I love when you reveal, “behind the curtain.” Here are the words Midjourney ignores. I can’t believe how complicated the prompts are. My experiments with Leonardo AI were a couple of lines…the one you show-a novel. Lots to learn on my part. Thanks for this video.
90% of the time MJ will turn out a beautiful image, even with a two word prompt. However, if you're after a specific outcome that involves a combination of elaborate elements, characters, foreground, background scene etc, forget about trying to achieve it in one prompt. *Create your amazing character/subject in one image and your amazing background/setting in another image, then combine them using image prompts. It provides MJ with clarity about your desired style and composition. MJ is a visual medium that responds better to images than words. I'd be interested to learn how MJ weights a single image prompt against the actual words it took to create it. My guess is that it uses less tokens for the image, than the actual prompt. *Ed typo.
Excellent explanation. I would think it's a matter of taste and creativity as to deciding to use a long or short prompts. One additional consideration seems to be processing time for each token which I would think that depending on your computer (memory, ram speed, cpu , etc) which may or not be noticeable. I favor shorten prompts, but I look for words that gives as clear vision as I can. Then go for variations.
How do u add text weight into a shorten command? And would midjourney reevaluate which is more important after we added the text weighting? thank you for the clear presentation of the video tutorial,...
I did some testing with weighted words and /shorten did seem to pick up on the weights. It's been a while since I did a video on weighting words in prompts. I should probably update it. "Multi-Prompts in Midjourney :: Should You Use them?" ua-cam.com/video/ayIp712u-vQ/v-deo.html
6:56 We know this is not true as demonstrated in this video with "purple" being indicated to have 0.00% impact....... but the actual results are clearly drastically different to what the AI claimed. We cannot trust/Shorten at this stage. Hopefully somewhere down the road it will be SIGNIFICANTLY improved to the point where it can be trusted. 🙏 I hope they can fix the flaws in /Shorten.
I think one issue is that the /shorten command was designed for long prompts. But I'm putting short prompts through it. This is worth watching as the tool improves.
it has been helpfully in figuring why certain words don't get used as much as i would like, right now im trying to figure out how to hide images inside another image and you can image how that is going. For instance high school girl Japanese, standing in front of busted worn down series of mirrors in a run down bathroom, she is holding a camera up to take a picture in the image you can see a creature looking at her that not there. Something like that
👋 I've done a fair amount of testing of the /Shorten command / directive and my conclusion is that it's not very useful since the results of using the prompts /Shorten generates are so frequently noticeably different and not necessarily better that it's actually not a very good tool to shorten very long prompts to consistently enough get good or better similar results.
My prompts are really terse, all of the words seem to matter. Midjourney's suggestions tend to change the images wildly. Don't even start me on /describe, Midjourney doesn't know where to start. Generally the results turn out to be some kind of crazy mess!
4:18 👍 👍 👍 That's precisely what I pointed out to you in a previous video and why the analysis by the AI can NOT be trusted........ that and other reasons including the fact that shorten never really produces results very similar to the original long prompt. So shorten is virtually completely useless if you want to consistently produce the same results as the original with a /Shorten prompt.
4:40 No, they're actually NOT necessarily better. Better is entirely subjective and I actually prefer at least 2 of the images on the right produced as a result of the longer prompt. I'm definitely not one of those people who thinks that all long prompts are necessarily better or good........I've seen plenty of very long prompts which are contextually absurd and produce either crappy or at best average results. *My point is......there are much better and much worse very long prompts / prompting techniques........we need to distinguish between them and try not to lump all very long prompts into the same waste basket of poor prompting technique.* This seems to be especially true when considering very long prompts which use weighting. 👍 👍 👍
In my testing, I was usually ok with the 1st two suggestions. Were they always better than the original long prompt? Not necessarily. Similar maybe. But sometimes I would get a surprise and the suggested prompt was better than the long prompt. I'm watching this tool because I can see a potential usefulness. And I want to know what the AI is thinking. This might change how I prompt especially if I'm getting crappy results.
I ussually use short prompts they may have unecessary words there though.. Results are good, Did not know about this feature, many thanks for bringing it to surface. Will be very useful to understand how Midjourney works.
Another amazing video!
Thanks again!
These videos are the best on the internet for this stuff.
Thanks!
I love when you reveal, “behind the curtain.” Here are the words Midjourney ignores. I can’t believe how complicated the prompts are. My experiments with Leonardo AI were a couple of lines…the one you show-a novel. Lots to learn on my part. Thanks for this video.
ChatGPT prompts can be wordy. I've been working on a script that shortens the prompts. But if you let ChatGPT go - it will write a novel!
90% of the time MJ will turn out a beautiful image, even with a two word prompt. However, if you're after a specific outcome that involves a combination of elaborate elements, characters, foreground, background scene etc, forget about trying to achieve it in one prompt. *Create your amazing character/subject in one image and your amazing background/setting in another image, then combine them using image prompts. It provides MJ with clarity about your desired style and composition. MJ is a visual medium that responds better to images than words.
I'd be interested to learn how MJ weights a single image prompt against the actual words it took to create it. My guess is that it uses less tokens for the image, than the actual prompt.
*Ed typo.
Good point about using images over text
Good point about the image prompts. They work differently in MJ than text prompts. You pose an interesting experiment!
@@MakingPhoto Thanks. Over to you on that one. 🤩
Excellent explanation. I would think it's a matter of taste and creativity as to deciding to use a long or short prompts. One additional consideration seems to be processing time for each token which I would think that depending on your computer (memory, ram speed, cpu , etc) which may or not be noticeable. I favor shorten prompts, but I look for words that gives as clear vision as I can. Then go for variations.
I agree. I tend to favor short prompts, too. /Shorten (and /Describe) is giving me great ideas that I hadn't thought about.
How do u add text weight into a shorten command? And would midjourney reevaluate which is more important after we added the text weighting? thank you for the clear presentation of the video tutorial,...
I did some testing with weighted words and /shorten did seem to pick up on the weights. It's been a while since I did a video on weighting words in prompts. I should probably update it. "Multi-Prompts in Midjourney :: Should You Use them?" ua-cam.com/video/ayIp712u-vQ/v-deo.html
6:56
We know this is not true as demonstrated in this video with "purple" being indicated to have 0.00% impact....... but the actual results are clearly drastically different to what the AI claimed.
We cannot trust/Shorten at this stage. Hopefully somewhere down the road it will be SIGNIFICANTLY improved to the point where it can be trusted. 🙏 I hope they can fix the flaws in /Shorten.
I think one issue is that the /shorten command was designed for long prompts. But I'm putting short prompts through it. This is worth watching as the tool improves.
Best explanation 👌 thank you
it has been helpfully in figuring why certain words don't get used as much as i would like, right now im trying to figure out how to hide images inside another image and you can image how that is going. For instance high school girl Japanese, standing in front of busted worn down series of mirrors in a run down bathroom, she is holding a camera up to take a picture in the image you can see a creature looking at her that not there. Something like that
Wow! You've got a real challenge on your hands! This is complex for the AI. Let me know if you find a prompt that works.
@@MakingPhoto yep and would make for good products as well
👋
I've done a fair amount of testing of the /Shorten command / directive and my conclusion is that it's not very useful since the results of using the prompts /Shorten generates are so frequently noticeably different and not necessarily better that it's actually not a very good tool to shorten very long prompts to consistently enough get good or better similar results.
Good to know! I sometimes get better results after shorten, but sometimes the results are similar - at least with the 1st two suggestions.
My prompts are really terse, all of the words seem to matter. Midjourney's suggestions tend to change the images wildly. Don't even start me on /describe, Midjourney doesn't know where to start. Generally the results turn out to be some kind of crazy mess!
I'm a short prompter, too. I'm not sure the suggestions improve my prompt, but I have found some OTHER prompts through these tools that I quite like.
4:18
👍 👍 👍 That's precisely what I pointed out to you in a previous video and why the analysis by the AI can NOT be trusted........ that and other reasons including the fact that shorten never really produces results very similar to the original long prompt.
So shorten is virtually completely useless if you want to consistently produce the same results as the original with a /Shorten prompt.
What if your prompt includes an image?
Some of the prompts will keep the image link as part of the prompt. But I noticed that 4 and/or 5 may not have the image url.
👍
4:40
No, they're actually NOT necessarily better. Better is entirely subjective and I actually prefer at least 2 of the images on the right produced as a result of the longer prompt.
I'm definitely not one of those people who thinks that all long prompts are necessarily better or good........I've seen plenty of very long prompts which are contextually absurd and produce either crappy or at best average results.
*My point is......there are much better and much worse very long prompts / prompting techniques........we need to distinguish between them and try not to lump all very long prompts into the same waste basket of poor prompting technique.* This seems to be especially true when considering very long prompts which use weighting. 👍 👍 👍
In my testing, I was usually ok with the 1st two suggestions. Were they always better than the original long prompt? Not necessarily. Similar maybe. But sometimes I would get a surprise and the suggested prompt was better than the long prompt. I'm watching this tool because I can see a potential usefulness. And I want to know what the AI is thinking. This might change how I prompt especially if I'm getting crappy results.
@@MakingPhoto
I'm guessing that the devs will improve this tool...... it has promise!
👋
I ussually use short prompts they may have unecessary words there though.. Results are good, Did not know about this feature, many thanks for bringing it to surface. Will be very useful to understand how Midjourney works.
Glad it was helpful! I often use short prompts, but it's interesting to see what the AI thinks is important.