The Most Creative Midjourney & ChatGPT Prompt Method!

hi everyone today we're going to take a look at probably the most creative way that you can use chat EBT with mid-journey and this isn't about creating some intricate elaborate prompt formula in chat apt that like spews out hundreds of mid-journey images so I've mostly stayed away from the idea of using chat EBT as a mid-journey engine you know I realize there probably are some novel uses for it but for the most part it all kind of feels a little too automated for me if you've been following the channel for a while you know that I always like to stress the human element of using these AI tools so the idea of using chat GPT like whole cloth to make prompts just seemed a bit reductive to me and also kind of not fun but both chat gbt and mid Journey are very powerful tools so I knew that there would be a potential to synchronize the two of them in a unique way and indeed I did run across a method that I think is phenomenal and I'm going to share that with you today run you through the whole thing we're going to take a look at some images and then to coin a Billiards phrase we're going to put some English on it all right let's go Dive In so I ran across this technique from a guy named David Hardy on a mid-journey forum David by the way if you're watching and you want me to link something a website or anything just let me know and I'll pin it below basically the technique Begins by running a describe on an image in this case I just took one that I previously generated uh cinematic design imagination Montage no text copy words ar169 and then we run a describe on it describe returns to us four prompts uh an imaginary City and a fantasy atmosphere there's an old mystical fantasy town with lots of buildings and decorations I'm not going to read through the whole thing but yeah I mean you know how describe works and if you don't know how describe works I did do a video on it kind of recently it's linked below quick note as well in our recent office hours mid Journey did let us know that the describe results are not connected to the mid-journey prompt language model so it's kind of its own independent thing kind of like clip interrogator that's not to say that it isn't really really useful though it just does explain why sometimes describe will give you these prompts that mid-journey clearly is just ignoring I should say for V6 and V7 they are working pretty hard on the language model so I do expect those to be connected at some point or another it's it's just it's not for right now so now that we have our four described prompts we're going to take them over to chat GPT and just as a quick note I did try this out on Bing chat as well it seems to work although you got a finesse Bing chat a little bit more anyhow once you're in chat EBT you're going to want to run the following prompts these create a truncated tokenized summary that blends specific details and cultural influences from four descriptions I will supply when ready emphasizing the hybridization of these descriptions prioritize specific details over generalizations and use varied sentence structures the resulting summary should be around 30 keywords long and should not include any reference to the prompt that generated it or the methodology used to arrive at it and chat gbt returns to us with a hybridization of all four of those prompts so the plus side in having chat gbt do this is that by adding in the truncated tokenized summary command at the top it's outputting something that should probably fit within the token limit length of mid Journey if you're kind of confused about the whole token system um I get it it is a pretty confusing topic I do go over that in the describe video that I mentioned earlier once again linked below so let's take a look at the four describe outputs plus our chat gbt hybridized version so just as a quick refresh this was our initial image uh describe one got us this this was our image for described two this was our image for describe three which is really cool it's very different than all of the other ones I I think it's pretty neat actually and this was our image 4 described four so overall pretty cool and this was our output for our hybridized version which pretty much looks in line with our other outputs minus number three um so now we've proven that it works so let's kick it up a notch so as you know when you run a prompted mid Journey you end up with a grid of four images in this case cinematic still idyllic sci-fi and we get these four Images so what if we were to run a describe on all four of those images at once and then hybridize them in chatgpt so running all four Images at once in describe led to some pretty interesting results in the first describe it mentions a tree and a pond in the second description it talks about uh cinematic stats mysterious jungles Xbox 360 Graphics as well but whatever um and you know in the fourth one we've got emotional Landscapes quiet contemplation so it's an interesting blend of all four of those images per prompt so then having chat GPT take all four of those and kind of of compressing them all together we end up with the prompt spaceship Tree Pond on and on it goes let's take a look at what we got here so our first describe was pretty interesting it really focused in on that tree I think it lost a lot of the sort of sci-fi alien elements but you know it still ended up as a cool image our second describe brought us these images which I think are kind of cool it kind of gave us that image within an image thing that mid Journey sometimes does um yeah pretty neat definitely back to the alien Landscapes here's another output from that describe that I think kind of leaned more towards the first image in our initial describe set as it kind of has the same kind of color palette to it and our last described which I think looks really really nice so now let's take a look at our chat gbt hybridized version and so here it is and it's interesting because it kind of ended up becoming a whole other thing and going into a different direction um definitely much more in the jungley side than the previous images the spaceship tends to be much more prominent and then in the bottom two images characters um in the other ones there were were characters they were just much more distant but in that third image we're much closer to a characters so I think this shows that we can use this technique to get outputs that differ from describe which got me thinking well maybe we could use this almost as a better blend so let's take a few of the most famous paintings of all time and see what we can do with it so I initially tried to experiment with Van Gogh's starry night but I think that image is so iconic that the describes come back as Van Gogh starry night or starry night is depicted on the screen so it felt like that would be too strong an image so I ended up grabbing his Cafe Terrace at night which described described as a building with table on it by Van Gogh in the style of atmospheric Urban escapes painted scene of people outside of Cafe with lamps you know on and on a ghost I wanted to see if we could combine that with Dali's Persistence of memory and just for fun we'll throw in the Monet so what I ended up doing was taking just one of those prompts from each of those describes and having chat merge them together the results were pretty interesting I think in this image it definitely kind of leaned into the van goghisms but there's a lot of dolly-isms as well in the center of that frame and then in this one I think the sidewalk definitely has a watery look that you know is reflecting Monet so after doing some more experimentation I did more or less feel that I think having three artists with such iconic Styles was just a little too much for a mid-journey to juggle all at once though I have to admit there were some cool outputs they may not necessarily represent like a art jam between Monet Dali and uh and Van Gogh but you know still these are kind of cool so I had an idea to take Monet out and instead replace his prompt with a famous photograph in this case the National Geographic cover uh girl with the green eyes I think it's called Uh which described gave back to us as a young woman wearing a red shawl looking at a camera red scarf with blue eyes blue eyes wow that's interesting and a red band National Geographic photo so and then add adding that into the chat gbt mashup and seeing what we got the results ended up being pretty cool like this one in particular I think had some nice van goghism some nice dolly-isms and then had I wouldn't call it a photorealistic character but you know a highly rendered character I guess because of the photographic prompt so because we were losing the sort of overall Vibe and style of the artists when we had three of them I thought maybe let's take it down to two and see what happens and that's where we get some magic so I took two of my favorite paintings um Edward Hopper's Nighthawks and duchamp's descending staircase number two which if you've ever seen this painting and you're like I don't get it uh it's basically Duchamp was trying to create motion and the movement of time through the painting So if you look at it there's a woman walking down the stairs in this photo It's Just Happening all at once so there you go look at that the art degree did something so describe gave us a couple of options I ended up taking number three Philly's coffee shop 1950 Edward Degas uh wrong painter uh in the style of photo real detail dark Navy Emerald to describe on Duchamp came back super surreal like the first one was figure 27bd person The Lyricist uh and then the second one it actually calls out the year as being 18 million um so yeah that's it's pretty great and got these images which I do think are pretty cool one thing to note and something we've discussed here before is that mid Journey tends to prioritize words that are at the front of your prompt um just the same chat GPT does as well in that it reads your prompt but it isn't reading or thinking ahead so if you change the order of your describe prompts you will end up with a different output in this case I took the Duchamp and Hopper Nighthawks prompts and reverse them and we ended up with this image because Chad TPT organized The Prompt starting with Philly's coffee shop 1950 Edward de gospel once again this is not Edward Degas um but yeah it ended up kind of a really cool image I think it lost the duchampiness of it all but still a you know really nice looking painting overall I think this is the most fascinating use case for combining chat EBT and mid-journey together it certainly miles more interesting to me than the you know prompt formula of make me a hundred pretty pictures robot and there are countless different ways to approach this like I didn't even touch on cinematic Styles which I'm very curious to try later on tonight so I do hope you enjoyed this video if you made it this far and have not hit the like And subscribe button I do invite you to do so and if you want to check out that describe video that is coming up next thank you for watching my name is Tim foreign [Music]

Channel: Theoretically Media

Views: 13,664

Keywords: Midjourney, midjourney prompts, midjourney ai, midjourney ai tutorial, midjourney tutorial, AIgeneratedArt, CreativeProcess, ChatGPT, ai tools, openai, ai art

Id: 7ttBvM8i9_o

Length: 10min 22sec (622 seconds)

Published: Tue May 02 2023