Support for ref images RoPE index (omni-kontext)

[stduhpf]

Support for models like omni-kontext

How to use:

Download omni-kontext character LoRA and add it to your lora model directory.

example command:
sd.exe --diffusion-model models\flux1-kontext-dev-q5_1.gguf --clip_l models\clip\clip_l\clip_l.safetensors --t5xxl models\clip\t5\t5xxl_fp16.safetensors --vae models\vae\flux\ae.f16.gguf -W 1024 -H 512 -p "Add the character to the scene <lora:character_5000:1>" -r 'scene_02.png' -r 'boy_reference_512.png,2' --cfg-scale 1 --sampling-method euler --color --lora-model-dir models\loras\flux\kontext\

note the ,2 after the path to the second image. That's the desired RoPE index of the character's image (defaults to index 1).

Results:

Background	Character

	Without LoRA	With omni-kontext character LoRA
master (index 1 for both images, using 2D RoPE offsets)
PR (index 2 for character)

[leejet]

I think it's not necessary to manually assign a number to each ref image. Instead, we can use a single parameter, such as --increase-ref-index, to control whether to increment the index.

[stduhpf]

I like having the option to set it manually, but you're right, a --increase-ref-index would be more convenient.