[debug] FluxControlImg2ImgPipeline 和 FluxControlImg2ImgPipeline 运行报错：

[whisky-12]

import paddle

from ppdiffusers import FluxPipeline
from ppdiffusers import ControlNetModel,,FluxControlPipeline,FluxControlImg2ImgPipeline,AutoencoderKL
from ppdiffusers.utils import load_image


pipe = FluxControlPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev",paddle_dtype=paddle.float16, low_cpu_mem_usage=True, map_location="cpu",
)


image = load_image(
    "robot.png"
)
control_image=load_image(
    "多啦A梦.jpeg"
)

# 运行推理
prompt = "a photo of a cat robot"
output = pipe(
    prompt=prompt,
    control_image=control_image,
    num_inference_steps=50,
)


# 或者 

pipe = FluxControlImg2ImgPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev",paddle_dtype=paddle.float16, low_cpu_mem_usage=True, map_location="cpu",
)


# 运行推理
prompt = "a photo of a cat robot"
output = pipe(
    prompt=prompt,
    control_image=control_image,
    image=image,
    num_inference_steps=50,
)

报错：

ValueError: (InvalidArgument) The 'shape' in ReshapeOp is invalid. The input tensor X'size must be equal to the capacity of 'shape'. But received X's shape = [1, 16, 128, 128], X's size = 262144, 'shape' is [1, 8, 64, 2, 64, 2], the capacity of 'shape' is 131072.
[Hint: Expected capacity == in_size, but received capacity:131072 != in_size:262144.] (at ../paddle/phi/infermeta/unary.cc:2246)

同时：

# 加载vue模型
vae = AutoencoderKL.from_single_file("models/flux-vae.safetensors")



# controlnet模型我们选用lineart类别的来提取线条
controlnet = ControlNetModel.from_pretrained("lllyasviel/control_v11p_sd15_lineart", paddle_dtype=paddle.float16)

上述加载flux-vae会报错，希望官方可以给出支持接在flux vae 模型地址，并且controlnet不能适配FluxControl管道，也需要给出优化解决方案