AI filmmaking is still in its infancy, but it is developing very rapidly. As a historian, I would like to be able to use AI to bring the past to life, and video is one way to do this. However, at the moment, it is still very difficult.
To make video clips with AI, you have to either use a text prompt or develop a video from an image.
Text prompts do not work well for historical periods because the AI has not been trained on enough data to know what many people and places looked like in the past. Meanwhile, to develop a video clip from an image, you have to have a good image, and for many historical periods we do not have good images, or we have no images at all.
So, I have been trying to find a way to create new images of the past, that I could then use to take “characters” from who would be able to “act” in videos.
While it’s possible to do this with popular AI image generation tools like Midjourney, I find that the images that Midjourney creates are too “Hollywood.” That may change with time, but presently that for me is a limitation.
Over the past couple of days, I tried to create historical people using something called Flux in ComfyUI. My computer is not strong enough to do this, so I found a way to use a cloud service (Lightning AI), and with the help of YouTube tutorials and ChatGPT (this one, this one, and this one, for anyone interested), I was finally able to get it to work.
I took the above picture from 1924 of the regent of Kudus (in central Java) and his family and had Flux recreate it. I just ran it with the default settings and did not attempt to tweak anything.
From this initial experiment alone, I can see that this is going to be possible. Sure, there are some obvious problems with the images, but again, this is a first attempt just using default settings.
There are also going to be issues to consider. For one, I don’t feel right to include babies/infants. I actually used the generative fill function in Photoshop to remove the infant from these pictures, that’s why the mom’s arm looks a little weird.
Yes, I know that there were babies/infants in the past, and I’d be ok with creating comic versions of babies/children, but not lifelike ones.
In any case, now I can see that there is an AI future for the past that doesn’t have to be all glossy Hollywood style.
