says and specify left, right, or center alignment; move text to completely different places; bold or italicize text; Adjust the vertical scale of text; all in a custom node that I believe feels more intuitive (and looks nice!) to use than Comfy's native bbox editor. Boilerplate prompts are generated based on the bboxes you placed and the modes they have, meaning you don't have to fiddle around with prompts much! As this is a proof of concept LoRA, it is unstable in some generations. Temper your expectations, but this is a good start, and I will curate more data and train a better version. Please, let me know what you think. Thanks, and have fun! :3 Here is the huggingface link: [https://huggingface.co/lilylilith/FLUX.3\_Klein\_FontBox](https://huggingface.co/lilylilith/FLUX.3_Klein_FontBox) and here is the Github link for the custom node: [https://github.com/sillylilithhh/ComfyUI\_FontBox](https://github.com/sillylilithhh/ComfyUI_FontBox)
document fetched 2026-10-31 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xj7fwy/hi_i_made_a_proof_of_concept_lora_for_flux3_klein