I recently re-subscribed to Midjourney, to test out some of its new video functionality and I discovered a little VFX/ Post-Production hack that really helps out when isolating images. But first, if you haven’t already seen it, the video is below (there’s no sound as it sits on the Website’s homepage).
Why Make This Video?
There’s a big project I’m working on behind the scenes which is going to require the blending of traditional design/ motion graphics/ filmmaking techniques but mixed with modern AI tools. There’ll be more about that in a later post but while I’m paying £50 a month to Midjourney, I wanted to familiarise myself with the tool. So thus began the challenge of rehashing what I’ve been up to over the last half decade in a visual format that would sit well on my website. The limitations I set was to keep the text in 2 containers around a central video component, make the copy rhyme throughout and keep the imagery relevant to the copy while complying with the blue and purple colour scheme of the site.
Is it all A.I?
No, I wanted to have full control of the copy for this, I often use tools like ChatGPT to bounce ideas off or to troubleshoot but as this piece was about me, I wanted the written part to be as close to unedited as possible. There was a challenge in getting certain lines to fit within the containers, so a lot of sections did change.
There’s a hybrid of Midjourney and Photoshop throughout this one, for example the director’s clapperboard was without text, which was later added in Photoshop. The 3D camera is a gen that required heavy editing in Photoshop (Midjourney tends to generate images with blemishes that need fixing before you can run it through their video gen). The laptop, phone and book are all pulled from previous work I’ve done and are modified After Effects templates. Finally, everything has been colour graded to fit the brand guidelines of the website.
It’s Not Possible Without Keylight
Here’s the secret sauce to making a video like this, first of all you’re going to need to prompt your images ‘floating’ in front of a flat green background. You’d end up with something like this:
It’s not perfect, but it’s good enough to tweak, the following step is to take it into a tool like Photoshop and remove the background, then add a chroma green backdrop afterwards.
You would then add this back into Midjourney as the reference image before prompting the video component to spin on its axis extremely slowly. If you’re lucky you’ll get something like this:
And once you have that, you can take it into After Effects and run an effect called ‘Key Light’ to take away the green, allowing you to place it over any new background you like.
What Are The Limitations?
You’ll still have to generate multiple iterations to find a clip that you like. You have to be incredibly specific with your prompts or the A.I is going to run off and create a lot of weird visual artifacts. You have to be eagle eyed when looking for mistakes and even with ‘good’ generations, they’re still not perfect. A.I should still be treated like a tool and is worth exploring, but the important factor is blending it with the other creative skills you’re developing. I wouldn’t have been able to create this video without having experience with motion design, typography, Photoshop and After Effects. As these creative tools become more accessible, those with a solid background and understanding of the theory will still have an edge.
