Doing video background replacement with H3 Inpainting + LTX Alpha Matte LoRA, and it's surprisingly good 🔥 > LTX 2.5 LoRA to extract matte > H3 Ref inpainting (+ Turbo LoRA) to generate a new background > Mask drop knob to reduce flickering and improve lighting match - drops the inpainting mask at step t I added the workflow to 🧨diffusers modular + built Spaces for it 👇

made an open source app version of photoshop relight with native Qwen Image 2.1 and it's kinda good 👀 > fast inference with Viggle / Pruna turbo LoRAs - I like pruna better for skin texture and realism, but viggle seems better at prompt following > no LoRA - quality is comparable, but a (slider) LoRA to control intensity, angles etc better would help with fine-grained control. My private relighting dataset needs an upgrade > added an intensity slider (when viggle is enabled) that blends the original with the relighting output - works surprisingly well app▶️ huggingface.co/spaces/linoyt…

This is native Qwen Image 2.1, no LoRA - move and resize objects seamlessly in an image🔥 I was planning to finetune on this task to showcase the new 🧨 diffusers LoRA support for Qwen 2.1 (go try it!), but while testing the base model I realized it's actually so good at it out of the box (pun intended haha) Try it ▶️ huggingface.co/spaces/linoyt… Train 🧨 github.com/huggingface/diffu…

I'm loving the character swap LoRA for H3, I think it might be even better than animate and inpainting huggingface.co/spaces/huggin…