How can i change generate image size?

[s-show]

I am using Stable Diffusion in this branch to generate images, but I get an error when I try to generate a 768x512 size image.

The code to generate the image is as follows

prompt = "1girl, \
blue eyes, \
cat ears, \
full body, \
looking at viewer, \
school uniform, \
simple background, \
silver long hair, \
under rim eyewear, \
white background"

image = pipe(prompt, height=768, width=512, num_inference_steps=50, guidance_scale=7.5, eta=0.0, execution_provider="DmlExecutionProvider")["sample"][0]

Running this code returns the following error

Traceback (most recent call last):
  File "C:\Users\hoge\stable-diffusion-dml-branch\dml_onnx.py", line 261, in <module>
    image = pipe(prompt, height=768, width=512, num_inference_steps=50, guidance_scale=7.5, eta=0.0, execution_provider="DmlExecutionProvider")["sample"][0]
  File "C:\Users\hoge\AppData\Roaming\Python\Python39\site-packages\torch\autograd\grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "C:\Users\hoge\stable-diffusion-dml-branch\dml_onnx.py", line 171, in __call__
    noise_pred = unet_sess.run(None, inp)[0]
  File "C:\Users\hoge\AppData\Roaming\Python\Python39\site-packages\onnxruntime\capi\onnxruntime_inference_collection.py", line 200, in run
    return self._sess.run(output_names, input_feed, run_options)
onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Got invalid dimensions for input: latent_model_input for the following indices
 index: 2 Got: 96 Expected: 64
 Please fix either the inputs or the model.

I think it conflicts with the setting of the example code in examples/inference/dml_onnx.py, but due to my lack of knowledge I cannot determine the cause.

Please let me know how to generate a 768x512 sized image with Stable Diffusion in this branch.

[zz1014449672]

https://www.youtube.com/watch?v=Z9PFh2Us3vk
This might help you.

[Patometro06]

You need to rebuild the data base to the proportion you want, but at more resolution you need more VRAM, see the steps to do it here #10 (works to higher sizes too)

PD: You need to calculate based in any divible number by 64, not by 8 as the comments in the source code.