Step By Step…
On the heels of Barbiefied, I decided to explore the other toy genres of mini-figures and action figures. Next on the list is the LEGO Group. First, we need a reference image (if our goal is to make the toy figure match a real person). This is a photo of me sitting outside enjoying the beautiful weather!
Reference Image
• Upload your image into Midjourney – drag a JPG or PNG onto Midjourney and press return. It will upload the image.
• Select your image. When it opens larger, right click and choose “Copy Link”
Prompt.
– Start it the same way you have been doing – /imagine
– Paste your link
– Following the link, succinctly describe the person
– Add pertinent details to call on the Legoworld elements.
• https://lnkd.in/gU3AeycG [this references my uploaded image] Lego version of Brian, middle aged friendly figure with beard, wearing ball cap, orange shirt. blue pants. sticker, no background, in the style of Lego –ar 1:2
[Note that I gave it my name in the description. While that information does not carry over to other prompt renders, it has benefit in the prompt being worked on.]
Zoom Out to EXPAND image
•
I decided to work with the image that had black bars to see what Midjourney would do with the bars on an expanding zoom out…
• Clicked on [ 🔍 Zoom Out 2x ] and selected my fave.
Once you have your base image, you can use the expanding features of Zoom Out or Pan to show more of the body. For these, I first did a Zoom Out 2x, and chose my favorite from the 2×2 grids rendered.
• Pan (down arrow) expands the image to provide more of the body in the shot.
Looking a bit like Bob the Builder…
I did not stop here… I wanted to take this mini-me, and put him in a different setting. SO, I went back to Midjourney, and prompted for an office setting.
• Macro Photograph of a Designers office made of Lego. Lego set of bright, colorful bricks. On display near a window in a real office –ar 16:9
I brought mini-me into Photoshop, and knocked out the background. Imported my favorite of these 2 office spaces, and added the newly cut-out mini-me that I scaled to fit.
On my LinkedIn post, Daniel Huss commeted:
This is actually really impressive. It accomplished this with a single image? I’ve been working on trying to get style transfer to work after training Stable Diffusion on 100s of images of myself, but its never this good.
Daniel Huss
At the request of my friend (and fellow AI enthusiast and educator) Tridib Ghosh – I re-ran the steps above and created a version of him!
