Future or Fake? A.I. in War Films.

Is AI going to replace filmmakers? Grab your movie prop guns and buckle in for a quick and dirty look at basic AI creation tools!

DALL-E
This is Microsoft Bing's AI image creator and its pretty decent when you have very generic requirements to create. So far it is free but falls down on the fine detail of images and when you want to start making variation of the same scene. It works by creating a detailed text prompt. 

Below you can see the variations and errors from this simple tool.

Text Prompt "A photorealistic wide shot picture above a convoy of army trucks on a desert road in Libya in 1943 with blue sky and a few scattered clouds"

So overall the AI creates a typically Hallmark-looking glossy image that looks fairly decent except there's thousands of trucks that look more or less vintage. Some images have what appears to be planes flying in the distance except they may also resemble pterodactyls! Picture No3 has soldiers on top of the truck with obviously AK47s which are not WW2 vintage but what does an AI tool know anyway!

Text Prompt "A photorealistic wide shot picture above a convoy of 20 army trucks on a desert road in Libya in 1943 with blue sky and a few scattered clouds and some black smoke in the far distance"

If we add some variation to the AI creation text we can see that DALL-E starts to struggle. The AI picked up the need for black smoke in the distance but ignored the '20 trucks' (I tried both numeric and text commands with the same result). At least the vehicles seem generically 1940s anyway.

For me as a filmmaker, the utility of Bing is likely in the creation of fast very generic images either as story board or simplistic shot blocking ideas to demonstrate a concept. So far my research indicates that the images will get you about 60^ of what you are after.

​Text Prompt: "A storyboard style black and white sketch of ten army trucks on a desert road in Libya in 1943 and some black smoke in the far distance"

Text-to-Video
SORA, the much feared or desired text-to-video tool from Chat-GPT creators is not widely available yet and is being red teamed to make sure no-one creates videos of nasty stuff (but then who is to define nasty - that's a real rabbit hole of free speech etc). This tech below (Gen 2 by Runway) is certainly not SORA and seems to only generate static images with some horizontal or vertical movement. As can be seen, the objects in the video start to morph like a bad Salvador Dali painting and the AI has a trouble placing priority on each asset in the image.
​
Video Prompt: "Camera fly in over a convoy of 20 army trucks on a desert road in Libya in 1943 with blue sky and a few scattered clouds and some black smoke in the far distance."

As you can see the tech is pretty limited for this kind of application so we'll have to wait and see what SORA can or cannot do. Ultimately use of AI will be up to the filmmaker and I think its utility in indie war film production will be very limited. If you are an indie sci-fi filmmaker, the text-to-video will create totally surreal videos that would be impossible to create in practical fx (even if you're Christopher Nolan!) so for us historically accurate war film nerds - we'll have to stick with the old fashioned methods (for now).

Thoughts and ideas - put them in the comments below or subscribe to us below!