What's The Prompt For That AI Image? (Here's the Trick)
HTML-код
- Опубликовано: 27 май 2024
- In this video, I explore how to uncover the prompt that was used to generate a piece of art.
Hugging Face space:
huggingface.co/spaces/fffilon...
🛠️ Explore hundreds of AI Tools: futuretools.io/
📰 Weekly Newsletter: www.futuretools.io/newsletter
😊 Discord Community: futuretools.io/discord
🐤 Follow me on Twitter: / mreflow
🐺 My personal blog: mattwolfe.com/
🌯 Buy me a burrito: ko-fi.com/mattwolfe
🍭 My Backgrounds: www.futuretools.io/desktop-ba...
Outro music generated by Mubert mubert.com/render
#generativeart #MidJourney #aiart - Наука
Answer to common question in the comments: Yes! You can upload any images and have it give you a prompt to describe it. They do not need to be AI-generated images to have it attempt to create a prompt around the image. I uploaded a headshot and, well, it tried... LOL. See example here: twitter.com/mreflow/status/1643125694099251201
Thanks for the video and all your great videos. I wonder if MJ is now broken about /describe since all I got are errors. I have tried many kind of images. Images done with MJ and only errors. Very weird.
Heyy, what can you use to improve profile picture dating ptofile
I want to buy this guy a beer
Good because I like to steal😊
Tips are on for the channel. Called (Thanks)
Beer is so 90's. You mean, a bag of Xylozine.
You can. It's in the description.
Buy him a Burrito 🌯
For me the /describe feature is more useful for real photo’s. I struggle for instance to get motion blur on a face in Midjourney but now I just insert a real image and now it gives me a better prompt for the result I am after so it really helps me to better prompt.
this just destroyed the emerging ai art prompt market
That market would have been destroyed regardless. The thought of selling prompts always seemed dumb to me. But that's just my opinion. LOL
Obviously there was a market for it if people couldn’t come up with their own….why wouldn’t the only creative element in ai art be open to plagiarism? (I know that’s ironic…)
Dang. The timing on this was crazy. Midjourney just released "/describe" to generate a prompt from an image.
Yeah I figured that was what this video was about and was surprised how quickly Matt made this.
He talks about the /describe so it's almost certain he saw that first then looked for alternative systems which have been around for longer.
Google LM has a system where it can turn a photograph into a song.
That's why he gets paid the big bucks 💯🍄🔥😂
I saw the announcement in MidJourney this morning and made a video about it along with other ways to help find the prompts. :)
your videos are amazing 🔥 I come to your channel every day to see if there are any new videos up now. just signed up for your weekly newsletter too! Thank you so much for making such helpful content! 👏
Keep ‘‘em coming Matt! Thank you for brightening up my day - everyday 😊✨
It doesn't have to be an AI generated image either. I've used photos of my cats in all three tools to test the tools. I also have used photos when I needed a hint for how to make a certain look. The new Midjourney /describe tool is the best because it gives 4 suggestions and you can just click them to try them all. I love it so much!
Love your videos bro, as always, thanks for the info!
This was helpful for me today. It's awesome. This is helping me quite a bit for real. First bit of progress I'm actually seeing tbh.
Amazing work Matt, thank you for sharing.
Excellent information! Thank you 🏆
MATT how are you missing some of these new ai papers. where is a video on huggingGPT? The Reflexion paper? all the stuff with iterative criticism and refinement. DERA????? i feel like agi is around the corner and nobody is covering the very obvious path to agi sitting right in front of us! check it out if you can.
I read the papers. I can only put out so many videos. LOL. Some topics are harder to make videos about than others.
@Matt Wolfe We will humbly and patiently await
I found it interesting that after telling us to write prompts as full sentences in V5,. what I got back from Midjourney's /describe feature was more like a v4 prompt and there were a lot more words than my original. The results were similar but not what I'd classify as reversed engineered.
My own experience is too much sentences it gets lost but lots of descriptive words is working for me.
Even if you could get the exact wording for a prompt, you still need the AI Model, Guide settings, Seed Number, Negative Prompt, etc. The prompt is just one ingredient in the recipe. I think even the equipment being used (PC/Laptop) has some influence on output. It's all math after all.
True. The idea isn't to get an identical image. It's too get something in the same vein.
You can use img to img anyway, if u get the prompt then the image to image will work even better
You've probably noticed that the image/link from Midjourney includes part of the prompt used.
I discovered your content on SubStack and then googled how to find prompts for images and you were first in the results. You’re awesome! I could literally watch your tutorials all day - week.
Thank you for your enthusiasm
Good looks thank you!!
Thank you! Amazing content!
Fantastic! It is just too cool all the stuff you can do.
So fun 🎉
Thanks a lot! Really a grat via with all the infos are needed for most newbie!
Super helpful channel love his stuff
Its because of the metadata of the image. every image gets its prompt as metadata after being generated.
Does that mean you can open the image in noted pad or something similar to find the information?
@@OliverJonCross i tried an online meta data viewer . it worked . default windows viewer won''t show it. i used a site called jimpl
@@OliverJonCross Never mind notepad works
I wanted to see you get recursive, taking the generated images, getting it's description, generating that, getting the description, generating that, etc. I am curious about the drift.
This is great. I’ve been having fun with Bing image creator but can’t do anything with the images (personal use only), so it will be great if I can recreate something similar enough that I can then sell.
Ps. I looked at Sam Spratt art and it’s detailed pencil drawings or dry media, it doesn’t look like vector art.
This is great information i really love your videos
If you see names like Wojtek, Benjamin, with a J in the name, and you don't know how to read it, then probably it's read as a Y, and it's probably Polish. J is read as Y - I think it comes from German where the letter is used in words like Ja, Jawohl, Jacht, Jakob, etc.
another cool video that helped AI enthusiast like me... great vide and thanks a lot for sharing bud...
THANK YOU VERY MUCH !!
Wasted almost 3 hours of my life trying to figure out what prompts I used 5 mins ago T_T brutal.
JUST WOW thanks a lot Man!
Bro, I saw the feature 8 hours ago, your video came out 4 hours ago ... damn actual :D.. thx..
Great work brother keep it up 👍👍👍👍👍
The timing is very coincidental.
Do they talk to eachother, or did MidJourney reactively release a feature to prevent users from leaving the platform?
Midjourney is based on StableDiffusion, so they likely talk to each other, or at least there is some significant cross pollination going on :)
Midjourney is not based on Stable diffusion though, they did say that last month when they were releasing MJ version 5
I'm not sure what you mean. What's the coincidence? That MidJourney stopped allowing free trials and they release a new feature to make it more valuable? I'm not sure that's coincidental. That's just them proving that's there value in paying.
@@robiaster MidJourney is not Stable Diffusion. They use their own technology. You can actually see it while generating. Stable Diffusion will start with white, add noise and diffuse from there. MidJourney will start with a black image, add noise and diffuse from there. At one time, MidJourney used Stable Diffusion but their recent model is completely proprietary to them now. You prompt the two tools in entirely different ways too.
MJ probably knew of alpha/beta testing of reverse prompt engineering going on, and hrrmmmn promptly commenced work on their own flavor
Hey Matt, I just watched your video and I must say that it was really informative and well-made. I was wondering if I could help you edit your videos and also make highly engaging shorts for you?
More fantatasticness!! Thanks Matt. Great stuff.
In the future, we will be able to turn a photograph into a video game!
Can't wait!
Thanks very much men, your information is very useful 👍👍👍😃
Matt Your The Boss!!! Thank you for this!
Thank you 😊
Thanks for sharing all these tips. So far, the midjourney one is only for subscribed members. Clip interrogator is the only one that could generate similar prompts. But when I applied these prompts on dreambooth model, it does not generate the same styles on the trained model. I wonder if it only works on random characters.
Thank you so much
Thanks!
Good job, greetings from Poland
when using adobe firefly i have the series of prompts saved and i re-generate the image each day. and every day its somewhat different in the 4 results i get.
if i start over and paste in each prompt then 'generate image' one by one it will also come out differently.
again, if i start over and paste them in a different order, it will also have a different set of results.
however, they will all have similar themes as its using the same prompts but they get interpreted differently depending on those factors.
LETS GO!
LFG
Describe feature just made midjourney great again!!
🎉
Love your videos and your posts on twitter!!😊😊😊 Greetings AI Einstein
Isn't his stuff so helpful ❤
@@aiartlounge It really is, huge Fan here :)
@@AIEinstein ❤️🧡💛💚💙💜
bro, you're on fire
Would be nice if Elevenlabs or Tortoise TTS would generate Scottish, German, French accents of English, and other languages.
Hopefully in time they will ❤
Thank you.
yes that was great info
I’m an artist but I also like this AI stuffs, kinda geek n like to nerd out with you 😂
Great Hat!
time to leave midjourney and try other ai images like bluewillow
what if you want to take a photograph of someone and turn it into AI art photo, how to generate prompts for that, will drag and drop image work
People think they can hide their secret sauce but AI will get better at decoding images.
The secret to MidJourney is that it pretty much makes good images no matter what you type these days. "Secret sauce" is barely even necessary anymore. :)
Maybe a quantum based AI.
How to do batch image process?
Matt! A bigger question: I want to know if I create an image in Midjourney, take to GIMP, alter in some way, export to PDF, then use for commerical use, say a book cover on Amazon, is the image still containing any metadata that will identify it as AI generated? In other words, I'd like to be able to maintain that it's a product of my labor and thus a legit use commerically (of course, anyone could use the image to generate new similar images) but I'm not worried about attacking, I'm worried about defending (I mean in a commerical copyright sense).
If there is no metadata after going through GIMP and exporting to Amazon (or other place) as a PDF, sure, people can know all day long that it's at least partially AI generated, but so what? If they can't prove it, then the work is essentially protected. Thoughts?
I think it’s hilarious you are worried to have an image that essentially ripped off others’s original creations in order to be generated protected….
@@user-cp3kh5xo2h All work is "ripped-off" from someone else, nobody is original.
It's a dumb analogy.
If you put work into a specific piece that piece is unique, but they style? The design? Nah, hasn't been any original art since a few hundred years ago at least.
@@michaelsbeverlyconsidering it was a simple prompt, already low effort, yeah they should at the very least try to create their own…
And people are original…inspiration for a human is quite different…the ai was literally fed images owned by others to “train” it which means it had stored those images in its vast stores of reference materials which it then cannibalises in order to create its composites that it spits out…
@@user-cp3kh5xo2h Fair enough.
One of the things comes down to protecting the consumer, i.e. if you copied a book cover of a best selling book, in hopes to trick readers, the idea that the cover was AI generated wouldn't matter over the idea, at least to me, that someone was trying to trick readers
In any case, a couple months back, Amazon asked indies to answer questions about AI usage, but it was only about content, not covers.
WOW!
Damn, the value of creativity is in serious danger these days.
You can do this with /describe in midjourney now 🚀🚀
Anyone else running into a glitch with the newsletter link? Page goes blank for me after like half a second
Its about time MJ are releasing their /describe tool. Ppl at prompt engineering community are gatekeeping knowledge and hide their prompts and images in a race to the top
I am getting "Internal Server Error" When I try to use 'Midjourney /describe' feature. How to fix it?
It may have been temporarily down.
pretty daaaang close!!!
when i used midjourney to get the prompt using the /describe command I got this message "Subscription required"
1:26 😅😅those negative prompts are hilarious
And midjourney did it officially
What are negative prompts, and how do we use them?
anyone know which AI image tool is best choice for people who having a little money but want to deploy the massive strenth of AI tools??
Stable diffusion
So cool do you happen to know how we can find all of the images that we’ve generated in discord? Can we search our username? Do you have a tutorial on how we can sell AI art ?
If you go to the MidJourney.com website and login, you'll see all the images you've generated.
Don't be too sure, cause it doesn't account for extensions like control net or composable lora too.
I literally mentioned that in my video.
@Matt Wolfe the tricky part in conflating all the AI art platforms is that while Stable Diffusion is increasing its specificity and control of image creation, Midjourney and for profit platforms are now disallowing more and more words to "paint" with. Even with the process you outline here, the prompt that is uncovered may not be used in its entirety as certain words or phrases won't be allowed. It's interesting and important to point out this divergence, especially as more for profit AI image creation systems appear.
Anyone else going to name their firstborn after Matt Wolfe?
Isn't there a first person perspective view prompt? Because somehow the ai doesn't understand me what first person perspective even is
Trying the /describe option and just getting 'Apologies, something went wrong.' error
I think it might have been temporarily down for a bit.
@@mreflow Tx. It does happen occasionally. Still, not out for long
btw. you dont have to write:" a photo of dominikcooper person". just dominikcooper does it..
Just tried it but not working! :< Midjourney says there is internal server error. Tried many times. Does not work!!!! :
It works. MidJourney had a short outage because so many people were trying it at once.
@@mreflow I got in. It kind of works..but I always record my prompts in a text file to copy over if needed and then change words as fit. But it may help so you do not bother people on what prompt they do.
@@senju2024 Problem I'm having is that it works find in the MJ bot, but when I created a text room and tried it, it wasn't in there. Any clues?
@@DeankutI do not know. I only using MJ bot.
@@mreflow I can't seem to get that room you have to work. I created it, but it doesn't work in that room. 🤷🏼But it works fine in the MJ bot. Any clues?
Hi Matt, I wonder if you could be so kind of pointing in the direction of finding a solution to my task. As an owner of a painter and decorator company I always have the problem to get customers imagining a make over of a room, for instance. There are some mobile app from paint producers where you can take a picture and change colors of a wall... And so on. But that's it. Doesn't implement wallpaper nor furnitures. A solution for this would also be of great help for estate agencies to show more possibilities to potential buyers. Long story short... Do you know about a solution. Thank you ever so much for your comment in advance. Greetings from 🇩🇰
Cool idea, seems like a way to upgrade some of the technology already out. To be able to change a wall color on your phone shouldn't be too far away though if it's not available yet. ❤
Why do use MJ4 as your default and not 5, Matt?
He changes it up I'm sure ❤ I know v5 is better but it is a lot different, and more photorealism based maybe?
Holy SH*T!
can it describe images that weren't made by Midjourney?
I still confuse how to add audio track for my video😂
OK now I'm gonna have to subscribe with my main account as well there's just too many videos I'm missing.
And done I'm double subbed now mate can't get any more committed than that, can't wait to share all the things I've made with the tools you've shared just amazing stuff Ive been in IT since the 80's and this is the biggest thing to ever hit our collective minds. IT CHANGES EVERYTHING! the possibilities are infinite 😎🖖
SDSU almost did it! Go Aztecs!
Do you ever noticed that midjourney can't generate 2 specific character in an image ?
How do you get any results at all for your name when you have "ugly" in your negative prompts? ;)
So that's why it always comes up blank!
Ist there an API?
coming soon
Has anyone read the recent Time Magazine article on AI published by Elizier Yudkowski? Thoughts? Time to "shut it all down my all means necessary"?
Did the name of the MidJourney pictures containing the prompt give MidJourney an extra clue what to look for when you reverse engineered?
No. It doesn't look at the file name. I could have renamed the files anything I wanted and gotten the same result. It looks at the contents of the image itself.
Could you just ask chat JPT what prompt created this image
10:09 The deer is created in a style of art called WPAP.
What happens when you feed it original artwork made by a human?
It will try to guess the prompt to generate something similar. This doesn't only work with AI generated images. You can toss any image in and it will try to guess a prompt to get you close.
You can modify any image, regardless of how it's made.
@@mreflow Ty both. I've always wondered what REALLY makes those classical art pieces so special. Maybe such tools could help uncover those secrets, at least for the non-art savvy like yours truly...
I really want to talk to you about it
👋
Honestly I think this would work for literally any image, not just AI generated ones, no? I believe the only reason this is so accurate is because your images are in the database it is using. So it can precisely tell you what you did. If it wasn't. It would just give you the closest prompts to achieve that image in midjourney (and or whatever AI they are using) So depending on the AI you use, you may get wildly different results if it's working of entirely different data.
Another reason why they're potentially not as precise if you're like myself and use your own art to help modify outputs. will give you an image that has vague semblances of the one in question. But not nearly as close as others. Because it is losing what ever % of the image you used to train the ai to create the desired image. via the image strength toggle. So the parts it keeps when it gives the output is the only information of that reference that it will remember for this purpose. Even though the parts that weren't used, were part of what trained it to know what you were actually looking for by eliminating low potential desires. But this is just a guess as there really shouldn't be any reason that it wouldn't be able to give you the direct seed of the image otherwise with a simple algorithm to do the calculations based on the datasets the AI uses to generate the art to begin with.
It works with any images. They don't have to be AI generated. It's not looking at what's in the database to check it. It's trying to detect what's in the actual image. I could have uploaded an image I made in Stable Diffusion and MidJourney would have figured out what prompt to use to get something that looks close. The 1st tool only works with images generated in Stable Diffusion. The second two would work with literally any image at all, AI generated or not.
@@mreflow I see, thanks for the clarification and response. Especially so quickly. Keep up the great content! I find this all incredibly fascinating and your content is great for information and insight.
Please use the correct terminology, it's PROOMT, not prompt, other than that, great video!
The more I learn about Ai generated art the less it concerns me as a graphic artist. I feel sorry for those Ai only users with no art skills digital or traditional who think that they are going to make a career from this Ai art.
Nobody will be making any career soon enough... We will all just curate (free) art and (free) food all day long...
People who use Ai have a completely different mindset. These are lazy people who won't take the time to learn the skills needed to become a good artist. They don't understand how amazing it is to create something with your own hands and imagination. The program is the artist, not them. They can't even draw a matchstick man without these program.
@@janwelander4110 or maybe they didn't want to become an artist. They just wanted to create something.
@@truthseeker1934but they’re not creating anything….they’re so lazy that the only real effort that was required (prompting an image) is being made idiot proof and being copied as well…
This does NOT work !!!!
Google "Image to prompt"
I just did. It said nothing about MidJourney's new /describe feature.
@@mreflow I know but the first two search results are good alternatives if required.
Why can’t people just create their own imagery instead of copying? I mean, it’s already such a low effort endeavor….people are so lazy….