Support Flux Klein peft (fal) lora format

[asomoza]

FAL trained loras use the default peft key naming, this PR just rename them to the diffusers naming.

Fixes #13162

Tested with this lora

code

import torch

from diffusers import Flux2KleinPipeline
from diffusers.utils import load_image


device = "cuda"
dtype = torch.bfloat16

pipe = Flux2KleinPipeline.from_pretrained("black-forest-labs/FLUX.2-klein-4b", torch_dtype=dtype)
pipe.load_lora_weights(
    "fal/flux-2-klein-4b-spritesheet-lora",
    weight_name="flux-spritesheet-lora.safetensors",
)
pipe.enable_model_cpu_offload()

prompt = "2x2 sprite sheet"
image = load_image(
    "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/edit/julian-hochgesang-1fDZ4BijtG0-unsplash.jpg"
)

image = pipe(
    prompt=prompt,
    image=image,
    height=1024,
    width=1024,
    guidance_scale=1.0,
    num_inference_steps=4,
    generator=torch.Generator(device=device).manual_seed(42),
).images[0]
image.save("flux-klein.png")

source image	result

Who can review?

@christopher5106 @sayakpaul

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[christopher5106]

Works for me but your tests seem to fail

[asomoza]

yeah but test is unrelated, it fails with wan vace.

[sayakpaul]

Sweet!