12:12 The ‘consistent characters’ is the most interesting part of this video. It deserves a video of its own. I hope we will see truly ‘consistent’ characters, not just ‘similar’ characters that we’re seeing in Midjourney, where the character looks similar, but Midjourney adds different characteristics to that character in every image, such as changing their clothes to something else. A truly consistent character will look consistent in every scene, and their clothes won’t change either, unless we prompt for their clothes to change. Also, I’d love to see an AI tool that can not only do consistent characters, but also pose multiple consistent characters together in a scene. For storytelling, we often need to create images of scenes where two or more of our consistent characters are interacting with each other.
there are certain tools/extensions that are getting dangerously close to this 'consistent character + compositing characters'. Once it is done, i think industry pros will have no choice but to make use of it.
@@MattVidPro Thanks Matt. I would love to see you dive deeper into this. These types of features would be very useful to anyone who wants to use AI for storytelling.
You can do consistent characters in Comfy UI with an IP adapter. Nerdy rodent has a couple of videos on it. (Probably any SD but I have no idea how it works in A1111)
ITs strange to think by 2030 we could have AI generated games and worlds more detailed and beleivable than anything a triple A game developer can make with a team of a few thousand people.
We will have more ways to waste time than ever. I can also see steam popularity fading since many of these projects will be open source and indie games of today being really easy to make with AI in a few years. The holodeck thing I am sure was about VR
I've been saying for some time that these image generators need to build a full 3D model of a recurring character for easier generation of those characters.
Yes! And those models need different, adjustable layers for accessories, clothing and of course: hairstyles. It's like merging Midjourney with Blender or Unreal Engine etc.
@@jarblewarble already happening. There's a Text to Video AI that does entire short SCENES with persistence in the background and characters from a script. We'll Text to Video for entire sequels to past movies and TV shows in the future, and pass them around like Tiktok vids to our friends.
Since you asked, it's not even the obvious weird ish that gives them away. Hair/fur, skin, fabrics, textures, reflections, coloration, have subtle plastic/ painterly/ unified qualities in all these pictures that aren't present in real life. Object placement and alignment would be the next things. Then obviously when things melt together lol
The zero stylization and raw mode really surprised me. I've been producing extremely photorealistic shots by setting stylize to 1000 and making my prompts short and non descriptive using Midjourney V6
One of the best realistic checkpoints of Stable Diffusion is Epic Realism.. It produces images of normal people (not the typical profesional models), with a lot of details and imperfections on skin, with spots, wrinkles, etc... Specially if combined with upscalers like 4x_NMKD-Superscale-SP_178000_G
Very cool. I was one dalle and chatgpt day one. And was using Google colabs before that.from watching a few of you longer videos, I can tell you understand really understand the image generation community. Thanks for the videos!
I think by the end of this year or early 2025, we will all be in shock at the generative tools we will have available. Then we will all be lost in the Elmo sauce. 😂 For some reason, I think I might start using that phrase. 🤔
🎯 Key Takeaways for quick navigation: 00:00 🖼️ *AI Image Generation Trends* - Overview of AI image generation trends in 2024. - Highlights DALL-E 3, Mid Journey V6, and Stable Diffusion. - Discusses the realism of images generated by Mid Journey V6. 05:03 📸 *Mid Journey V6 Realistic Image Prompt* - Explores the Mid Journey V6 prompt for creating realistic-looking images. - Demonstrates the settings and results of the prompt. - Emphasizes the community's desire for creative expression in AI-generated images. 10:12 ✒️ *AnyText: Editing AI-Generated Text in Images* - Introduces Alibaba's AnyText, enabling text editing on AI-generated images. - Showcases the ability to modify text in different languages. - Discusses the impact of AnyText on AI art generation workflows. 11:50 🌐 *Future of AI Art: Visual Text Generation* - Explores the potential of visual text generation in AI art. - Highlights the significance of combining AI-generated imagery with editable text. - Considers the implications of this technology for the future of AI art. 13:00 🤖 *Consistent Characters in AI Art Generators* - Discusses the emergence of consistent characters in AI art generators. - Showcases an example of a consistent character created with Runway. - Explores the potential impact of consistent characters on AI image generation. 15:16 🚀 *Mid Journey's Bold Claims for 2024* - Addresses Mid Journey CEO's statements about achieving the "hollow deck" by 2024. - Discusses the potential development of a real-time open world by the end of 2024. - Expresses skepticism and curiosity about Mid Journey's ambitious plans.
Great overview, thanks! I hope we can at some point create consistent virtual characters that we can pose, animate and direct (playing out a script or convincing dialogue.) All the seeds to make this possible are already out there, but the scattered workflows have to be brought together into one platform with easy UI.
Dude, if we get holodeck level of immersion I want to actually walk around on a holodeck Enterprise and then talk to the AI Star Trek Next Generation cast and just chill out eating some Gagh with Picard. Awesome video.
Definitely so excited for the field of AI art moving to 3D. It indeed is the only way to go for creating believable videos. Though, we still need more flexibility in adjusting AI outputs, including purely manual adjustment for last touches. Still a few years away, but it's heading in the right direction.
Character consistency is such a high-demand item that SOMEONE is going to figure it out, eventually--probably all the AI image generators will have it, someday.
Professional artists, take a moment to absorb how incredible all of this is, but please remember to protest the use of these tools on the job (even if incrementally, because who tf can stop it). Stand with your union or unionize. If the convenience of AI Art can reduce a team of 100 to a team of 10, the chances of you being one of the 90 are pretty high.
I find the detail amazing, that in the „first date“ picture, both look slightly off camera, as if they watch themselves on the phone screen and do not look into the camera - just as we doofuses all do. 😅
Consistent character creation wasnt really working before! If its been solved then content will definitely moving from static to dynamic which means authors will write worlds and characters in place of just stories! Cant wait!
I can't do the Sesame Street ones in GPT-4 (Dalle3)--it pings me for copyright. How did you not get hit? I switched out the copyright info for something similar, and it gave me an image, but it wasn't amazing, just OK. Then I went to the Dalle standalone site, and used the copyright prompt, and it worked great--much better than within ChatGpt...is there a reason? Aren't they using the same model?
Is it the first time you wear that glasses? They look great! Not ai generated at all. Great vid, as per usual, especially your enthusiasm that is contageous
I think midjourney ceo meant it more along the lines of prompt generated nerf style worlds that can be navigated but with a vr headset we’re again realistically 1 step closer to holodeck
A similar effect can be achieved on Dalle-3 using the prompt ‘photo taken in 1996’ or any other year. Really outstanding results can be achieved if you tell it the details of what you want to see.
@mattvideopro, if I want to mix non-cell filming with A.i. such as midjourney, what do you suggest is the kind of camera that can get the job done without costing a fortune?
if you upscale the image and downscale for sharing you can fake perfectly because you are creating a lot of detail in the background 4:06 and then pixelating it when downscaling afterwards thus faking the image perfectly
The fun irony that AI is going to produce more "realistic phone pictures" when nearly every company is rushing to add AI upscaling to their phone cameras.
That second DALL-E 3 image of Elmo looks like it could have been shot on an iPhone. The latest ones are capable of automatically applying portrait mode to people and pets when you take a photo. Not sure Android flagships can do something similar (probably can), but if not they'll be able to real soon. So those kinds of "slightly blurry" photos of subjects are gonna get more believable as they get uploaded to social media more and more.
As incredible as those new Midjourney images look, I'm not too interested in the progress of Midjourney/DALL-E simply because they're both online-only models with limitations and restrictions. But I'm VERY excited for the possibility of good offline 3D model generation coming this year. Stable Zero123 is a good start, but with some refining and improvements by the community (mainly to reduce the VRAM needed), it could truly become something amazing.
i cant wait for character consistency frustration to be a thing of the past. i have been spending way too much time with my workarounds/workflow for getting around current ai's shortcomings in this regard and it just gets exhausting ... and makes art less fun
I draw art for a living. I draw to create visial depictions of my imagination. I enjoy the result, not the process. AI is getting closer and closer to being able to take seconds to produce what it takes me 10 hours to produce. I LOVE AI!
Exactly. I don't know why anyone enjoys the process of figuring out proportions/tediously drawing out microscopic details to make art. Maybe some people find it relaxing, but I find it more enjoyable to see an image I wanted without wasting an absurd amount of time and effort on it.
And this is why you'll get ahead. Learn these tools and go above and beyond with them. We are living incredible times in terms of creativity. In fact, this is the stuff of dreams. The harsh reality is that people who sell single static illustrations will not survive. But on the other hand, you can create entire universes. Imagine hand drawing a particular character and being able to replicate it in multiple poses and insert it in different scenarios in minutes, instead of redrawing the same char over and over? You can create comics quickly. And create entire set of concept art rather quickly. You can have your studio for doing stuff you love with 3 people instead of 30, because output willl increase so much. You can still introduce manual work in the workflow, in fact I think that's how AI is used for the best. People who reject AI will be left behind for good.
It won't because more of the applications as people upgrade hardware will be done locally. I have a decent computer setup and can do most of what done in ChatGPT locally with a combination of opensource models. However these next 3 years will be big for them
Mat, I appreciate your optimism and excitement about the progress in AI development. It's what I enjoy most about your content. I used to feel the same way in the beginning but not anymore. But it has come to light that many of these technologies are actually ripping off content creators. In light of that, how do you keep the same energy? I would really like to hear your point of view on that. Not that I'm trying to be a party pooper but I feel like we all need to appreciate the complexity of ai development.
I read that post for the reditt portrait image of the lady holding the paper. he used photoshop to add the text. Shame you missed the historical image 3 lol...may have been why he was banned.
TBH, most of these photos only trick you if you take a quick casual glance. All are easily identifiable as AI if you look closer. Maybe MJ 7 will be indistinguishable from real photos.
The original post from reddit had used those words taken from a photo of actual handwritten pen and then photoshopped into the file. There was also some inpainting involved I believe and the post got taken down because the third image was of h!tler as a shitpost. Still was really great and the character was consistent. It was all done locally with stable diffusion.
All of these AIs trying to fly before they can walk is the biggest issue. Focus on getting consistency i.e. characters that look the same in different environments AND pose from any angle. Fixing the hand, feet, limbs and weird eyes thing. Once they do that, they can do anything from there.
Midjourney would need to 10x their Art generation quality and power for me to justify subscribing to their platform. I can create some impressive art on my PC for pennies with SDXL that rivals midjourney, especially since midjourney does not have controlnet or other bells and whistles that the opensource does. Ive already sunk $4k into a 4090 PC this year, so Midjourney would need to destroy the AI world for me to pay money for it.
Lightfield Labs has created real 3D. It's physical three dimensions. I've seen a couple clips on this platform. If there's a demand for holodecks then they will make them.
The place to look to see if an image is AI-generated is still the hands. It's getting better, but hands and (to a lesser extent) feet are still a problem for AI.
Would this help bring more features at a faster rate to existing apps? There are some apps that I like, but the developers take forever to implement popular features that everybody is asking for. I wonder if AI can help with this.
@MattVidPro which is crazy, it feels like coding would be what ai would be good at, not art, but your so right. On the flip side, I can already use gpt 4 to doing the majority of my coding, I just have to clean it up to make it actually run, but it does >80%. I just want to be so lazy I dont have to do any of it... 🫠😆
Hello there, may I offer a couple of tips? These came to mind even before I started watching your video. Firstly, could you consider using more accurate and less sensationalized video titles? They often set expectations that aren't met, and let's face it, not all your viewers are children. Secondly, while I appreciate your reactions to news and events, try not to exaggerate their importance. I value well-reasoned critiques and a clear breakdown of what works and what doesn't. A more constructive and organized approach would be more beneficial, in my opinion. Feel free to take or disregard my advice. I'm merely sharing my thoughts on how to enhance the content of your channel. No offense, but currently, it feels a bit too adolescent-focused. However, I still follow your updates because you're quite informed about various studies and research. I just wish your analysis was taken more seriously and presented in a more structured manner. This is my honest feedback. You can either consider it or ignore it. It didn't take me long to write, so I haven't invested much time here.
I'm not having much luck with AnyText. If I don't specify a position, it puts the text in weird places. If I specify a position, the text is usually unreadable. Ah, well.
Dude! The lack of consistent characters is the biggest draw back from AI. It’s progressing so fast
12:12 The ‘consistent characters’ is the most interesting part of this video. It deserves a video of its own. I hope we will see truly ‘consistent’ characters, not just ‘similar’ characters that we’re seeing in Midjourney, where the character looks similar, but Midjourney adds different characteristics to that character in every image, such as changing their clothes to something else. A truly consistent character will look consistent in every scene, and their clothes won’t change either, unless we prompt for their clothes to change.
Also, I’d love to see an AI tool that can not only do consistent characters, but also pose multiple consistent characters together in a scene. For storytelling, we often need to create images of scenes where two or more of our consistent characters are interacting with each other.
there are certain tools/extensions that are getting dangerously close to this 'consistent character + compositing characters'. Once it is done, i think industry pros will have no choice but to make use of it.
I will do a deeper dive at some point!
@@MattVidPro Thanks Matt. I would love to see you dive deeper into this. These types of features would be very useful to anyone who wants to use AI for storytelling.
You can do consistent characters in Comfy UI with an IP adapter. Nerdy rodent has a couple of videos on it.
(Probably any SD but I have no idea how it works in A1111)
We have consistent characters in stable difusion. Lora nad dreambooth. Dreambooth makes 100% copy of person pr character. Super consistent
ITs strange to think by 2030 we could have AI generated games and worlds more detailed and beleivable than anything a triple A game developer can make with a team of a few thousand people.
I sure hope we do.
More like by 2025
Would be easy for ai. 3000 devs apparently aren't enough to make a finished game
2024-2028
We will have more ways to waste time than ever. I can also see steam popularity fading since many of these projects will be open source and indie games of today being really easy to make with AI in a few years. The holodeck thing I am sure was about VR
I've been saying for some time that these image generators need to build a full 3D model of a recurring character for easier generation of those characters.
That would also improve the temporal consistency of text-to-video models. They could generate 3D animated scenes instead of 2D animations.
Yes! And those models need different, adjustable layers for accessories, clothing and of course: hairstyles. It's like merging Midjourney with Blender or Unreal Engine etc.
@@jarblewarble already happening. There's a Text to Video AI that does entire short SCENES with persistence in the background and characters from a script.
We'll Text to Video for entire sequels to past movies and TV shows in the future, and pass them around like Tiktok vids to our friends.
@@brianmi40 What's the name of this AI? I don't think I've seen it yet.
Some do
Since you asked, it's not even the obvious weird ish that gives them away. Hair/fur, skin, fabrics, textures, reflections, coloration, have subtle plastic/ painterly/ unified qualities in all these pictures that aren't present in real life. Object placement and alignment would be the next things. Then obviously when things melt together lol
The zero stylization and raw mode really surprised me. I've been producing extremely photorealistic shots by setting stylize to 1000 and making my prompts short and non descriptive using Midjourney V6
I basically only use MJ 6 in raw for the extra coherency
One of the best realistic checkpoints of Stable Diffusion is Epic Realism.. It produces images of normal people (not the typical profesional models), with a lot of details and imperfections on skin, with spots, wrinkles, etc... Specially if combined with upscalers like 4x_NMKD-Superscale-SP_178000_G
I also think using the Turbo XL versions of checkpoints on Stable Diffusion are super good at creating candid type images with only 4-5 steps.
Very cool. I was one dalle and chatgpt day one. And was using Google colabs before that.from watching a few of you longer videos, I can tell you understand really understand the image generation community. Thanks for the videos!
Stable Diffusion and ComfyUI is the best. You have so much control over your compositions with ComfyUI
I think by the end of this year or early 2025, we will all be in shock at the generative tools we will have available. Then we will all be lost in the Elmo sauce. 😂 For some reason, I think I might start using that phrase. 🤔
Thank you Matt! 😀
It's odd now that the amateur aspect from pictures became the new high tech from ai generated images.
3:00 You forgot to mention software compression in your digital camera compared to a phone camera explanation.
6 Years from now, this will look like Ancient Tech from the 80s or 90s compared to today's Tech.
Amazing like always! ❤
Go on matt
Thank you!!
Cool content! So is midjourney the best for photorealism?
Great video, Matt. I appreciate you breaking things down and sharing the news.
My pleasure!
about the ai influencer post, the note text was added in photoshop
Came here to say this. I saw the original post and comments.
🎯 Key Takeaways for quick navigation:
00:00 🖼️ *AI Image Generation Trends*
- Overview of AI image generation trends in 2024.
- Highlights DALL-E 3, Mid Journey V6, and Stable Diffusion.
- Discusses the realism of images generated by Mid Journey V6.
05:03 📸 *Mid Journey V6 Realistic Image Prompt*
- Explores the Mid Journey V6 prompt for creating realistic-looking images.
- Demonstrates the settings and results of the prompt.
- Emphasizes the community's desire for creative expression in AI-generated images.
10:12 ✒️ *AnyText: Editing AI-Generated Text in Images*
- Introduces Alibaba's AnyText, enabling text editing on AI-generated images.
- Showcases the ability to modify text in different languages.
- Discusses the impact of AnyText on AI art generation workflows.
11:50 🌐 *Future of AI Art: Visual Text Generation*
- Explores the potential of visual text generation in AI art.
- Highlights the significance of combining AI-generated imagery with editable text.
- Considers the implications of this technology for the future of AI art.
13:00 🤖 *Consistent Characters in AI Art Generators*
- Discusses the emergence of consistent characters in AI art generators.
- Showcases an example of a consistent character created with Runway.
- Explores the potential impact of consistent characters on AI image generation.
15:16 🚀 *Mid Journey's Bold Claims for 2024*
- Addresses Mid Journey CEO's statements about achieving the "hollow deck" by 2024.
- Discusses the potential development of a real-time open world by the end of 2024.
- Expresses skepticism and curiosity about Mid Journey's ambitious plans.
Hey Matt. What’s the news with midjourney’s 3D image setting?
There is already holodecks in Australia from a point cloud company axiom holograms which used to be Euclidean holographics
Can you do more on pika ? Also runway released aspect ratio customization now too so you can change to 9:16 etc
Great overview, thanks! I hope we can at some point create consistent virtual characters that we can pose, animate and direct (playing out a script or convincing dialogue.) All the seeds to make this possible are already out there, but the scattered workflows have to be brought together into one platform with easy UI.
Dude, if we get holodeck level of immersion I want to actually walk around on a holodeck Enterprise and then talk to the AI Star Trek Next Generation cast and just chill out eating some Gagh with Picard. Awesome video.
Fascinating stuff. Agree 💯
Looking most forward to consistency, particularly for costumes. Can't wait for it to become available for Stable Diffusion.
Did you get a new editor or are you just getting better at this?
Ive been a decent editor forever, just trying out some new editing tricks to see how it resonates with the audience
Live long and prosper.
Hollowdeck is a Star Trek reference
Definitely so excited for the field of AI art moving to 3D. It indeed is the only way to go for creating believable videos. Though, we still need more flexibility in adjusting AI outputs, including purely manual adjustment for last touches. Still a few years away, but it's heading in the right direction.
Speaking of character consistency, Insight Faceswap with Midjourney is pretty good.
I will make a note of that!
Character consistency is such a high-demand item that SOMEONE is going to figure it out, eventually--probably all the AI image generators will have it, someday.
Just a matter of time 4 sure!!
Professional artists, take a moment to absorb how incredible all of this is, but please remember to protest the use of these tools on the job (even if incrementally, because who tf can stop it). Stand with your union or unionize. If the convenience of AI Art can reduce a team of 100 to a team of 10, the chances of you being one of the 90 are pretty high.
I find the detail amazing, that in the „first date“ picture, both look slightly off camera, as if they watch themselves on the phone screen and do not look into the camera - just as we doofuses all do. 😅
I tried the same Elmo prompt in Dall-e and it said copyright infringement. How did he get it to work?
These images are unbelievably believable!!!
Consistent character creation wasnt really working before! If its been solved then content will definitely moving from static to dynamic which means authors will write worlds and characters in place of just stories! Cant wait!
I can't do the Sesame Street ones in GPT-4 (Dalle3)--it pings me for copyright. How did you not get hit?
I switched out the copyright info for something similar, and it gave me an image, but it wasn't amazing, just OK. Then I went to the Dalle standalone site, and used the copyright prompt, and it worked great--much better than within ChatGpt...is there a reason? Aren't they using the same model?
Amazing 😍 🔥
We had all that since the first Loras in Stable Diffusion... but the hands are impressive.
Matt! Happy to see a vid today! Ps. Idk if you got new glasses or a haircut, but lookin good!
just a haircut thanks!!!
Is it the first time you wear that glasses? They look great! Not ai generated at all.
Great vid, as per usual, especially your enthusiasm that is contageous
Thank you! And no actually!
Great Video. I might try AI one day
Apparently people got an email fom openai saying that the GPT store is coming next week. Do you know which day?
No but I also got this email
I think midjourney ceo meant it more along the lines of prompt generated nerf style worlds that can be navigated but with a vr headset we’re again realistically 1 step closer to holodeck
A similar effect can be achieved on Dalle-3 using the prompt ‘photo taken in 1996’ or any other year. Really outstanding results can be achieved if you tell it the details of what you want to see.
Will give that a shot! I think D3 has some stylization settings that can’t be “switched off” like midjourney
@mattvideopro, if I want to mix non-cell filming with A.i. such as midjourney, what do you suggest is the kind of camera that can get the job done without costing a fortune?
10:11 'workflow included' means its a comfyui workflow thats in the image. All you need is the original image and pull it into comfy.
if you upscale the image and downscale for sharing you can fake perfectly because you are creating a lot of detail in the background 4:06 and then pixelating it when downscaling afterwards thus faking the image perfectly
Great video!
Glad you enjoyed it
How much caffeine do you hit before you film? :D
Enjoyed the vid. Keep it up, Matt.
Edit: Ah! I was watching on 1.5x
😮😄
The fun irony that AI is going to produce more "realistic phone pictures" when nearly every company is rushing to add AI upscaling to their phone cameras.
That second DALL-E 3 image of Elmo looks like it could have been shot on an iPhone. The latest ones are capable of automatically applying portrait mode to people and pets when you take a photo. Not sure Android flagships can do something similar (probably can), but if not they'll be able to real soon.
So those kinds of "slightly blurry" photos of subjects are gonna get more believable as they get uploaded to social media more and more.
+1 for the X-E4. 😀
Nice, very informative
Consistency, realism, truth, facts are some of my criteria
As incredible as those new Midjourney images look, I'm not too interested in the progress of Midjourney/DALL-E simply because they're both online-only models with limitations and restrictions.
But I'm VERY excited for the possibility of good offline 3D model generation coming this year. Stable Zero123 is a good start, but with some refining and improvements by the community (mainly to reduce the VRAM needed), it could truly become something amazing.
DALL-E is better
i cant wait for character consistency frustration to be a thing of the past. i have been spending way too much time with my workarounds/workflow for getting around current ai's shortcomings in this regard and it just gets exhausting ... and makes art less fun
Similarity in general look between multiple people is usually a tell. Also, the situations created can give away whether or not an AI created it.
Some of these AI images were simply insane. I could've sworn those were real legit photos.
12:21 Consistent characters are coming to Midjourney very very soon. I. Can't. Wait. 😃
The candid photos got some great ones had alter the prompt but love it n did it better in dalle 3 lol
Why do I feel personally attacked at 13.41😅😅😅
Real-time open VR worlds in 2024 is not a stretch actually. Open source... UEVR was just released a couple days ago.
I draw art for a living. I draw to create visial depictions of my imagination. I enjoy the result, not the process. AI is getting closer and closer to being able to take seconds to produce what it takes me 10 hours to produce. I LOVE AI!
You are one of the few artists that aren't hating or being salty about that. I actually believe it helps artists more than taking them credit or so
Exactly. I don't know why anyone enjoys the process of figuring out proportions/tediously drawing out microscopic details to make art. Maybe some people find it relaxing, but I find it more enjoyable to see an image I wanted without wasting an absurd amount of time and effort on it.
And this is why you'll get ahead. Learn these tools and go above and beyond with them. We are living incredible times in terms of creativity. In fact, this is the stuff of dreams. The harsh reality is that people who sell single static illustrations will not survive. But on the other hand, you can create entire universes. Imagine hand drawing a particular character and being able to replicate it in multiple poses and insert it in different scenarios in minutes, instead of redrawing the same char over and over? You can create comics quickly. And create entire set of concept art rather quickly. You can have your studio for doing stuff you love with 3 people instead of 30, because output willl increase so much. You can still introduce manual work in the workflow, in fact I think that's how AI is used for the best.
People who reject AI will be left behind for good.
15:50 "Holodeck' = real-time virtual environments.
Scrolling through those images, text is the easiest thing to spot, otherwise it's hard to tell at a quick glance.
Hey matt it's me drdimples 👋
I wonder when _Dall-E 4_ will be released...
2025 is my guess
Think about where we were were at this time last year - let that sit in your mind for awhile...
If the GPT store disrupts the appstore and google play, it might end up being one of the momentous times of this decade. The expectation!
It won't because more of the applications as people upgrade hardware will be done locally. I have a decent computer setup and can do most of what done in ChatGPT locally with a combination of opensource models. However these next 3 years will be big for them
Wait, Midjourney doesn't even Lora yet? Hmm, and here I am preparing a dataset for another consistent character.
Mat, I appreciate your optimism and excitement about the progress in AI development. It's what I enjoy most about your content. I used to feel the same way in the beginning but not anymore. But it has come to light that many of these technologies are actually ripping off content creators. In light of that, how do you keep the same energy? I would really like to hear your point of view on that. Not that I'm trying to be a party pooper but I feel like we all need to appreciate the complexity of ai development.
Nice! I've always been fascinated by ai art. Thanks for the video.
They also need to make characters consistent from frame to frame also for video generation.
I read that post for the reditt portrait image of the lady holding the paper. he used photoshop to add the text. Shame you missed the historical image 3 lol...may have been why he was banned.
oh this is.... scary to put it at the least
Love this optical AI image Brain trick discussion Matt gppd job the alternate universe of elmo lol
Whoa... MattVidPro really into News Anchoring Daddies, I'd never knew that. Whatever floats ya boat.
Dawg not like that 💀💀💀
TBH, most of these photos only trick you if you take a quick casual glance. All are easily identifiable as AI if you look closer. Maybe MJ 7 will be indistinguishable from real photos.
The original post from reddit had used those words taken from a photo of actual handwritten pen and then photoshopped into the file. There was also some inpainting involved I believe and the post got taken down because the third image was of h!tler as a shitpost. Still was really great and the character was consistent. It was all done locally with stable diffusion.
All of these AIs trying to fly before they can walk is the biggest issue. Focus on getting consistency i.e. characters that look the same in different environments AND pose from any angle. Fixing the hand, feet, limbs and weird eyes thing. Once they do that, they can do anything from there.
Midjourney would need to 10x their Art generation quality and power for me to justify subscribing to their platform. I can create some impressive art on my PC for pennies with SDXL that rivals midjourney, especially since midjourney does not have controlnet or other bells and whistles that the opensource does. Ive already sunk $4k into a 4090 PC this year, so Midjourney would need to destroy the AI world for me to pay money for it.
But why a 4090 for stable diffusion?
@@lollihonk I'm a pc enthusiast
Lightfield Labs has created real 3D. It's physical three dimensions. I've seen a couple clips on this platform. If there's a demand for holodecks then they will make them.
It was really easy to see the images at the start were fake, because the first thing I look at is the text.
The ceiling is a little weird?
There is an eye on it.
The place to look to see if an image is AI-generated is still the hands. It's getting better, but hands and (to a lesser extent) feet are still a problem for AI.
Text as well is always garbled
I wanna see where generative AI in gaming goes.
I cant wait for ai coding to be as good as ai art
Would this help bring more features at a faster rate to existing apps?
There are some apps that I like, but the developers take forever to implement popular features that everybody is asking for. I wonder if AI can help with this.
Same here. Big requirements for that though
@MattVidPro which is crazy, it feels like coding would be what ai would be good at, not art, but your so right. On the flip side, I can already use gpt 4 to doing the majority of my coding, I just have to clean it up to make it actually run, but it does >80%. I just want to be so lazy I dont have to do any of it... 🫠😆
If you look at the sign and text with the A.I. generated images it seems more often than not that they are gibberish.
Hello there, may I offer a couple of tips? These came to mind even before I started watching your video.
Firstly, could you consider using more accurate and less sensationalized video titles? They often set expectations that aren't met, and let's face it, not all your viewers are children.
Secondly, while I appreciate your reactions to news and events, try not to exaggerate their importance. I value well-reasoned critiques and a clear breakdown of what works and what doesn't. A more constructive and organized approach would be more beneficial, in my opinion.
Feel free to take or disregard my advice. I'm merely sharing my thoughts on how to enhance the content of your channel. No offense, but currently, it feels a bit too adolescent-focused. However, I still follow your updates because you're quite informed about various studies and research. I just wish your analysis was taken more seriously and presented in a more structured manner. This is my honest feedback. You can either consider it or ignore it. It didn't take me long to write, so I haven't invested much time here.
You're awesome bro.
no u
Somstimes I can tell it is AI, even when it looks very realistic. I make images, and if you do, it helps you spot them. Maybe on a subconscious level.
For what it's worth I really like the AI art, but I prefer to use it for upscaling textures and stuff like that
five fingers and a thumb with cross eyes gives most away
bing requires one to three generation o create a image like v6
there is a third post on SD subreddit that it's still alive
I'm not having much luck with AnyText. If I don't specify a position, it puts the text in weird places. If I specify a position, the text is usually unreadable. Ah, well.
Darn. I’m going to look more into it
Most powerful AI image creator....
Matt: Make a lemon character.
Yes
This is cool
Those pictures are pretty good. Only a few I could tell were AI. I think the giveaways are eyeballs and any text on screen.