50 Prompts, 5 NEW benefits: DALL-E 3 vs Midjourney 5.2 | The Ultimate AI Art Showdown

[@AppyDave]

Links in the about tab on my KZbin channel, but here they are as short cuts. twitter.com/AppyDave, youtube.com/@appydave, youtube.com/@flivideo, youtube.com/@winningprompts

[@jimmortalitytech]

Thumbnail art was beautiful, pulled me right in. Great video!

[@AppyDave]

:) Thanks

[@Bensboyforreal]

Your videos are very professional and clear. Love the structure and design of your videos

[@AppyDave]

Thank you Ben, I keep improving my design with each video.

[@AppyDave]

I plan to release a video this week on how I put together the video structure using ChatGPT

[@GiGiNevarez]

Thank you

[@AppyDave]

You are welcome

[@jimcrawley7363]

Very impressive! And I speak as someone who has generated over 26,000 Midjourney images. Just getting into Dall-E3. I have begun conversations with the system about how it interprets the text and provided examples of token weights from MJ. It confirmed that it works in a different way and I am exploring what it recommends as a prompt crafting strategy. Thank you for all the valuable information, so well presented.

[@ewallt]

A few things which might be helpful. 1. DALL-E doesn’t allow you to explicitly reference artists active in the last hundred years, but you can work around this pretty effectively by having ChatGPT write out the characteristics of the given artist’s style, and use that in place of the artist’s name. 2. DALL-E is a lot better at doing what you ask it to do, especially where locations are concerned. For example, if you ask for someone overlooking Yosemite with Half Dome and El Capitan in the distance, you’ll get that. And a prompt for Pont Alexander bridge in Paris. Unbelievable! I felt like I was there. 3. I’ve found the black and white things to be excellent in DALL-E, like old black and white photos, graphite drawings, etc. 4. Things like digital painting, vector art, line drawings, it captures the style more accurately than mj often, but more simplistic. It’s so much better than v2, I wonder what v4 will be like.

[@AppyDave]

Thanks Jim, I would be interested to hear how you go with your prompt strategy

[@Sweettooth1231]

Really good well put together video. I like how you explain stuff every now and then in the middle of the images instead of doing an info dump in the beginning. I hope you can do a similar video with stable diffusion XL! although that'd be unfair since XL is in most cases inferior to these two...well, if you can find it a worthy opponent that'd be awesome

[@AppyDave]

Much appreciated! I'm glad you liked the format with the different tangents along the way, I'm still developing my video style and this was new for me.

[@onlyyoucanstopevil9024]

THE BATTLE OF THE LEGENDS 😊😊😊

[@Enu_Vibe]

Great comparison, I personally enjoying DallE. The more your prompt are detail the better results you get.

[@AppyDave]

Thanks Enu, I am impressed by DALL-E3 as well

[@jonathanberry1111]

@@AppyDaveWhat about the "Slow is the new Sexy"! for the Sloth on the vogue cover, amazing that that is created by the AI, it literally thought up Sloth related magazine sounding catch phrase! That is pretty stunning!

[@lbzorz]

Great video!!! The effort shows!

[@AppyDave]

Thanks for the comment, I plan to release a video this week just talking about how I put this video together using ChatGPT

[@lbzorz]

@@AppyDave brilliant idea I think!

[@stefanguiton]

Excellent

[@AppyDave]

Thanks

[@teolita]

With the same prompt MJ will give us more different results each time. There is more variety in style, scene, colors... I prefer 90% the images of MJ. Master work

[@kimastor9753]

Totally agree! MJ is without competition from dalle3 in this regard. I have dalle3 through chatgpt plus, for me personaly, it is so limitet that i see no point in using it. Truly dissapointed...

[@kimastor9753]

Another thing that is quite annoying is that it modify the prompt by default, introducing specifics you did not ask for and kind of faking variety in output. Forcing dalle3 in chatgpt to use the exact same prompt for all four images and they will end up extremely similar..

[@AppyDave]

I agree that MJ is still a better product, but the way technology moves forward, I can see DALL-E 3 displaying it a little, not because it is a better image generated, but because it integrates easily into workflows via the API, Bing and ChatGPT. As a software developer, I have been begging MJ to releases an API and so far this request from a lot of people is falling on deaf ears

[@AppyDave]

That is an good point

[@ewallt]

@@AppyDaveFrom my experiments, it depends upon what you’re doing. There are some things DALL-E is much better at, including certain art styles and getting what you ask for. Given the improvement from v2 to v3, I wonder what v4 will look like. I also wonder if ChatGPT will get involved in the process of improving DALL-E. That can’t be far away.

[@jugibur2117]

Really great comparsion, thanks for your effort! My experiences with ChatGPT 4 in terms of programming were partly not so good, there were among other things Lua, Photoshop Scripts and Bash Scrips. I was amazed at how "confidently" ChatGPT suggested various solutions to me, some of which were completely wrong. Afterwards, however, she always apologized dutifully. However, the results were better with JavaScript, probably because it is more common. PS: At 7:53 I had to smile at how you (or you English-speakers) pronounce the german company BRAUN, because it's actually just our word for "brown" and is pronounced the same way.

[@nunokel]

For me, lua code worked very well, I was able to make some cool scripts for FiveM.

[@ewallt]

She? That’s the problem. My version is male.

[@isaaccrum3628]

Great work. I have a lot to learn.

[@AppyDave]

Thank you and, You can do it!

[@robo3383]

Good stuff, thanks for taking the time to put together this comparison. IMHO the AI thinks Dobby is Dobie which is a nickname for the Doberman breed, like Doxi is a nickname for the Daschund breed of dog. I could be wrong..

[@AppyDave]

I was thinking it might have been doberman myself

[@DreamDetective]

Links to your other KZbin sites would be good. I wanted to check them out.

[@AppyDave]

Sure things, I have goto the links in the about tab on my KZbin channel, but here they are again. Twitter: twitter.com/AppyDave, youtube.com/@appydave, youtube.com/@flivideo, youtube.com/@winningprompts

[@daveonpurpose]

This is gold!!!

[@AppyDave]

Thanks

[@FrankysChilloutTime]

I'm always amazed at the new applications of AI. The future is now!

[@stacyruiz5001]

Dobie is also short for Doberman breed of dog

[@AppyDave]

makes sense

[@isaaccrum3628]

#29 Dobey is for Doberman pinscher a type of dog.

[@AppyDave]

Thanks for the info

[@erikaol1091]

I really hope they do something about the constant unsafe image results and let us change the aspect ratio. These two things are the reason I'm sticking with midjourney.

[@AppyDave]

I'm sure these issues will resolve soon, the tool is only a week old, so just needs time I guess

[@b3rt33]

Yup it's the --ar functionality I really need

[@HCforLife1]

The pixelart works well in Dalle - comparing to midjourney is a lot better. I think Midjourney v4 was doing also decent ones.

[@AppyDave]

agree

[@1996gameking]

In number 17 Dall-E is actually doing better cause you asked for military emblems and the stuff on the left is way to detailed for emblems. Just cause it looks more like the reference doesn't mean it did better with the prompt.

[@AppyDave]

Thats a good observation, thanks for the feedback

[@Daz19]

Dalle 3 gets much closer to the promt espqiclly when it co tains mutiple subjects, but Dalle lakes the abilty to iterate as eaisily. I think MJ does slightly better lighting and composition, but its marginal.

[@AppyDave]

Good point

[@lawrenceclark7079]

dobey is short for Dobermann (a breed of dog)

[@AppyDave]

Thanks Lawrence

[@HCforLife1]

One thing to remember - dall-e won't generate the images of an artist who's last work was less than 100 years. So if you are referencing that artist - it won't be taken into consideration.

[@AppyDave]

That's interesting, people love that capability in MJ

[@kimastor9753]

It still refuses to use artists that died 100 plus years ago... at least when used within chatgpt

[@AppyDave]

@@kimastor9753 interesting, I have not tried it out like that

[@Brismo7]

19:54 Since Dobby isn't an actual word but a made up name from Harry Potter. I assume when Dalle sees "Dobby" and thinks you probably meant, "Doggy".

[@AppyDave]

That makes sense

[@teolita]

Maybe there is a copyright reason to not wanting to develop the text feature in MJ.. Putting Prada, Vogue, Braun on a image could be a copyright issue..

[@AppyDave]

I don't really know, that could be, but I know my use case is to put text on imagers for thumbnails and my own content and I look forward to any image generator that builds in that functionality

[@danielboeren7715]

Firefly images are on par if not better than both of these

[@AppyDave]

I havn't used it for a couple of months, but I will check it out

[@danielboeren7715]

they have a new v2 model which is way better than the original @@AppyDave

[@MikeKleinsteuber]

This is not a good comparison because both require different prompting techniques. And both have their own peculiarities, own requirements and generally produce different types of images. There is not a one size fits all answer here.

[@AppyDave]

I hear you Mike, that is a fair comment. When I recorded those comparisons DALL-E3 was released only 24 hours earlier so my focus was not on learning the idiosyncrasies of prompt writing for the new release, but instead just seeing what would happen if I used good quality MJ5.2 prompts on DALL-E3. I know that more balanced video would have taken into account prompts re-written to take advantage of the target platform. But to be honest I was very impressed with the generally quality of DALLE with prompts that were not tailored to its strengths, I guess I could have been a little clearer in the introduction about that

[@khanisbrown6424]

Is it just me or does dolle e 3 produce images with not very good colours and a lot of noise. Sometimes the colours are very good but most of the time the coloring and graphics isn't on point, try comparing the colours with ideogram, they are far more brighter and better even though Dall e produces better, accurate images but when ideogram gets the prompt right, i prefer it than Dall e3 just that ideogram doesn't give accurate generated images but the coloring still awesome.

[@AppyDave]

I didn't notice, but I think I will check into that

[@khanisbrown6424]

@@AppyDave thanks, please I'll love to get your feedback

[@sigmarealML]

links?????

[@AppyDave]

Sure things, I have goto the links in the about tab on my KZbin channel, but here they are again. Twitter: twitter.com/AppyDave, youtube.com/@appydave, youtube.com/@flivideo, youtube.com/@winningprompts

[@AdamTwardoch]

necesita avatar de estela soberano just means I need an avatar of Estela Soberano. Estela Soberano is a person. Amazing how human intelligence these days is often clearly inferior to AI.

[@chadclay1643]

You realise Ai gave him that dodgy translation ?

[@krzysztofparda6482]

The downside of Dalle 3 is that, according to the regulations, you cannot use images for commercial purposes, making it somewhat useless.

[@AppyDave]

I did not know that, thanks for the info