Oddbean

It’s amazing how quickly Ai generated images became old and largely uninteresting. You can still produce really cool and fascinating images, but it takes a *Lot* of exploration and prompting to get there, but I’ve always thought the real secret of the tool is in using it to piece together a more complex artwork and composite something with an explicit vision. Not merely typing a prompt and taking the output. The amount of work the “creator” puts into it will also be reflected in how they treat it. You can still just see the proof of work somehow. 😆🤔

Unfortunately you can’t do it entirely with MidJourney. Their tools is very “one size fits all.” We do use MidJourney a lot for individual elements, but everything we do is put together in photoshop, upscale with Magnific, and a lot is manually drawn and then used as ControlNet input in Auto1111. Using this you can outline a picture, put in individual elements, create a background or setting, composite them together, use that to generate a “first draft” image, then pull a ton of creative detail out of it with something like Magnific. It’s also just stupid fun to play around with it all. Ai is also great for just general ideation, but using the basic output is always going to just feel weak. The only times I use the standard is when it’s done simply to create stock images for a video, it helps reinforce an explanation or to paint a picture, so it’s less about the art than it is the image itself.