Midjourney Vs DallE-3 Prompt Shootout!

Рет қаралды 37,508

Күн бұрын

Off the heels of the recent announcement that Open AI's Dall-E 3 will be integrated into ChatGpt, I thought I would do an old fashioned Prompt Shootout between the two AI Image Generators.
I always enjoy seeing the aesthetic differences between AI Image Generators, and I do not think one is better than the other, but please do leave a comment and let me know which you prefer!
PDF Segments included in this Video:
Midjourney Prompting: • A Perfect Midjourney P...
All Midjourney Dash Dash Commands: • Midjourney Control for...
Midjourney Tokens: • Midjourney's Amazing N...
Join the Patreon here: / theoreticallymedia
Thanks for watching Theoretically Media! I cover a large range of topics here on the Creative AI space, Technology, Tutorials, and Reviews! Please enjoy your time here, and subscribe!
0:00 - Intro
0:33 - Overview of Dall-E 3
1:18 - Text in Dall-E 3
1:35 - Examples of Dall-E 3 Images
3:24 - A nod to Dall-E 1
3:45 - The Prompt Shootout
4:46 - Prompt 2
5:51 - Prompt 3
7:00 - Prompt 4
7:56 - Prompt 5
8:41.- Prompt 6
9:41 - Prompt 7
10:27 - Is Midjourney Dead?

Пікірлер: 144

@SirQuantization 9 ай бұрын

AI time really does move differently. I can’t believe how simultaneously fast and slow this year has gone

@TheoreticallyMedia 9 ай бұрын

I legit did a double take when I saw that last shootout was in March! I was like...but that was...wait, was that a long time ago or not? Haha...Rubber Band Time, I tell you!

@Sparkette 9 ай бұрын

Even more so for me. My spiritual awakening happened this year.

@StefanReich 9 ай бұрын

@@Sparkette Good for you ☮

@StefanReich 9 ай бұрын

Fast, sure But why slow?

@tomoriah 9 ай бұрын

After having watched many of your videos, it is about time for a big thanks, man, keep it up!

@TheoreticallyMedia 9 ай бұрын

Oh, thank you so much! Seriously, that means a LOT!

@morpheus2573 9 ай бұрын

That's very generous of you. And well deserved by Tim who does an amazing job. 👍

@jaysonp9426 9 ай бұрын

Absolutely incredible. Your channel has become one of my favorites

@TheoreticallyMedia 9 ай бұрын

That is fantastic to hear! Thank you so much!!

@ManiSaintVictor 8 ай бұрын

Wow! Thanks for this! I didn’t realize a lot of the details about the integration with ChatGPT until you pointed them out

@TheoreticallyMedia 8 ай бұрын

I think that’s the real “win” here. Apparently Dalle3 will be available for free on Bing, but I presume will utilize a more standard method of promoting. Having a conversation with an image generator just sounds like a ton of fun and likely leading to some really interesting outputs!

@RocketQueen71 8 ай бұрын

I rather have THIS KIND OF RACE THAN A NUKE RACE or even a NASCAR RACE ( NASCAR SUCKS NOW!! WENT WOKE!😤😡)

@johnfromberkeley 9 ай бұрын

I appreciated your deep dive on this comparison.

@TheoreticallyMedia 9 ай бұрын

Thanks so much! I’ll do an actual deep dive on Dalle-3 when it drops!

@AllanMcKay 9 ай бұрын

great work mate! Definitely want to check this one out

@TheoreticallyMedia 9 ай бұрын

Thank you Allan!! I'm really looking forward to Dall-E 3 dropping for all of us. The idea of narrative/conversational prompting is going to be REALLY interesting!

@trevorama 9 ай бұрын

Great episode! And thank you for the several shoutouts to your vegetarian viewers!! Also, terrific interview on the Nerdy Novelist channel recently. 🙏🏼

@TheoreticallyMedia 9 ай бұрын

Oh, I loved being a guest on that show! It was such a good time! Glad you caught it-- and sorry that I said "um" so much! haha!

@WadeMcMaster 9 ай бұрын

Awesome video, DallE 3 is looking pretty amazing so far! It's a race for sure.

@TheoreticallyMedia 9 ай бұрын

For sure it is-- the great part is: No matter who crosses the finish line, we ALL win!

@RocketQueen71 8 ай бұрын

I rather have THIS KIND OF RACE THAN A NUKE RACE or even a NASCAR RACE ( NASCAR SUCKS NOW!! WENT WOKE!😤😡)

@WadeMcMaster 8 ай бұрын

Such a positive outlook, I can definitely appreciate that!@@TheoreticallyMedia

@WadeMcMaster 8 ай бұрын

@@RocketQueen71 100% over the nuke race!

@betabob2000 8 ай бұрын

On the mahogany gaming chair, those aren't straps to hold you in. They provide tension to hold the back support pillow at whatever height you want it at.

@TheoreticallyMedia 8 ай бұрын

I was really hoping it was for VR Rollercoaster Tycoon!

@lemmingdot 7 ай бұрын

Awesome video, thank you.

@obeyjoe9534 9 ай бұрын

Nice one Tim, its amazing how fast time moves in Ai, better not have a day off lol!

@TheoreticallyMedia 9 ай бұрын

Haha, there are days where I wish it did! Well, at least it seems to take a break on the weekends....for now!

@mindbytedigital 9 ай бұрын

Thank you for another great video friend!

@TheoreticallyMedia 9 ай бұрын

And thank YOU for watching and commenting!!

@mindbytedigital 9 ай бұрын

@@TheoreticallyMedia I enjoy your videos alot.

@AG_before 9 ай бұрын

8:05 In case _any one_ missed that, please read the sign! 🤯 Great video, as always! 👍

@TheoreticallyMedia 9 ай бұрын

Oh, THAT'S a good catch!! Man, DallE is impressive!

@electricdreamer 9 ай бұрын

I think what's killing midjourney is their price tag. 30 dollars a month is too much. You can buy 2 netflix sub with that. If they decrease it to like 10 dollars a month, that'll keep them in the game better.

@TheoreticallyMedia 9 ай бұрын

They had the $10/mo plan, but did they get rid of it? I’m curious if (when they get to) 3d and Video, if they’ll lower the price to an “Image only” tier?

@electricdreamer 9 ай бұрын

@@TheoreticallyMedia They have the $10 tier but the function is so limited it's not worth it. I think it would be better if they price the three tiers to $2.99, $9.99 and $19.99.

@OggieDoggie 8 ай бұрын

Wow I better check my sub. I was subscribed earlier this year but didn't know prices have gone up

@BastianAI 9 ай бұрын

Will be interesting to see how being able to have a conversation about what you're prompting will affect recurring characters and locations.

@TheoreticallyMedia 9 ай бұрын

That will be MASSIVE. I am curious, since the demo (which admittedly is a children's story) does illustrate that it is capable of doing so.

@zrezaeimobin 8 ай бұрын

detailed & clear 👍

@TheoreticallyMedia 8 ай бұрын

Thank you!!

@dnewman6 8 ай бұрын

I feel like midjourney really needs to pull away from being in discord to even think about be more mass market. It’s such a big hurdle that put any non techy off. Putting dalle 3 in chat gpt is 100% the way to go too. I feel like writing prompts at the moment is a skill in its self

@nicdunz 8 ай бұрын

the straps connect to the lumbar support cushion. its how a lot of gaming chairs are set up.

@TheoreticallyMedia 8 ай бұрын

Ooooooh, that makes sense. And also explains why I had stupid lower back problems for years (switched to a standing desk, which was a real game changer) And here I was hoping it was from playing a Rollercoaster sim in VR!

@jjjaaaabroni 9 ай бұрын

Hey Tim, great video my man. I was wondering if you are able to help with Midjourney making art of popular characters from tv shows/anime. Since it's been very inconsistent in my experience. For example, using Goku, Vegeta, and Future Trunks (from Dragonball Z) doesn't really work well. I feel they can capture Goku but the other 2 tend to not show in so well. Especially if you're adding them in altogether in one prompt. Maybe I'd need to keep expanding the image and reusing the prompt but changing it to one character at a time. Not sure.

@AbhinavPrem-fg2jj 6 ай бұрын

I don't think they are allowed to do that. That's why it is unavailable.

@mikeheffernan 8 ай бұрын

Bring it on!

@TheoreticallyMedia 8 ай бұрын

When the giants battle, we all win!

@harrygoldhagen2732 8 ай бұрын

Interesting, Tim! I'm curious, how does the new Dall-e handle consistency? The same mess as usual, or have they addressed that? For me, the AI that creates consistent characters that can pose and emote will be the winner. By the way, when I went to the Theoretically Media youtube home page, it says there is no content. Clicking on the Videos tab brings them all up, but that was weird. Which reminds me, hope your youtube troubles have been tamed.

@TheoreticallyMedia 8 ай бұрын

I guess we're waiting to see with the consistency. The demo video makes it SEEM like you can, but...well, we've been sold that a few times! That said, I hope to have a video up this week that....well, I'm not ready to say "solves" but is a good home-brew work around for the time being! I'll look into that home page issue. That's super weird! Eh, still have the whole YT Invalid Traffic thing. I did play with a few things, like turning playlists off (someone said that would help), but at this point-- I'm just going to turn them back on. There's been some recent noise on Twitter about it, that (may) have finally gotten KZfaq's attention. Fingers crossed!

@Glowbox3D 8 ай бұрын

Excellent video. I already have a Plus membership, so it makes sense to me to now cancel MJ on Discord. And the prompt style of GPT will undoubtedly feel better to me personally since I use GPT daily. I save $10 and get an excellent text to image at the same time. Win!

@TheoreticallyMedia 8 ай бұрын

I feel like we’re very much entering the age of…almost like streaming service. One month of Netflix, binge, then onto Disney… There are so many different generators, you almost have to play that game!

@Glowbox3D 8 ай бұрын

True. These new multimodal models are just moving so fast it's hard to know who's going to be the best and the least expensive, and for how long. Who knows what multimodal AI assistant will be available in five years. Good thing for us, they'll have to be competitively priced, because the other brands will be on each others heals. @@TheoreticallyMedia

@gbpferrao 9 ай бұрын

I was just now in the middle of thinking about how I wanted it to be interactive and change things based on future comments like "make it more ______," without having to save the image and include it in a new prompt with another description from scratch.

@TheoreticallyMedia 9 ай бұрын

I think that might be coming in v6. They have mentioned a big update to the language model, but at the same time, there’s no way it is going to beat GPT4. Baby steps!

@TelcomTransmissions 9 ай бұрын

Nice to see they’re giving artists the ability to opt out the datasets. That should deal with some of the ethical qualms some folks seem to have.

@TheoreticallyMedia 9 ай бұрын

To play devil’s advocate here, I don’t think the actual training data is going anywhere, it’ll just be the name locked off. So, I won’t be able to say: “a photo of a movie by Tarantino” but, I think I could get “90s indie crime movie, starring seven men in dark suits with skinny black ties” and get the same result. …but you didn’t hear that from me!

@TelcomTransmissions 9 ай бұрын

🤔

@garypick 9 ай бұрын

If the 3D feature could be used in 3D printers, that could be amazing.

@TheoreticallyMedia 9 ай бұрын

So. I actually don’t think it is going in that direction. But I think it could be used as part of the process. You could generate a turnaround in MJ and then bring it in to 3D software to map it. But I don’t think it’ll happen out of the box. At least to start.

@FilmFactry 9 ай бұрын

DallE wins on composition and creative camera angles.

@TheoreticallyMedia 9 ай бұрын

That reflection/camera shot was really impressive, I gotta admit! The diorama as well!

@christopherd.winnan8701 9 ай бұрын

Alan Moore's Amazing appearance works well on many prompts on all ai art generators. Here is one to try - Alan Moore detailed atmospheric portrait, 2k uhd, intricate embellishments cosmic environment, rotten fallen archangel, ((yellow glowy eyes)), biomechanical, eerie, creepy, nightmarish, very bright colors, light particles, bioluminescence, Mschiffer, wallpaper art, UHD wallpaper I also enjoyed the mahogany chesterfield gaming chair. It does really well at taking ordinary household objects and transforming them into fantastical contraptions. More of these too please!!!

@TheoreticallyMedia 9 ай бұрын

I can only presume that Alan is cursing all of us with a pox right now. …eh, we’ll at least we’ll know it’s happening when the whole of DC turns into snakes. …and deservedly as well. The Watchmen deal is terrible. I’ll try out that prompt for sure!

@christopherd.winnan8701 8 ай бұрын

@@TheoreticallyMedia He is probably too busy installing megaliths into the labyrinthine network of tunnels that he has excavated under his Northampton terrace. I would be surprised if he even admitted to knowing about KZfaq's existence....;-)

@thewebstylist 9 ай бұрын

As much as love MJ 100 (I’m actually taking the Masters of MidJourney course rt now) LUV the OG Dalle joining forces w my pair ChatGPT platform, double the productivity!

@TheoreticallyMedia 9 ай бұрын

I love the example of using ChatGPT/Dall-E to make a kids bedtime story. It's just so damn cute. And I think that's going to be a really interesting approach with Dall-E, that idea that you're "talking" to it. I do love MJ's hybrid Prompt/Command format, but the idea of just freeform narrating an image does sound really cool!

@bellsTheorem1138 8 ай бұрын

The straps are there to hold the lumber support in place. I have a game chair that has those straps.

@TheoreticallyMedia 8 ай бұрын

Yup! Few people have mentioned it-- I was hoping it was for playing a Rollercoaster Sim in VR!

@bellsTheorem1138 8 ай бұрын

@@TheoreticallyMedia lol.

@sooool4716 8 ай бұрын

Lol I like how everybody just forgot about Midjourney v6, just wait :)

@TheoreticallyMedia 8 ай бұрын

Oh, I haven’t! I can’t wait for it! I do think that Dalle’s integration with ChatGPT is going to be a superior Language model to whatever V6 has (I mean, GPT4 is still a monster LLM), but MJ’s will likely be very good as well. It might be interesting to generate in Dalle using natural language, and then use that as an image reference in MJ to “finish” it. There’s something I can’t wait to try!!

@richardglady3009 9 ай бұрын

Wonderful and informative video. Great introduction to Dall E. I really like the ending…if you have the resources (i.e. money) to subscribe to both services. Thank you for this video and all the hard work required to create it. I really enjoy your videos.

@TheoreticallyMedia 9 ай бұрын

Hey, thanks so much! On the plus side (financially) it looks like Dalle-3 will be available via Bing as well. My guess is, the image generation model will be there, but the chat functions won't be...I mean, otherwise, why sub to ChatGPT? So, you'll be able to use it (for free), but rely on a more standard prompting style.

@richardglady3009 9 ай бұрын

@@TheoreticallyMedia Thank you for taking the time to respond…with even more information. Thanks for you willingness to help others.

@funcorporatelife2177 8 ай бұрын

for the thunderstorm image with the coffee mug MJ v 4 looks way way closer to the image than MJ v 5, try it and youll get much much better results

@TheoreticallyMedia 8 ай бұрын

I LOVE v4! I’ll give it a shot! I always say: v4 had some special sauce to it! And props to MJ for continuing to let us use the old models!

@OggieDoggie 8 ай бұрын

Best video but I think the straps are holding the lumbar cushion in place. My gaming chair has the same thing. And if that's the case that's even better that DallE-3 knew that and put them in the picture.

@TheoreticallyMedia 8 ай бұрын

Thank you! And yeah-- haha, I guess that goes to show where I am with chairs. I was thinking it was supposed to strap you in. Like you've got a VR headset on and you're playing No Man's Sky. DO A BARREL ROLL!

@Beyondarmonia 9 ай бұрын

The Dalle 3 images also missed stuff. For eg. the river was a normal river and not made out of meat like it was specifically asked for. Also no dumbstick trees.

@TheoreticallyMedia 9 ай бұрын

yup. And that's why I'm kind of concerned about inpainting in Dall-e 3. If it is text based, like "Make that Tree a drumstick" I just don't think it'll work well. You need contextual boxes in the image to direct the bot. I mean, heck, even if I'm talking to a person, I sometimes have to point and say "this thing!"

@commonsensegaming1756 8 ай бұрын

also blue willow is a nice image AI. COMPETITION IS GOOD

@user-tp8il8sd3b 9 ай бұрын

ahah, you get 12 points for the Alan Moore reference :))

@TheoreticallyMedia 9 ай бұрын

Haha. I knew someone would it be there for it!!

@louie8362 8 ай бұрын

If Dall-e 3 can portray 2 characters together accurately than I'm hopping ship, ⚓pun intended

@TheoreticallyMedia 8 ай бұрын

Haha, fingers crossed! We'll see. I'm cautiously optimistic about it, but not willing to put any money down until I can see it in action. We've all been burned by the hype train!

@Yottenburgen 8 ай бұрын

It would be interesting to attempt to sort of translate them to eachother. Like you take the DallE prompt and toss it in midjourney, then try and tune the prompt so that Midjourney achieves DallE's output and DallE achieves Midjourneys prompt. If its easy for DallE to achieve midjourneys look ontop of its new spacially aware generations then midjourney is pretty dead.

@TheoreticallyMedia 8 ай бұрын

The thing I can't wait to try is to natural language prompt in Dalle, then take that image to Midjourney and play with it there. Use it as an Image Source, in/out paint, etc...I think combining these two are going to result in some really creative works!

@Yottenburgen 8 ай бұрын

Yep have DallE spacially map the image and provide all of its nice little details, then "upscale" with Midjourney. Going to have great results on its own, but even better when paired as per usual.@@TheoreticallyMedia

@WifeWantsAWizard 8 ай бұрын

(2:38) My question is will DE3 tell you when it refuses a token because the artist has opted out or just ignore you and pray you don't notice.

@TheoreticallyMedia 8 ай бұрын

Probably not? But, given that it is hooked into ChatGPT, you could always ask....although it is also hooked into ChatGPT, so it might lie!

@1FotoJadamo 9 ай бұрын

Unstable Diffusion also disabled certain celebrity names. But I found a work around.

@TheoreticallyMedia 9 ай бұрын

Do tell, when you have a moment!

@1FotoJadamo 8 ай бұрын

I'm not gatekeeping for the sake of gatekeeping, however, every time I divulge hacks, programmers eventually get wind and fix the "bug" and the hack is no longer workable. @@TheoreticallyMedia

@RocketQueen71 8 ай бұрын

Don't let PETA see that one DallE-3 image!!!! 😆

@TheoreticallyMedia 8 ай бұрын

We also don’t want that department of sanitation seeing it! That place would get a F in the window for sure!!!

@RocketQueen71 8 ай бұрын

@@TheoreticallyMedia You mean the Health Department .😆

@TheoreticallyMedia 8 ай бұрын

@@RocketQueen71 yup! I could not for the life of me remember which agency handed those out! Also…some of the best tacos I’ve ever had came from a place with a B!

@RocketQueen71 8 ай бұрын

@@TheoreticallyMedia Yes! I can definitely agree with you on that. Alot of B or even C places have the best food They shuy down this 1 place by us , a Mexican resturant that was by a Clearview Cinema (that's shutdown also) my hubby and I went there alot. We didn't have to wait for a table, not too may people, they had beautiful Mexican deco and THE FOOD WAS DELICIOUS....Thanks to the pandemic, they had to close it. Sad

@Mixxwizard 9 ай бұрын

Here's my 2 cents, the images from Dalle3 look nice. After using mid-journey for a while now. If you use the words photo or image in the prompts. That makes a huge difference. Plus, describing the lighting you are looking for in your image also helps.

@Mixxwizard 9 ай бұрын

Great video, Tim, on another note. Chatgpt & Elevenlabs is a dream team for my work with clients. If they only knew😂

@TheoreticallyMedia 9 ай бұрын

100%! I don't know if you caught it, but I have a whole video on prompting strategies in MJ: kzfaq.info/get/bejne/oJN3jdmJzMzLkY0.html --But, I think we're already on the same bandwidth here!

@Mixxwizard 9 ай бұрын

Thanks for the link. Because of that video of yours. My images are amazing. I have a few clients of mines. That's blown away by them. 😂 Being around computers & new technologies is not their thing.

@Mixxwizard 9 ай бұрын

I might not comment on all of your Mid-journey videos. But I do indeed watch them all & learn something new each time. Thanks

@LouisGedo 8 ай бұрын

*It's truly a major misstep in direction for certain AI's to put a kibosh on prompting artist names. What's next, architect, sculptor, and photographer names as well?* Not good at all.

@TheoreticallyMedia 8 ай бұрын

oh, I'd imagine that is all happening. It'll just be work arounds. Since you can't say Guillermo Del Toro, it will be "Dark Fantasy Cinema, 2003 era, natural gothic" etc...you'll still get the same results, just without the name drop. Yeah...it'll be annoying.

@johnpapadopoulos5687 9 ай бұрын

I am not really sure that head to head comparison is the correct way to go about this. It is more than established by now, that Midjourney needs a specific way of composition when putting a prompt together and if you can tune to this, you get pretty good results, every time. But that doesn't mean that this "specific way' would work somewhere else (and vice verse).

@TheoreticallyMedia 9 ай бұрын

Oh, that's mentioned in the video. A few times in fact. This is more just for fun. In fact, I conclude by saying this is a win for everyone. It isn't really a "Battle"

@sidewaysdesign 9 ай бұрын

True - MidJourney steadfastly centres the subject, whereas there seems to be a more varied compositional arrangement in the DALL-E examples.

@sparklzdiamondgirl7502 8 ай бұрын

Dall-E is so limited through Chat GPT and I like the photos ins MJ so much better. I keep getting (that goes against standards). Um. It shouldn't! I don't do inappropriate pics!

@holdthetruthhostage 8 ай бұрын

Midjourney Lost no need to sugar coated, the fact that Midjourney can't understand longer text while saying it can, this is bad for them especailly since Dale 3 will be Chat GPT, Bing, & Windows 11. Midjourney 6 better be ground breaking because closing up their API is catching up to them

@TheoreticallyMedia 8 ай бұрын

Agreed. The thing I always caution is that MJ is a sleeping giant. Every time someone pronounces them dead, they come out swinging. But, agreed on V6. They need something that’ll wow. Also, while I don’t mind Discord, they need to get that web platform up and running.

@holdthetruthhostage 8 ай бұрын

@@TheoreticallyMedia thanks for the reply, I think Mid journey not opening up their API is what's leading them to obsoletion, they are to private with their tools. Could it be that Mid Journey is just a modified Stable Disfusion or not as new as we expect. Something is very fishy about Midjourney privacy of API

@3dus 9 ай бұрын

I understand why they are doing it but the art world works by naming references. Very hard to see something that has yet to be realized without any references.

@TheoreticallyMedia 9 ай бұрын

Agreed-- there is so much referencing with art...I'll say, I think it just turns into code-words. Figuring out alternate tokens for the look you're going for. Like, instead of G. Del Toro, it is something like "Dark Adult Fantasy Film with Natural Tones and an underlying element of horror, shot in the mid 2000s" ...that's a bad example, but you know what I mean...

@googleyoutubechannel8554 7 ай бұрын

So many softball prompts.... Why not try _any_ of the prompts we all know generative networks struggle with? Any specific multi-human interaction prompts, non 'straight-on' shot orientation, prompts that can't be pulled straight from a image label

@TheoreticallyMedia 7 ай бұрын

Haha, well for demo purposes, I’ll give it the biggest area of use case, but yeah- I get you. Trust me, the minute I run across an image generator that’ll give me a Cyclops, I’ll start shouting from the rooftops!

@JackMosel 7 ай бұрын

Sik Journey 🎉 Huh?!😅

@user-dn6gp2du2h 9 ай бұрын

No more midjourney time to use dalle 3

@TheoreticallyMedia 9 ай бұрын

I think there will be a lot of others that also follow suit. But, I also think the inverse: people will start on Dalle3 and then move to MJ. It’ll be a bit of a churn for sure.

@swoletech5958 9 ай бұрын

Midjourney is in trouble. Also most folks are not vegetarian or vegan…that meat looked deeeeelish.

@TheoreticallyMedia 9 ай бұрын

Haha, I eat meat as well, but I have a number of friends that are Vegetarian and Vegan, so I always like to throw them a....ahem, bone. To be honest, I once went a bit as a Vegetarian, mostly when I became obsessed with Indian food...It wasn't too bad!

@swoletech5958 9 ай бұрын

Haha, I got you. Great video by the way too man. I’m excited to try out Dalle3 once it’s available.

@TheoreticallyMedia 9 ай бұрын

@@swoletech5958 same! News just broke that it’ll be available in Bing as well, but probably more as a standard prompt kind of thing. The “chat with it” will likely stay relegated to GPT4 (where the money is!)

@rendermanpro 9 ай бұрын

Everything is very simple. MJ is a community, people, friends, warm and nice relationship like fellows or a family. DallE is just a cold soulless corporate machine. Kind of alive thing vs piece of equipment.

@TheoreticallyMedia 9 ай бұрын

The most unintentional brilliant thing about MJ is that it began on Discord. I don’t think the social aspect was part of the master plan, but it ends up being their Excalibur.

@quercus3290 8 ай бұрын

"pick out key tokens from it" See, thats the problem with all these text2image tools, you can kid yourself on and write a highly descriptive prompt, however half of its never understood, grammar is pointless. You need to just punctuate a few key descriptors the LLM understands and thats that. Whatever words the model was trained on during tagging, thats all that matters.

@TheoreticallyMedia 8 ай бұрын

That’s 100% true. I do think that an LLM will be better at contextualizing what you’re looking for in a prompt written in natural language (MJ isn’t really great at that), but it does ultimately boil down to the fact that the image generation side knows what it knows. I recently did a thing where I needed a cyclops. For some reason, most image generators hate a cyclops. I could write a 70k description in ChatGPT for Dalle3 about a cyclops, and it still couldn’t give me one. I ended up bringing a standard 2-eyed Troll into PS and tried Gen-Fill. Still: Nope. So; it was finally down to old school Photoshop work to attain the result. If anything, I think that is the mantra of this channel: use AI as a tool, but you still gotta know the manual labor tools! The alternate mantra: Ai hates Cyclopses.