Stable Diffusion model clearing

It would be nice to have a function that unloads modules from memory, otherwise when repeatedly using the StableDiffusion class, RAM overflow occurs

import keras_cv

import matplotlib.pyplot as plt


def plot_images(images):
    plt.figure(figsize=(20, 20))
    for i in range(len(images)):
        plt.subplot(1, len(images), i + 1)
        plt.imshow(images[i])
        plt.axis("off")

    plt.show()

for i in range(2):
    model = keras_cv.models.StableDiffusion(img_width=512 // (i + 1), img_height=512 // (i + 1))
    images = model.text_to_image("photograph of an astronaut riding a horse", batch_size=2)
    plot_images(images)

# Some other stuff 

[freedomtan]

move the line

model = keras_cv.models.StableDiffusion(img_width=512, img_height=512)

out of the loop. You only need to construct the pipeline once.

[Neizvestnyj]

or if you do need to clear the model, simply del model would work?

No, I try different methods, but RAM is overflow. I know that model optimized for 512*512, but it is not unloaded from memory after execution, which is why 4GB of memory is just so full.

Would you mind trying this instead? https://stackoverflow.com/questions/52133347/how-can-i-clear-a-model-created-with-keras-and-tensorflowas-backend

Yes, I’ve already tried it, it doesn’t help, 2 starts already takes about 8 GB of RAM