This is a model trained on 161 assorted kodachrome photo scans verified to be from the 1940s. The training data did not have any captions, I'm not sure if captions are necessary as the results are already good at a mere 640 steps.

It won't always produce period accurate results; the training dataset is small and limited, kodachrome film was very expensive in the 1940s and surviving images do not depict a great variety of settings. You might see electronic billboards, modern cars, LCD screens, anachronistic fashion pop up from time to time. This is normal, there's infinitely more high quality color images in the Anima base dataset than there are high quality color images from the 1940s.

Nevertheless, I am very pleased with the results. I may train more models on kodachromes from other time periods -- though it is not a fun process acquiring and sorting the images.