IP-Adapter attention masking warning logs

[Honey-666]

diffusers/src/diffusers/models/attention_processor.py

Line 517 in d8d208a

if len(unused_kwargs) > 0:

diffusers/src/diffusers/models/attention_processor.py

Line 1202 in d8d208a

def __call__(

hi!
Is the call method of the AttnProcessor2_0 class missing the parameter ip-adapter_masks? When I use it, it will print a warning log cross_attention_kwargs ['ip_adapter_masks'] are not expected by AttnProcessor2_0 and will be ignored.

os:

diffusers==diffusers-0.27.0.dev0

this is my code:

import torch
from PIL import Image
from diffusers import StableDiffusionXLPipeline, AutoencoderKL, DPMSolverMultistepScheduler
from diffusers.image_processor import IPAdapterMaskProcessor
from transformers import CLIPVisionModelWithProjection

model_path = '../../../aidazuo/models/Stable-diffusion/sd_xl_base_1.0'
vae_path = '../../../aidazuo/models/VAE/sdxl-vae-fp16-fix'
ip_adapter_path = '../../../aidazuo/models/IP-Adapter'
ip_img_path1 = '../../../aidazuo/jupyter-script/test-img/ip_mask_girl1.png'
ip_img_path2 = '../../../aidazuo/jupyter-script/test-img/ip_mask_girl2.png'
mask_path1 = '../../../aidazuo/jupyter-script/test-img/ip_mask_mask1.png'
mask_path2 = '../../../aidazuo/jupyter-script/test-img/ip_mask_mask2.png'
ip_adapter_img1 = Image.open(ip_img_path1)
ip_adapter_img2 = Image.open(ip_img_path2)
mask_img1 = Image.open(mask_path1)
mask_img2 = Image.open(mask_path2)

processor = IPAdapterMaskProcessor()
masks = processor.preprocess([mask_img1, mask_img2], height=1024, width=1024)
# vae = AutoencoderKL.from_pretrained(vae_path, torch_dtype=torch.float16)

image_encoder = CLIPVisionModelWithProjection.from_pretrained(ip_adapter_path,
                                                              subfolder='models/image_encoder',
                                                              torch_dtype=torch.float16).to('cuda')

pipe = StableDiffusionXLPipeline.from_pretrained(
    model_path,
    safety_checker=None,
    variant="fp16",
    torch_dtype=torch.float16,
    image_encoder=image_encoder
).to("cuda")

adapter_file_lst = ["ip-adapter-plus-face_sdxl_vit-h.safetensors"] * 2
adapter_weight_lst = [0.7] * 2
adapter_img_lst = [ip_adapter_img1, ip_adapter_img2]

pipe.load_ip_adapter(ip_adapter_path, subfolder="sdxl_models", weight_name=adapter_file_lst)
pipe.set_ip_adapter_scale(adapter_weight_lst)
# pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)


images = pipe(
    prompt='2 girls',
    ip_adapter_image=adapter_img_lst,
    negative_prompt="monochrome, lowres, bad anatomy, worst quality, low quality",
    # num_inference_steps=30,
    num_images_per_prompt=1,
    width=1024,
    height=1024,
    cross_attention_kwargs={"ip_adapter_masks": masks}
).images
pipe.unload_ip_adapter()

for img in images:
    img.show()

[yiyixuxu]

It is not a bug but we need to find a better way to handle this argument so we won't spam your console 😭

[Honey-666]

It is not a bug but we need to find a better way to handle this argument so we won't spam your console 😭

hhh ok 👍

[LinchuanXuTheSEAAI]

Also encountered this warning when using StableDiffusionXLControlNetPipeline.
Does this means that the IP-Adapter attention masking feature is not enabled in the pipeline because they are ignored?
Thanks in advance!

[yiyixuxu]

ip adapter uses the default attention processor for self-attention and the IPAdapterAttnProcessor for cross-attention

the ip_adapter_mask is only used in IPAdapterAttnProcessor and ignored when it was passed to the default attention processor for self-attention, hence the warning. It is still enabled!

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[detkov]

Not stale.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[ghunkins]

Not stale and still visible in diffusers==0.28.2.

A simple fix would be to add a bypass exception to the logging for AttnProcessor2_0 when the key equals ip_adapter_masks.