Proper way to return tensor from node

[WASasquatch]

I am having issues converting a numpy to a proper tensor that doesn't result in 768 individual slices on the Y axis x.x

I am adding a MiDaS depth node to my custom nodes, but can't get a workable image for other nodes. I can save out the image just fine from PIL (below).

Example:

        depth = prediction.cpu().numpy()
        depth = (depth * 255 / (np.max(depth)+1)).astype('uint8')
        depth_image = Image.fromarray(depth)
        depth_image.save('TEST_DEPTH.png')
        tensor = torch.from_numpy( depth )
        tensors = ( tensor, )

[comfyanonymous]

See the LoadImage node: https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py#L863

[WASasquatch]

Oh gawd! Thank you! I was looking at like every other node (decoders and stuff) but that, which is silly in hindsight. Lol

[wwdok]

But sometime the tensor is not image, it can not be visualized simply like image

[WASasquatch]

You'd want to give a custom type to a tesnor that is not a image.