#### Win a RTX 3080Ti #### Sign up FREE to Nivida GTC for a Chance to Win a RTX 3080Ti: nvda.ws/3D7nNY4 #### Follow Robomar #### Follow Robomar on Instagram: instagram.com/robomar.ai.art/ #### Join me #### On Facebook: facebook.com/groups/theairevolution On Discord: discord.gg/XKAk7GUzAW
I wish they actually gave away 3090 because I already have got 3080ti. The only reason why I would upgrade is the need for more VRAM. I bought my card before AI image and language models were really a thing so I didn't foresee the situation where 3080 TI's 12 gigs wouldn't be enough for me. At least the AI stuff for us doing on my computer a year ago is way different from what I'm doing these days. And for me 12G definitely ain't enough at the moment. In fact I will probably even save money to get myself a 80 GB A100. For many years I've already valued VRAM size more than te number of cuda cores. For a while I even used Nvidia Tesla k40 with custom 3D-printed cooling attached to it as Tesla cars are usually used in servers in data centers.
@@Dazreil True. Discord as a whole was at breaking point yesterday, wouldn't even load some images for me, so who knows, maybe there were many more of the 13+ million users playing with the new model.
Great video as always Olivio and a mention to Robomar for his excellent prompts as always. Again, I've typed up the prompts for the community to save time, enjoy folks. catholic nun from 1780, old Whitestone village background, ultra realistic, octane render --ar 6:4 --v 5 bald man with a beard, wearing a tuxedo --iw 1 --v 5 intricate details, skull made of gold --ar 4:5 --v 5 woman fashion model wearing 60s clothing, comic book, sharp focus --ar 4:5 --v 5 Lovely anime Cyberpunk City Painting, Diagonal Camera Angle, Dutch Angle, Volumetric Lighting, Digital painting, night, Marc Allante, Studio Ghibli, Impressive, Intricate detail, Alejandro Burdisio, Dan Luvisi --v 5 --v 5 --ar 1:3 --chaos 20 ar 1:2 --chaos 20 a set of 9 startup icons with a modern, flat icon design --ar 4:5 --v 5 yellow Mercedes Brabus standing dirty in the jungle, moisture and haze in the air, professional vibrant grading, soft shadows, no contrast, clean sharp focus, commercial photography --ar 6:4 --v 5 The hospital room of the future, in a futuristic spaceship, with calming bright lights and a welcoming wooden floor, a doctor and her patient look at a computer screen showing medical graphs --ar 6:4 --v 5 modern hotel room --ar 16:9 --v 5 imperial Palladin in the detailed filigree armour, fine illustration, ultra sharp --ar 4:5 --v 5 Sprite sheet with game visuals, game assets, various prompts, 3D exploring game, cinematic lighting, designed, pc game, beautiful, manga, unreal, 64k, enhanced, intricate pizel art --v 5 detailed blacksmith room interior, sharp focus, isometric diorama, octane render --v 5 expression character sheet --ar 4:5 --v 5 sea town, white background, cartoon editorial --ar 4:5 --v 5 Futuristic woman, sharp focus, white background, comic book --ar 4:5 --v 5 old car from 1970, color pencil drawing --ar 6:4 -- v 5 mountain rescuer backpack, symmetrical placement, knolling, sharp focus --ar 4:5 --v 5 --v 5 Full body shot of a cute little anthropomorphic fox wearing a adventurer outfit, style of Pixar, character design art, by Jean-Baptiste Monge, highly detailed, digital art, cinematic lighting, epic, on white background, ultra realistic, vray render, expressive pose, professional color grading, soft shadows, no contrast, clean sharp focus --ar 4:5 --v 5 crouching tiger on the hunt for his pray, national geographic photography -- ar 6:4 --v 5 modern street hustler, sharp focus, renaissance painting --ar 4:5 --v 5 impressionism dark cloudy sunset over the stormy ocean shore, huge waves, the ocean playing jazz in coalt colors, TT --ar 2:5 --v 5 dress made of mirrored scales, high fashion, elegant, beautiful, model --q 5 --ar 9:16 --v 5 ancient aztec teenagers, traditional clothing, waiting around --v 5 A smiling little girl in a red coat and hat walks in front of a giant panda with a bamboo forest in the background, delicate features, rich colors and details, art by shinkai, studio Ghibli, highly detailed, HDsmooth --v5 --q 2 --s 750
So viele V5 Videos die ich jetzt gesehen habe, aber keines war so smooth und informativ wie deines. Danke für deine Arbeit und die Mühen! Zumindest für mich bist du auf AI-Art Ebene ein echter Mentor. Das wollte ich einfach mal gesagt haben, weil ich nach lengerem zögerlichem rumprobieren in den letzten Monaten, vor einigen Wochen so richtig in die Materie eingestiegen bin und einen Großteil meines Wissens dir verdanke. Danke Mann, Liebe geht nach Wien! 😉🤙❤
Danke, das freut mich sehr das ich dir mit meinem Video helfen konnte und es genau das ist was du gesucht hast. Viel Spaß in der Welt von AI :) Ich hab noch viele andere Videos zum inspirieren. Schau dir mal mein Hyper-Realism in Midjourney video an :)
lol, how cute. Wait til the mass layoffs start and then the riots and mass shootings because people have no purpose or livelihoods and are left in the gutter by corporations and the Government. We're not going to last another 50 years.
Yes for everything you'll get the results as you described in the prompt. As you say octane render will give you cgi look, photography will give you photo look.
yes, sorry. I wanted to do a quick video and then it took ages to record and edit this video. I also had a day filled with tasks. Really happy i could publish this
I simply can't give enough praises to you and others like you that understand the potential and power of AI. It's amazing how I see other artist actually believing that AI is just a phase, and it will pass over soon. AI is here to stay and it's evolving at a pace that few can even keep up with. I can say with comfort that you'll be sitting in the captain's seat when AI is in full control.
Thanks so much for this tutorial, Olivio! I've been trying to design cartoon backgrounds for my animations. I'm trying to figure out codes/ways to prompt Midjourney for certain perspectives of a room (for instance, Midjourney often shows me overhead views when I'm looking for straight on, eye level view). I'm also wondering if there are good prompts for Midjourney to share multiple walls of the same building space so they all work together congruently in the animation. So far I've been struggling to find the right words in midjourney to make this happen. Thanks for your help!
Loved this video! Thank you for showing us what V5 can do. You mentioned that game assets are something you enjoy making in Midjourney. Could you maybe make a video dedicated to showing the types of prompts that have worked well for you in the past?
@@OlivioSarikas That's great! I've tried making icons/buildings/units and stuff to be used as board game assets, but I think my prompts need refining. Looking forward to it!
The main source of any ai image generator are the images already made all across the internet. MJ is not rendering something from scratch, but rather from compilling pre made photo elements and combining them.
That's like saying you can't make anything new, because all you know is learned from the people who lived before you. When in fact we have seen many many new works generated by AI in a way a human would not do that. Sure, that was by chance, but still that counts
Mid-journey is a new medium but they are not photos. I call them digital illustrations. I also recently used Blue Willow V2 and I got more better-quality images with more aesthetic outputs
V5 made major strides in dealing with hands. The problem I'm having is how realistic everything is. Even specifying digital painting doesn't work out that well. I was mostly using Niji since it's more in the line of the style I'm going for. Hopefully their V5 version is just as good.
Thanks for amazing content, I noticed something, I have generated the prompt for MJ 5 in ChatGPT4 when you change the order of the description in the prompt, the output change completely, so the focus will be on the beginning of the prompt Example 01 Sharp focus, isometric diorama, octane render - Introduce the main character, John, as he walks through a bustling city street filled with people and vehicles. He is wearing a blue jacket and carrying a briefcase. The scene is viewed from an isometric perspective, showing the tall buildings and cityscape in intricate detail. The lighting is natural and balanced, highlighting the rich textures and vivid colors of the environment. John looks determined and focused as he navigates the busy street. --ar 16:9 Example 02 The main character, John, as he walks through a bustling city street filled with people and vehicles. He is wearing a blue jacket and carrying a briefcase. The scene is viewed from an isometric perspective, showing the tall buildings and cityscape in intricate detail. The lighting is natural and balanced, highlighting the rich textures and vivid colors of the environment. John looks determined and focused as he navigates the busy street. Sharp focus, isometric diorama, octane render --ar 16:9 --v 5 Please advise if you have any tips regarding this
I got amazing results for the following prompt: In the mind's abyss, where thoughts reside, Images take shape, and truth they hide. You believe you think, and so you see, A world of shadows, where light should be.
V4 was meant to replace artists. V5 was meant to replace photographers. I don’t know if the realism makes things better or worse. The mutilated fingers, extra limbs, or missing limbs is more unsettling to me now that the images look more realistic. It’s like you’re looking at a real mutilated person. No thanks. If anyone needs me I’ll be in the lab with stable diffusion. I’ll check black when V6 is out
I’m very indifferent with V5. Sometimes it’s great & other times, not so much. Feet are still a foreign concept but the devs consider feet NSFW, so stupid. The major issue is the upscaling is disabled, luckily they said it’s temporary. Lastly, they just soft banned Chest with the latest update. I hate HATE fake artists that are offended by the female or male body. I’m not talking about XXX I’m talking artful skin pictures. I’m looking forward to your video on Sunday bro.
mind boggling. I know AI generation comes from Billions of images but it is quite amazing that it gets the lighting, color, shadows right with newly generated images. The bedroom for example gets it like it had path tracing ray tracing technology but no. It could compete with architecture CAD for quick, inspirational designs.
Those that realistic have weird stuff when you look at the detail - again not usable. The best one is the hotel room you created, with the shortest prompt lol
Midjourney has come a long way for sure. However, I'm still surprised by what it can't draw like scissors for example. They appear to be from another planet at times. Also with Stable Diffusion's many options and now Adobe Firefly, will MJ become the "Corel Draw" of ai image generating software?
Looks like it's a great for those generating human subjects. A huge step back with other subject types though. Played with trying to create a few dragons and the results are hilariously bad! V4 was much better. Hopefully they will update to fix this soon.
hm.... i just created a dragon and it looks very nice to me. But i'm not a dragon expert cdn.discordapp.com/attachments/1004066178861383873/1086368586827833364/Olivio_fantasy_dragon_digital_drawing_highly_detailed_2_30dd7927-6411-4936-ace0-fccf74dcf7a8.png ps.: you can't post links for spam prevention
It is easy to get MJ to make cool images but hard to get it to make very specific complex images. It still screws up hands about 50% of the time. You can get some amazing images but if you are trying to make something specific that is complex it will struggle. An example is making a cartoon truck monster with a face breathing fire. I tried so many different prompts and nothing close. You get better images with a Google image search.
Looking at these I must say I am not entirely sold. But do know MJ is promoting their service with cherry picked results everywhere else so thanks for this review Olivio. In the end average user will still only get average results with some improvements.
Those results aren't cherry picked, but you need to learn a little bit about prompts to get good results. However the MJ community stream shows you the full prompts of all the best works presented there. So you can get these results with ease yourself
@Olivio Sarikas I didn't mean the one you shared were cherry picked.but some on other social networks only perfect images. I agree the community helps alot to get better images. 👍
@@YogaBlissDance This has been the same for ages. New technology displaces a large number of people in particular professions, but never really on the scale feared. People still enjoy classic computer games, despite all the bells and whistles of modern games, people still walk and bicycle despite of cars, people still wash their clothes by hand, despite washing machines etc. Though. This technology will require people to get creative, no pun intended, in a general sense - as this doesn't just threaten artists, ( AGI ), but also 90%+ of all professions. By the way. I'm an animator and 3d-designer by profession.
If I comissioned an artist to create these images there would be far more correspondence and text in emails to create these images. The art is completely amazing. I think we are only a large language model away from a natural descriptions of these seems vs a comma delimited keywords
Imressive! Your way of showing us around is highly instructive and exciting, thanks! Question: Do you know if all parameters already work in Version 5? (I´m not sure, but I think that --seed, for example, did not work out in my case, but maybe it was my fault...)
This has been so so helpful at helping me to develop a prompt / result collection I can call upon to create my own masterpieces. Thank you, thank you ! Please keep making more of these.
Olivio, I'm trying to get a consistent pattern image that I can use for a fabric design. Do you have any tips? I've tried "tile" with not much luck. I've ended up with some really nice singular cherry blossoms, but I want that image repeated a bunch of times at differing angles with equal shadowing. Tall order I know but any tips appreciated. Thanks!!
It will be interesting to compare it to Firefly (when I get into the beta)...from your video comparing them Firefly was more accurate to the prompt but Midjourney producted better images. I tried three MJ prompts...in two similar cases it confused what I had asked (thats probably on me.) In the third case I wanted an image of a particular place (a famous baseball stadium that was torn down 60+ years but many images of it exist) and none of what came back was right..the main entrance looked nothing like the real thing and in one case half the structure was cut off and the lights were pointing in the wrong direction?! I will be trying the same thing in Firefly once I'm in the beta.
Do notice that the Aztec teens are good, but the middle teen only has one leg. Also - how do you typ “pi” on a normal keyboard? Oh, and in the ocean painting, I think the word is supposed to be “cobalt”.
Robomar seems to like 6:4. I'm not sure if that does anything for the quality of the image outcome. The Ratio is the same. But MJ also doesn't exactly play by these rules and the images aren't exactly 3:2. I will ask him why he prefers that
true, but for SD you need the right models for each style and know how to prompt for that model. MJ can do all styles with the same prompting. That said, of course SD still offers a lot more control
really? i didn't feel like it's slower. Also keep in mind that the images you get are now the same resolution as the upscaled you whould have to create manually before
Yes, SD models are on par with MJ now. More compllicated though and requireing your own hardware. MJ is still the iPad of AIs. Easy to use, but limited in control
The nun looks like an unreal engine 3 render. Impressive all things considered (re: AI art), but definitely doesn't look realistic or even thgat impressive compared to Stable Diffusion. Gold skull looks cool and the teeth are great, I've had many issues with teeth. Hands are nuts when they work. Icons are cool. Overall, not worth paying over using Stable Diffusion for free. Plus the amount of finetuning you can do using all the models, extensions, etc. in SD is unmatched.
Agreed, SD is still unmatched, the amount of extensions/scripts, LOTS of different models (+custom merging), great img2img and inpainting, control and freedom is just way too powerful. Also, SD has a custom Ultimate Upscale script, which generates even more details on an upscaled image (requires some experimenting though). Sure, MJ can create some realistic images, while in Stable Diffusion you have to put 100+ tokens of negative prompt for similar results (depending on the model). I don't know, maybe MJ is already using lots of pre-set negative prompts under the hood too with each request. Stable Diffusion is not that far behind though, it just needs some more time and a very long negative prompt, but it's still just a luck of the draw and there is no golden rule to photo-realism. Every seed is different for each sampler and parameters. An example without ControlNet, with DPM++ 2S a Karras 150 steps (the only way for me to reduce hand artifacts) - still, I needed around 10 tries to get a somewhat okay looking hand: i.imgur.com/4x2dSz8.png In the end, it still doesn't matter, I use SD for a lot of game asset work, which you can't do with midjourney, but that's a whole different story.
@@MarvinCZ Yes, I get the prompt asked for a render, but to praise the result as new and realist and magical is just wrong. I can and have gotten much better results with SD. It's really not that impressive overall. Especially when you consider all the extensions in A1111. Just being realistic,
@@snoweh1 You were criticizing a render for not looking realistic. Those were your words and it is absurd. It can also do "photorealistic" images if they are prompted. That just wasn't the case there. I don't think anyone is saying it's "magical". It is however quite impressive. It's definitely not as adjustable as SD, and again nobody claims it is. That's also not its goal. Its goal is to be accessible, which it does better than SD. Also keep in mind that this is the first alpha release of the new model.
I stand corrected. At the moment Upscale does NOT cost GPU time, because it does not upscale. But this is just for the moment until they implement the new upscaller that will actually give you a bigger image.
I was really hoping for a stronger form of image editing, perhaps something like in-painting to be available in Midjourney. The new improvements are great, but it's so frustrating to have an image that's really close to what you want. And then you have to go over to DALLE 2 to do further editing.
export the photo from iphoto. upload the photo into discord to the MJ chat bot. right click on the uploaded photo and select "copy link". write /imagine and paste the link into the prompt flied. write the rest of your prompt and hit enter
Don't know if its just my stuff being wonky but the envelope thing doesn't seem to work to get the seed on v5 and none of my v5 stuff show up yet on the site.
upload img, upscale the most similar look (to subject) pic, then rate it with emoji " 😍 " then click re-roll the prompt. and do it again, in 3 or 4 re-rolls you would see that the look is closer and closer near the subject in the img input.. enjoy ^^
I’ve been doing that with SD. It’s not perfect so I have to use photoshop to solve the imperfections. So it’s not perfect, but if you’re a professional artist and know photoshop, you can do it. I’ve made professional portrait photos for my LinkedIn, and professional photos for Instagram with a dog and amazing locations haha.
If you have a premium subscription click on the name of the Midjourney bot and send it a message. Now you are in a private Chat with the bot. Everything works the same, but now it's just your images
@@rudimcloughlin3627 lol, i know experts from all industries and i am a speaker about ai on art events. So yes, artists are very interested in adopting AI into their work process. are you a artist?
What would be more useful For you to show is images created that are wow realistically, that people won’t notice they are fake. These look all fake - like gaming style..
@@OlivioSarikas It is all progress though and I look forward to trying out the latest MJ. Hands have been becoming better with SD for a while especially with the use of negative prompts.
@@blacklightiteh5052 It sounded like you needed it explained when you said that v5 had a photorealistic look, since that is not true. It is of course okay to not like it as much.
That's like saying artists shouldn't reference other artists in their work. When cleare some of the most iconic artworks in history are direct refereces to other artsits works and even build on their composition and idea directly.
because i made more followers with AI in 6 months, than i made with Photo editing in 4 years. Last month i made mode followers than in a complete year of creating content about Affinity Photo
@@staceykimbell9324 it isn't a trend, AI is here to stay and it will continue to grow, photos is the beginning, hopefully we can have neural driven virtual actors soon! 😉
After having played around with, I am not sure, that V5 is actually that great. You can clearly see, that in terms of photorealism, they made a big step forward. And ya, it's of course awesome, that V5 finally gets hands right most of the time. Unfortunately, that's about it really. As soon, as you want to force a certain artistic style, things are going downhill... fast. V5 is less able to 'anticipate' the expected result... which for me is really 'less intelligent' than V4. People are always saying that it is a matter of prompting, but that is utter nonsense. The problem is, that in oder to get the result that you want, you need to become much more extensive in your description and prompting. And that's where the trouble really starts... for me it almost seemed, that the AI gave up at some point and just painted something. I am really disappointed I have to say. Especially when it comes to consistency. Producing several pictures in the same art style (which meet your expectation) other than photorealism is very difficult and a lot less fun than in V4.
#### Win a RTX 3080Ti ####
Sign up FREE to Nivida GTC for a Chance to Win a RTX 3080Ti:
nvda.ws/3D7nNY4
#### Follow Robomar ####
Follow Robomar on Instagram: instagram.com/robomar.ai.art/
#### Join me ####
On Facebook: facebook.com/groups/theairevolution
On Discord: discord.gg/XKAk7GUzAW
I wish they actually gave away 3090 because I already have got 3080ti.
The only reason why I would upgrade is the need for more VRAM.
I bought my card before AI image and language models were really a thing so I didn't foresee the situation where 3080 TI's 12 gigs wouldn't be enough for me. At least the AI stuff for us doing on my computer a year ago is way different from what I'm doing these days. And for me 12G definitely ain't enough at the moment. In fact I will probably even save money to get myself a 80 GB A100.
For many years I've already valued VRAM size more than te number of cuda cores. For a while I even used Nvidia Tesla k40 with custom 3D-printed cooling attached to it as Tesla cars are usually used in servers in data centers.
@@harlycorner I have a 3080 ti right now and it's really nice for AI Image render :)
👋
Today they had 1.3 million concurrent users on their Discord, that's crazy. 😅
wow, that's amazing!
I was one of them 😂
@@bastianrivero Yeah, we were quite the crowd yesterday. 😉
And that might not count users prompting in the DMs or on their own server.
@@Dazreil True. Discord as a whole was at breaking point yesterday, wouldn't even load some images for me, so who knows, maybe there were many more of the 13+ million users playing with the new model.
Great video as always Olivio and a mention to Robomar for his excellent prompts as always.
Again, I've typed up the prompts for the community to save time, enjoy folks.
catholic nun from 1780, old Whitestone village background, ultra realistic, octane render --ar 6:4 --v 5
bald man with a beard, wearing a tuxedo --iw 1 --v 5
intricate details, skull made of gold --ar 4:5 --v 5
woman fashion model wearing 60s clothing, comic book, sharp focus --ar 4:5 --v 5
Lovely anime Cyberpunk City Painting, Diagonal Camera Angle, Dutch Angle, Volumetric Lighting, Digital painting, night, Marc Allante, Studio Ghibli, Impressive, Intricate detail, Alejandro Burdisio, Dan Luvisi --v 5 --v 5 --ar 1:3 --chaos 20 ar 1:2 --chaos 20
a set of 9 startup icons with a modern, flat icon design --ar 4:5 --v 5
yellow Mercedes Brabus standing dirty in the jungle, moisture and haze in the air, professional vibrant grading, soft shadows, no contrast, clean sharp focus, commercial photography --ar 6:4 --v 5
The hospital room of the future, in a futuristic spaceship, with calming bright lights and a welcoming wooden floor, a doctor and her patient look at a computer screen showing medical graphs --ar 6:4 --v 5
modern hotel room --ar 16:9 --v 5
imperial Palladin in the detailed filigree armour, fine illustration, ultra sharp --ar 4:5 --v 5
Sprite sheet with game visuals, game assets, various prompts, 3D exploring game, cinematic lighting, designed, pc game, beautiful, manga, unreal, 64k, enhanced, intricate pizel art --v 5
detailed blacksmith room interior, sharp focus, isometric diorama, octane render --v 5
expression character sheet --ar 4:5 --v 5
sea town, white background, cartoon editorial --ar 4:5 --v 5
Futuristic woman, sharp focus, white background, comic book --ar 4:5 --v 5
old car from 1970, color pencil drawing --ar 6:4 -- v 5
mountain rescuer backpack, symmetrical placement, knolling, sharp focus --ar 4:5 --v 5 --v 5
Full body shot of a cute little anthropomorphic fox wearing a adventurer outfit, style of Pixar, character design art, by Jean-Baptiste Monge, highly detailed, digital art, cinematic lighting, epic, on white background, ultra realistic, vray render, expressive pose, professional color grading, soft shadows, no contrast, clean sharp focus --ar 4:5 --v 5
crouching tiger on the hunt for his pray, national geographic photography -- ar 6:4 --v 5
modern street hustler, sharp focus, renaissance painting --ar 4:5 --v 5
impressionism dark cloudy sunset over the stormy ocean shore, huge waves, the ocean playing jazz in coalt colors, TT --ar 2:5 --v 5
dress made of mirrored scales, high fashion, elegant, beautiful, model --q 5 --ar 9:16 --v 5
ancient aztec teenagers, traditional clothing, waiting around --v 5
A smiling little girl in a red coat and hat walks in front of a giant panda with a bamboo forest in the background, delicate features, rich colors and details, art by shinkai, studio Ghibli, highly detailed, HDsmooth --v5 --q 2 --s 750
Thank you :)
How can people still in good conscience use artist names when prompting, especially Monge, who is vocal against this software
freakin legend! i am annoyed that he let his viewers do his job instead of just type it in description.
I love the new version, is crazy how fast everything is moving. Also this is only the alpha of version 5, hands will get even better I assume.
true, this isn't even the full release yet
Teeth is still messed up though. No distinct molars, incisors, and canines. Just a lot of varied incisors maybe a canine here and there.
@@Mimeniia I had a shocking result with teeth, so added something like, healthy white teeth, to it next time, which helped.
So viele V5 Videos die ich jetzt gesehen habe, aber keines war so smooth und informativ wie deines. Danke für deine Arbeit und die Mühen! Zumindest für mich bist du auf AI-Art Ebene ein echter Mentor. Das wollte ich einfach mal gesagt haben, weil ich nach lengerem zögerlichem rumprobieren in den letzten Monaten, vor einigen Wochen so richtig in die Materie eingestiegen bin und einen Großteil meines Wissens dir verdanke. Danke Mann, Liebe geht nach Wien! 😉🤙❤
Danke, das freut mich sehr das ich dir mit meinem Video helfen konnte und es genau das ist was du gesucht hast. Viel Spaß in der Welt von AI :) Ich hab noch viele andere Videos zum inspirieren. Schau dir mal mein Hyper-Realism in Midjourney video an :)
This is definitely the most exciting release we have seen in a while! It is hard to believe how far the technology has come in the past few months!
I'm starting to wonder if those that say we are living in a simulation are actually correct. Imagine this tech a thousand years from now....
true, i wonder the same
lol, how cute. Wait til the mass layoffs start and then the riots and mass shootings because people have no purpose or livelihoods and are left in the gutter by corporations and the Government. We're not going to last another 50 years.
They said with V5 using terms like "octane render" or "unreal engine" will now give you results that look more like video games rather than real life.
Yes for everything you'll get the results as you described in the prompt. As you say octane render will give you cgi look, photography will give you photo look.
I was waiting for you to talk about v5. I hope to get more tips from you soon!
yes, sorry. I wanted to do a quick video and then it took ages to record and edit this video. I also had a day filled with tasks. Really happy i could publish this
I simply can't give enough praises to you and others like you that understand the potential and power of AI. It's amazing how I see other artist actually believing that AI is just a phase, and it will pass over soon. AI is here to stay and it's evolving at a pace that few can even keep up with. I can say with comfort that you'll be sitting in the captain's seat when AI is in full control.
Thanks so much for this tutorial, Olivio! I've been trying to design cartoon backgrounds for my animations. I'm trying to figure out codes/ways to prompt Midjourney for certain perspectives of a room (for instance, Midjourney often shows me overhead views when I'm looking for straight on, eye level view). I'm also wondering if there are good prompts for Midjourney to share multiple walls of the same building space so they all work together congruently in the animation. So far I've been struggling to find the right words in midjourney to make this happen. Thanks for your help!
Amazing informative video. Thank you for the mention.
Thank you for your help
Loved this video! Thank you for showing us what V5 can do. You mentioned that game assets are something you enjoy making in Midjourney. Could you maybe make a video dedicated to showing the types of prompts that have worked well for you in the past?
I second this comment! 😊
I will do a video soon about creating a Video Game with the help of AI :)
@@OlivioSarikas That's great! I've tried making icons/buildings/units and stuff to be used as board game assets, but I think my prompts need refining. Looking forward to it!
I simply can't get any interest in midjourney as long as it can't be run locally. I want to have full control of my generation.
The main source of any ai image generator are the images already made all across the internet. MJ is not rendering something from scratch, but rather from compilling pre made photo elements and combining them.
@@felipemagalhaes567 People still believe this lie in 2023? That is not how AI works my friend.
That's like saying you can't make anything new, because all you know is learned from the people who lived before you. When in fact we have seen many many new works generated by AI in a way a human would not do that. Sure, that was by chance, but still that counts
@@felipemagalhaes567 I don't get your point. Stable Diffusion can be run locally...
Would love to have local version of MJ
Mid-journey is a new medium but they are not photos. I call them digital illustrations. I also recently used Blue Willow V2 and I got more better-quality images with more aesthetic outputs
Mr Blue Willow? Is that you? :)
I’m hoping that they will have in a future update, two consistent characters in one image. That would be awesome especially when doing a comic book.
If you draw your comic book characters and put those in your prompts you could get consistency.
@@chrishenricks6405 lol if I knew how to draw, I'd just do that instead of typing out what I want.
V5 made major strides in dealing with hands. The problem I'm having is how realistic everything is. Even specifying digital painting doesn't work out that well. I was mostly using Niji since it's more in the line of the style I'm going for. Hopefully their V5 version is just as good.
Yes, i agree. V4 and V5 are less playful in their style than V3. This can be tackled with prompting to a certain extend though
Helps a lot to specifiy certain famous artists in the style you're looking for. "In the style of so and so" seems to be working much better in V5.
Thanks for amazing content, I noticed something, I have generated the prompt for MJ 5 in ChatGPT4 when you change the order of the description in the prompt, the output change completely, so the focus will be on the beginning of the prompt
Example 01
Sharp focus, isometric diorama, octane render - Introduce the main character, John, as he walks through a bustling city street filled with people and vehicles. He is wearing a blue jacket and carrying a briefcase. The scene is viewed from an isometric perspective, showing the tall buildings and cityscape in intricate detail. The lighting is natural and balanced, highlighting the rich textures and vivid colors of the environment. John looks determined and focused as he navigates the busy street. --ar 16:9
Example 02
The main character, John, as he walks through a bustling city street filled with people and vehicles. He is wearing a blue jacket and carrying a briefcase. The scene is viewed from an isometric perspective, showing the tall buildings and cityscape in intricate detail. The lighting is natural and balanced, highlighting the rich textures and vivid colors of the environment. John looks determined and focused as he navigates the busy street. Sharp focus, isometric diorama, octane render --ar 16:9 --v 5
Please advise if you have any tips regarding this
I got amazing results for the following prompt:
In the mind's abyss, where thoughts reside,
Images take shape, and truth they hide.
You believe you think, and so you see,
A world of shadows, where light should be.
Colors, hands, details, very good now. Thank you for the review. You are a champion.
my pleasure. thank you for watching
V4 was meant to replace artists. V5 was meant to replace photographers. I don’t know if the realism makes things better or worse. The mutilated fingers, extra limbs, or missing limbs is more unsettling to me now that the images look more realistic. It’s like you’re looking at a real mutilated person. No thanks.
If anyone needs me I’ll be in the lab with stable diffusion. I’ll check black when V6 is out
@23:25 Guy on left has wrong left foot (It's a right foot), one in middle only has one leg...hands are a bit off too, esp left and middle...
Very beautiful and realistic ❤
Thank you so much 😀
I’m very indifferent with V5. Sometimes it’s great & other times, not so much. Feet are still a foreign concept but the devs consider feet NSFW, so stupid. The major issue is the upscaling is disabled, luckily they said it’s temporary. Lastly, they just soft banned Chest with the latest update. I hate HATE fake artists that are offended by the female or male body. I’m not talking about XXX I’m talking artful skin pictures.
I’m looking forward to your video on Sunday bro.
yes, the cesorship is way too much. i wanted to create a photo of the famous bathing monkeys in japan, but bathing is not allowed as a word
@@OlivioSarikas Really...I wish they'd tell us the banned words, I happen upon them.
The three latin kids definitely have feet problems. The one in the middle has only one leg, the left has two right feet.
mind boggling. I know AI generation comes from Billions of images but it is quite amazing that it gets the lighting, color, shadows right with newly generated images. The bedroom for example gets it like it had path tracing ray tracing technology but no. It could compete with architecture CAD for quick, inspirational designs.
Those that realistic have weird stuff when you look at the detail - again not usable. The best one is the hotel room you created, with the shortest prompt lol
Midjourney has come a long way for sure. However, I'm still surprised by what it can't draw like scissors for example. They appear to be from another planet at times. Also with Stable Diffusion's many options and now Adobe Firefly, will MJ become the "Corel Draw" of ai image generating software?
Looks like it's a great for those generating human subjects. A huge step back with other subject types though. Played with trying to create a few dragons and the results are hilariously bad! V4 was much better. Hopefully they will update to fix this soon.
hm.... i just created a dragon and it looks very nice to me. But i'm not a dragon expert cdn.discordapp.com/attachments/1004066178861383873/1086368586827833364/Olivio_fantasy_dragon_digital_drawing_highly_detailed_2_30dd7927-6411-4936-ace0-fccf74dcf7a8.png
ps.: you can't post links for spam prevention
awesome ! thanks for sharing and your commitment ^^
you are very welcome, my friend
It is easy to get MJ to make cool images but hard to get it to make very specific complex images. It still screws up hands about 50% of the time. You can get some amazing images but if you are trying to make something specific that is complex it will struggle. An example is making a cartoon truck monster with a face breathing fire. I tried so many different prompts and nothing close. You get better images with a Google image search.
Looking at these I must say I am not entirely sold. But do know MJ is promoting their service with cherry picked results everywhere else so thanks for this review Olivio. In the end average user will still only get average results with some improvements.
Those results aren't cherry picked, but you need to learn a little bit about prompts to get good results. However the MJ community stream shows you the full prompts of all the best works presented there. So you can get these results with ease yourself
@Olivio Sarikas I didn't mean the one you shared were cherry picked.but some on other social networks only perfect images. I agree the community helps alot to get better images. 👍
Cherrypicketfence! Picked! Cherryontopsyturvy! OhNooooo!
Yep so hopefully some folks will still have work as artists to help them??
@@YogaBlissDance This has been the same for ages. New technology displaces a large number of people in particular professions, but never really on the scale feared. People still enjoy classic computer games, despite all the bells and whistles of modern games, people still walk and bicycle despite of cars, people still wash their clothes by hand, despite washing machines etc. Though. This technology will require people to get creative, no pun intended, in a general sense - as this doesn't just threaten artists, ( AGI ), but also 90%+ of all professions. By the way. I'm an animator and 3d-designer by profession.
If I comissioned an artist to create these images there would be far more correspondence and text in emails to create these images. The art is completely amazing. I think we are only a large language model away from a natural descriptions of these seems vs a comma delimited keywords
Great video and thank you for the insight.
My pleasure!
Awesome,thankyou.
Imressive! Your way of showing us around is highly instructive and exciting, thanks! Question: Do you know if all parameters already work in Version 5? (I´m not sure, but I think that --seed, for example, did not work out in my case, but maybe it was my fault...)
This has been so so helpful at helping me to develop a prompt / result collection I can call upon to create my own masterpieces. Thank you, thank you ! Please keep making more of these.
Thank you very much. I'm happy i could help and inspire you :)
Olivio, I'm trying to get a consistent pattern image that I can use for a fabric design. Do you have any tips? I've tried "tile" with not much luck. I've ended up with some really nice singular cherry blossoms, but I want that image repeated a bunch of times at differing angles with equal shadowing. Tall order I know but any tips appreciated. Thanks!!
It will be interesting to compare it to Firefly (when I get into the beta)...from your video comparing them Firefly was more accurate to the prompt but Midjourney producted better images. I tried three MJ prompts...in two similar cases it confused what I had asked (thats probably on me.) In the third case I wanted an image of a particular place (a famous baseball stadium that was torn down 60+ years but many images of it exist) and none of what came back was right..the main entrance looked nothing like the real thing and in one case half the structure was cut off and the lights were pointing in the wrong direction?! I will be trying the same thing in Firefly once I'm in the beta.
Do notice that the Aztec teens are good, but the middle teen only has one leg. Also - how do you typ “pi” on a normal keyboard? Oh, and in the ocean painting, I think the word is supposed to be “cobalt”.
Enjoying your videos, didn’t MJ5 state that prompts should be written in normal sentences rather then a bunch of separated keywords
I tried both and I kind of get better resutls with the commas. I think the real language model is still comming. this is V5 Alpha
I cannot believe the skull image. How is that not real????😱
I think you can select v5 form /settings so you don't need to type that in the prompt.
i would love to see the day when it can be able to do consistent characters that you can reuse for each individual picture like for a picture story.
yes, that would be great. and also explore consitent locations.
You can do that by using seed
Can't wait until I get home and can try some of these things.
you will have sooooooo much fun. no time to sleep tonight
Jesus christ that's mind blowing!
totally is!
Here’s to hoping that Stable Diffusion will bring the fire, ‘cause screw paying for gimped ass censored AI.
Gooood... thanks always!
thank you so much
Can you please explain the difference between aspect ratios 3:2 and 6:4?
none
What The Algomist says, they're exactly the same.
Robomar seems to like 6:4. I'm not sure if that does anything for the quality of the image outcome. The Ratio is the same. But MJ also doesn't exactly play by these rules and the images aren't exactly 3:2. I will ask him why he prefers that
the image processing of sd with its addons still beats this. but its not too bad.
true, but for SD you need the right models for each style and know how to prompt for that model. MJ can do all styles with the same prompting. That said, of course SD still offers a lot more control
Its indistinguishable from real life , is this the singularity ?
Olivio they have NOT fixed the problem with hands completely yet.
that's what i said in the video
V5 takes a lot longer, going to be extra frustrating when an image doesn't work out, but should be more rewarding when it does.
really? i didn't feel like it's slower. Also keep in mind that the images you get are now the same resolution as the upscaled you whould have to create manually before
I found it to be significantly faster.
ooh that icom example makes me think they should make it able to output svg files.
Love this
thank you
Hi Olivio, Is the v5 only for purchased plan user? If I purchased a plan, can I use the v5 immediately? THanks.
That was a tuxedo - the prompt should have been 'wearing a black tuxedo with a white tuxedo shirt and black bow tie'
Just starting dabbling in AI a week or so ago. Having a ball and loving v. 5
welcome to the AI world
mid v5 indeed getting better, but I think the IlluminatiDiffusion model have the similarity output model
Yes, SD models are on par with MJ now. More compllicated though and requireing your own hardware. MJ is still the iPad of AIs. Easy to use, but limited in control
@@OlivioSarikas Can any SD model do hands and teeth as well as MJ 5, though?
Its really amazing!
yes, totally love it
Hi Olivio, can you tell me what the highest resolution I can get from V5 is compared to V3 or 4? thanks
Thanks!
Thank you very much, that means a lot :)
It looks like 60% of amazing images are generated pure based on coincidence? Because the prompts are different from the results
08:42 Oops. Your bias is showing. According to the prompt later in this video, that is not a nurse, she's a doctor. 😅😂
The nun looks like an unreal engine 3 render. Impressive all things considered (re: AI art), but definitely doesn't look realistic or even thgat impressive compared to Stable Diffusion. Gold skull looks cool and the teeth are great, I've had many issues with teeth. Hands are nuts when they work. Icons are cool.
Overall, not worth paying over using Stable Diffusion for free. Plus the amount of finetuning you can do using all the models, extensions, etc. in SD is unmatched.
If you look at the prompt, that i show in the video. you can see that is says octane render.
Agreed, SD is still unmatched, the amount of extensions/scripts, LOTS of different models (+custom merging), great img2img and inpainting, control and freedom is just way too powerful. Also, SD has a custom Ultimate Upscale script, which generates even more details on an upscaled image (requires some experimenting though).
Sure, MJ can create some realistic images, while in Stable Diffusion you have to put 100+ tokens of negative prompt for similar results (depending on the model). I don't know, maybe MJ is already using lots of pre-set negative prompts under the hood too with each request.
Stable Diffusion is not that far behind though, it just needs some more time and a very long negative prompt, but it's still just a luck of the draw and there is no golden rule to photo-realism. Every seed is different for each sampler and parameters.
An example without ControlNet, with DPM++ 2S a Karras 150 steps (the only way for me to reduce hand artifacts) - still, I needed around 10 tries to get a somewhat okay looking hand: i.imgur.com/4x2dSz8.png
In the end, it still doesn't matter, I use SD for a lot of game asset work, which you can't do with midjourney, but that's a whole different story.
The prompt asked for a render and the result looked like a render. The horror.
@@MarvinCZ Yes, I get the prompt asked for a render, but to praise the result as new and realist and magical is just wrong. I can and have gotten much better results with SD. It's really not that impressive overall. Especially when you consider all the extensions in A1111. Just being realistic,
@@snoweh1 You were criticizing a render for not looking realistic. Those were your words and it is absurd. It can also do "photorealistic" images if they are prompted. That just wasn't the case there.
I don't think anyone is saying it's "magical". It is however quite impressive. It's definitely not as adjustable as SD, and again nobody claims it is. That's also not its goal. Its goal is to be accessible, which it does better than SD. Also keep in mind that this is the first alpha release of the new model.
I was going to quit mid journey and only use stable diffusion cause of hands. So many wonderful art destroy due to lack of hands.
Hands are waaayyyyy better in V5. I’ve noticed that most hands are actually perfect or close, at least in my experience playing around today.
yes, they are really amazing now. even in complex situations. really impressive
Do I need to install V5 for Midjourney, or will it update itself to the newest version once I use it on Discord ?
it's updated for you, but you need to use --v 5 at the end of the prompts or write /settings in the chat and change it to V5 for all your renders
Do you lose credit when you upsclae? I realised in v5 the upscale is the same as the original grid image
I stand corrected. At the moment Upscale does NOT cost GPU time, because it does not upscale. But this is just for the moment until they implement the new upscaller that will actually give you a bigger image.
I was really hoping for a stronger form of image editing, perhaps something like in-painting to be available in Midjourney. The new improvements are great, but it's so frustrating to have an image that's really close to what you want. And then you have to go over to DALLE 2 to do further editing.
I would love that too, but i think that's a long time away
I still can’t figure out how to get a url link to one of my own photos in iPhoto, to use in MJ ?
export the photo from iphoto. upload the photo into discord to the MJ chat bot. right click on the uploaded photo and select "copy link". write /imagine and paste the link into the prompt flied. write the rest of your prompt and hit enter
Don't know if its just my stuff being wonky but the envelope thing doesn't seem to work to get the seed on v5 and none of my v5 stuff show up yet on the site.
With V5 (and V4) the seed should always be right under the finished results, next to the job ID
there's a way to make ur input imgs looks closer to the subject in photo :D
upload img, upscale the most similar look (to subject) pic, then rate it with emoji " 😍 " then click re-roll the prompt. and do it again, in 3 or 4 re-rolls you would see that the look is closer and closer near the subject in the img input.. enjoy ^^
ohhhhhhh! I didn't know that. Thank you
@@OlivioSarikas you're welcome. let me know here if you found it useful
Yes. But how can I play with it today before some next big thing comes around like e.g. tomorrow?
You have to be very very quick 😂
@@OlivioSarikas Erwischt, ich hab doch geahnt die Sache hat ein Haken... 😅
Did you find any other AI that is better at getting your face accurately? Maybe training SD with your own photos?
I’ve been doing that with SD. It’s not perfect so I have to use photoshop to solve the imperfections. So it’s not perfect, but if you’re a professional artist and know photoshop, you can do it. I’ve made professional portrait photos for my LinkedIn, and professional photos for Instagram with a dog and amazing locations haha.
Is there a way of stop loading other users images on discord? It's really a PITA finding my own prompts when tons of images keep loading non stop.
If you have a premium subscription click on the name of the Midjourney bot and send it a message. Now you are in a private Chat with the bot. Everything works the same, but now it's just your images
Looks like ai will replace artists. I’d love to see how this is going to continue to be a tool for people.
Rather enhance artists in what they are doing :)
@@OlivioSarikas how will it enhance their process?
@@OlivioSarikas Yeah my guy it’s taking artists work. They still ain’t happy about it.
@@rudimcloughlin3627 lol, i know experts from all industries and i am a speaker about ai on art events. So yes, artists are very interested in adopting AI into their work process. are you a artist?
When did v5 come out?
the same day i made my video
jaw dropping AI
i'm in love with this
Imagine V6 when it comes out lol.
That will be insane!
What would be more useful
For you to show is images created that are wow realistically, that people won’t notice they are fake. These look all fake - like gaming style..
It’s just sad that the prompts are grammatically incorrect because the user is an analphabet but the AI still knows what to do. Humanity’s doomed.
The hand at four minutes isn't that great as an example. Still dodgy 😊
True, i should have picked another example
@@OlivioSarikas It is all progress though and I look forward to trying out the latest MJ. Hands have been becoming better with SD for a while especially with the use of negative prompts.
still renders strange fingers, and got a lot of watermarked stuff, and even the chairs look weird, this is still a hobby tool...luckily
yes, fingers aren't 100% yet, but it's a lot lot better than bevor
✋️ 🤚 ✋️ 👐 🤝🏻
👍🙏💪✋👏👋👊✊✌👌🤘🖐😂
Midjourney servers need more gpus.
keep in mind that a lot of people are testing V5 right now. this will be better in a couple of days
just when my subscription expired ;(
There is always the re-sub button ;)
I find a lot of hype that has a lot of time wasting and not really usable
Monetizing a 25 minute video of you just reading prompts of other people’s MJ work.
Jesus.
To be fair it isn’t even the person’s work.
The V5 is not good for futurism, V4 was better.
What do you mean by futurism here? I know what it is in general but I don't know what you mean in relation to AI-generated imagery.
@@MarvinCZ simply not so photorealistic style as v5 has.
@@blacklightiteh5052 The v5 can do a not-photorealistic style too, you just have to tell it quite specifically what style you want.
@@MarvinCZ you don't have to explain it to me, I work with it daily. I expressed my opinion and I don't like it as much as v4. That's all.
@@blacklightiteh5052 It sounded like you needed it explained when you said that v5 had a photorealistic look, since that is not true. It is of course okay to not like it as much.
People need to stop putting other artist's names in their prompts. Try creating something original.
That's like saying artists shouldn't reference other artists in their work. When cleare some of the most iconic artworks in history are direct refereces to other artsits works and even build on their composition and idea directly.
I am confused why Olivio would spend years building up his channel's following only to throw it away with a ton of AI videos. it doesnt make any sense
because i made more followers with AI in 6 months, than i made with Photo editing in 4 years. Last month i made mode followers than in a complete year of creating content about Affinity Photo
He should upload what he like, and a lot of people like what he is uploading so why not?
You adapt to the times, that's why.
@@Danuxsy I think you meant to say "follow the trend".
@@staceykimbell9324 it isn't a trend, AI is here to stay and it will continue to grow, photos is the beginning, hopefully we can have neural driven virtual actors soon! 😉
After having played around with, I am not sure, that V5 is actually that great. You can clearly see, that in terms of photorealism, they made a big step forward. And ya, it's of course awesome, that V5 finally gets hands right most of the time. Unfortunately, that's about it really. As soon, as you want to force a certain artistic style, things are going downhill... fast. V5 is less able to 'anticipate' the expected result... which for me is really 'less intelligent' than V4. People are always saying that it is a matter of prompting, but that is utter nonsense. The problem is, that in oder to get the result that you want, you need to become much more extensive in your description and prompting. And that's where the trouble really starts... for me it almost seemed, that the AI gave up at some point and just painted something. I am really disappointed I have to say. Especially when it comes to consistency. Producing several pictures in the same art style (which meet your expectation) other than photorealism is very difficult and a lot less fun than in V4.