This workflow is a test of Kontext for virtual try-on, and it has several strict requirements for the input images. The first input image must include both the head and the body — it cannot be just a headshot. The second image, if it includes clothing, must not contain a head. Also, the positions of the first and second input images must not be swapped.

The clothing in the first image needs to be removed using a mask. Alternatively, you can use the "not-flux-kontext-dev-clothes-remover" LoRA to remove the clothing.

https://huggingface.co/llama-anon/not-flux-kontext-dev-clothes-remover trigger "remove clothes"