I have the following Python code:

FONT = cv2.FONT_HERSHEY_SIMPLEX
FONT_SCALE = 1.0
FONT_THICKNESS = 2
bg_color = (255, 255, 255)
label_color = (0, 0, 0)

label = 'Propaganda'
label_width, label_height = cv2.getTextSize(label, FONT, FONT_SCALE, FONT_THICKNESS)[0]
label_patch = np.zeros((label_height, label_width, 3), np.uint8)
label_patch[:,:] = bg_color

I create a new blank image, with the size returned by getTextSize, and then, I add the text at the bottom-left point, according to the docs, which is x = 0, y = (height - 1) and with the same font, scale and thickness parameters used for getTextSize

cv2.putText(label_patch, label, (0, label_height - 1), FONT, FONT_SCALE, label_color, FONT_THICKNESS)

But when I use imshow or imwrite on the image label_patch, this is the results I get:

It can easily be seen that lowercase p and lowercase g are cut in the middle, such that g and a cannot even be distinguished. How can I make OpenCV's getTextSize return the correct size, and how can I make OpenCV's putText start drawing the text from the actual lowest point?

Found the solution to my problem, so sharing it here.

It turns out that there is another parameter that getTextSize returns, which is the baseline. It should have been taken into account when creating the box: the box height should be label_height + baseline:

(label_width, label_height), baseline = cv2.getTextSize(label, FONT, FONT_SCALE, FONT_THICKNESS)
label_patch = np.zeros((label_height + baseline, label_width, 3), np.uint8)
label_patch[:,:] = bg_color

Now, adding the text at the same point as before, which means that the baseline pixels will remain below (the point has actually moved one pixel down, as advised by @api55):

cv2.putText(label_patch, label, (0, label_height), FONT, FONT_SCALE, label_color, FONT_THICKNESS)

And the result: