Deepfloyd IF might win against Midjourney. I still want to do further testing. I prefer an image model that incorporates all aspects of the prompt first and foremost, while MJ might be clearer or more aesthetic, it often ignores parts of my prompt entirely.
@@jams2u786 yes, probably commercial license will be paid. They do need to make money somehow tho we could argue that ai generated art is not owned by anyone as it was created by tool.
Except it's a total pain to use. It's not totally local like A1111. What is this crazy 'notebook" stuff? It's all tangled up in Huggingface. It's like half-software, some coding required.
this is fake opensource, tying to attract free programmers that will test their sht for free. Check the license you cannot even publicaly use the images. Check the restrictions of the license. Same as facebook fake open source and openAI fake opensource and microsoft fake open source. Only google offers true unrestricted open source as a giant.
I can't believe this just came out after I spent 12 hours in front of a computer trying to keep up with ai tools. Insane progress. Thanks for the video!
:p Awesome! Yes, words have sucked so bad with regular AI, and I've tried several sites, and several apps. Hehe. This'll be fantastic for selling them, and should be good for consistent images that you want similae for children's books (what I have written by AI, but no images for it due to inconsistency), and other things.
I just tested it on Huggingface and i already noticed some other major difference, and that is generate images of cars. It does a way better job from the start then SD. It looks way more realistic there to generate a car in terms of design.
@@WeirdSmellyMan how is it open when you have to sign with name and email? That is by definition not open, how can you see that differently? My email and name have value that I have to trade in for a crappy license, uh no.
@@damien2198 it's still opensource. The menacing of opensource is, that the sourcode is public and can be read by anybody, the licence used dint chnsge anything in the fact that the code is opensource. Also who controls if you sell the AI images produced by this AI? There isn't any true indicator so if someone ask you, you can just say it's made with stable diffusion and so you are allowed to sell it.
@@electricz3045 you cannot do anything with it, even run if not for research purposes, and you are limited on what do can even run it for. not opensource.
:3 Awesome! Yes, words have sucked so bad with regular AI, and I've tried several sites, and several apps. Hehe. This'll be fantastic for selling them, and should be good for consistent images that you want similae for children's books (what I have written by AI, but no images for it due to inconsistency), and other things.
Hey Matt, hope you are great. Keep it coming, you put out absolutely awesome videos. I feel like I will never get to where I want to be, before I'm 3/4's in the gave, LOL.
The text output is definitely very impressive, and some of those output images aren't bad either. I think consistently Midjourney will still provide images that are on average better. Although, these new AI art generators will be better at a specific thing over Midjourney i.e. text, inpainting etc. Very keen to get started with this!
Many many thanks for the heads up on this! I just tried a few prompts on hugging face. Indeed, there is nothing like it in terms of handling text. However, the imagery it produces is way way behind midjourney or SD. In producing text, it often misses letters or misplaces them, but still that is way more than others can do. Let's give it time...
I love how AI news is happening so fast that by the time you finished the video there's already new updates to the topic lol Thx for doing what you do, you're the best !!
I've been playing with the HuggingFaces demo and... I'm not really impressed? It's better at text than other models, for sure, but... maybe there's some settings that aren't right, but the upscaled images are often distorted. Especially of human faces... even with things like "ugly, distorted, monster" in the negative prompt, the resulting faces range from mildly awkward to nightmare fuel...
..."with things like 'ugly, distorted, monster' in the negaqtive prompt" 😂😂😂 That was fking hilarious--and I totally agree. I'm playing with it for creating mockups for my design business (yes, I know, no commercial use yet, I'm just testing), and my prompt was "a 24 year old girl with blonde hair wearing a Bella Canvas tshirt with a boho background". It was still kind of impressive, but those god damned eyes will eat away at my soul.
This is cool but even more of a let down for the general user that the vram is so high when Nvidia (your only option for windows users) refuses to give any reasonable amount for non-ludicrous prices.
Claiming it beats MJ is a massive stretch. I tried it and honestly it feels like MJ version 1 or 2 at best. But again, it's just the beginning and I don't expect anyone to launch something out of the gate that can compete with MJ at this time.
You can tell he barely has a clue about MJ. That or he's getting kickbacks with all those affiliate links, clear embellishment, and out of place enthusiasm. With that being said, I hope this tool improves down the road because I love that it's open-source. In the meantime it's a candle to a flame
I like how AI made an accidental sarcasm there by putting a gun inside the cover of heart shaped box even it's not specified on that short prompt in 13:54
@@jopansmark That is true, but it's model is far more capable. For example, try to go make a well designed and very aesthetically pleasing horror monster in Stable Diffusion. It will take you quite a long time, if at all, unless someone comes up with a new trained model which contains relevant information. MJ's V5 can handle a lot of these monsters fairly easily, and is flexible enough to mix it well with other concepts. Inpainting, embeddings and other tools are powerful extensions of the base model, but are still limited to the initial training.
I'm willing to bet controlnet is the solution for near perfect text generation like almost on every prompt Cant wait to see someone apply the nvidia video ai method with this
This is very nice. But I really want an AI that will allow me to work with my own original characters and pose them together. I wish I could use it to make a comic. So, consistent characters & backgrounds, as well as posing, are all very important features for an AI to have.
@@rabanal_josh64 Hopefully, we will see this feature soon. Consistent characters, as well as posing of our consistent characters, would be a very important feature.
You can actually use the hyperlink to a preferred image in midjourney as a reference for any prompts you afterwards, essentially making a character you previously generated as the referenced character in future prompts, u just need to include the hyperlink in prompt and - - the image stock number
Darn, that is really good for a base model. Mostly solved text. The other major weakness of these models is getting multiple characters/things to interact with each other in a believable way. Is this model any better at that than stable diffusion?
There's node editor for StableDiffusion which allows manual composition using multiple generations, Z-buffer, vector poses and the rest of crazy tooling. But if you wantt something super complex from just one prompt, then no, it doesn't exist yet. I suspect it may be possible to train a model to create setups for this node editor, but I don't know whether anyone has tried.
Topaz Video Enhance AI seems to remain the best video upscaler at the moment, and I haven't heard about Topaz making a good progress for quite some time. And Diffussion models are too slow and inconsistent for video, so the current progress in generative models doesn't improve the state of video upscaling.
Peak of science! That's beautiful! Imagine paying 30$/month when people get these images for free(almost, at least without this subscription based business model(I hate it))
Never say never lol I have heard that Midjourney is working on lettering. DF is pretty incredible but I still give MJ the crown as far as quality. Based on what I have seen of DF so far.
I honestly believe that its clear the ai padora box has been opened and cant be closed. We better get on as a community and build our ai or its gonna be over for use. We got two future ahead of us, under the boot of corporate ai's, or we balance the field and live in megaman battle network world. Shiiieeet, imma start working on my megaman.exe
Love your channel, I tell others about it. One tip though, can you warm up your video portions of yourself? It’s so bright on our TV’s, it washes you out.
I personally think BlueWillow is far better than others free AI. BW is 100% free tool. Though BlueWillow is so early stage. I experimented with BW, and I'm really amazed by the results
I use BW a fair bit and for free it is excellent. I would just hope they include more features like Midjourney. If Midjourney opened up for free it would runaway with the lead out of the two, I have just cancelled my subscription to MJ because there are other AI doing a too similar job for free, Midjourney will die behind a paywall! I think of Blue Willow and Midjourney as a sort of VHS - Betamax battle, I just not sure which one is which right now!
Nah, it's not open source so I don't trust it. Also, is that bots? Like I seen two exact same comment about Blue Willow from two accounts. Really shady
So far, it definitely beats it when it comes to cohesiveness, but I don't believe DeepFloyd can beat Midjourney when it comes to making characters, yet.
bro, you are pumping out video after video after video, you are probably an ai yourself ;D love your videos, love your enthusiasm, love your excitiment and your energy. Do you mind me asking if you use dark mode on Twitter or everywhere possible? Your videos are so bright^^
Either I'm doing something wrong or I'm expecting something different. It does well at words but that's about it. People are severely distorted with messed up faces, arms coming out of the middle of their chest, etc. Typing any prompt with a known person like "Walter White" produces half of the results matching some random person and the ones that look like WW are distorted. Other than text, this looks like a very early AI text-to-image generator. Like in the realm of the old craiyon.
It's to be expected. Videos like this are always baseless hype. "I've used it before and it's amazing." Okay where's the proof? And where's the prompts to go with it? They always lie and show cherry picked results. Easy way to know if a video is clickbait: check if it has Midjourney in the title.
From what I could understand of their license agreement, you can't do any commercial use with anything made in it. That means, can't use it as a book or album cover, etc. That sucks :(
Well, DeepFloyd definitely beats Midjourney is some aspects and in some prompts. The same is true for StableDiffusion, especially specialized fine-tunes. The only problem is that Midjourney still looks better in 80% of cases. :) And doesn't look like DeepFloyd would be able to beat Midjourney with its heavy reliance on upscaling (4x + 4x), as textures are ruined. But it's "modular", so the IF-I-XL is a huge progress either way.
Oh no! If the community really gets into this I will have to buy a 4090. My eyes have grown hypersensitive to the typical SD coherence problems that have never been properly solved, and I just don't find it very fun unless I can run my models locally. But man! I am super impressed that something this advanced can run on consumer hardware at all. It looks extremely promising!
I happen to already have a 4090 in my PC, so if that's what it takes to run AI models like this locally with good performance then I may as well go ahead and do it. How convenient.
Or you could buy a Mac with Apple Silicon with the requisite amount of ram. The Video ram is shared with regular ram so much bigger than you'd get on Intel video cards by default. M1&M2 also have ML cores that contribute to FAST rendering for some models. I've got an M1 Pro with 16gb ram which can do many open source calculations. The M2 Max would be the one to consider today if 24+GB video ram is required.
This looks like a really good model but it looks like it has inconsistent lighting across some of the generated images. Looks kinda like photoshop jobs with the ability to add text where as midjourney csnnot do text but still looks like it is producing more consistency and realism in its generation. Very powerful tool but so far ill stick to midjourney v5. Also missing aspect ratio options and such
Looks like proper prompt generations but with a poor aesthetics. It should be great as a starting point for text2img in PF and then move it in to SD or MJ for img2img
Awesome! Yes, words have sucked so bad with regular AI, and I've tried several sites, and several apps. Hehe. This'll be fantastic for selling them, and should be good for consistent images that you want similae for children's books (what I have written by AI, but no images for it due to inconsistency), and other things.
From what I understand from the agreement is that all there images are subject to copywrite and your not allowed to change them either. sounds like a lot of future litigations.
This looks very exciting! Is Deepfloyd able to be run on a Mac as well as PC with dedicated graphics card? Technically the M1 Max should be capable enough, but has anyone tried setting this up on a Mac yet?
I wait for the day strong open source multimodal models become bidirectional 😍 - so you can input text, getting a rendering of your prompt (including written text), and input an image, getting a description (including the text written in the image)...
Nah, it will be used mainly to add voiceovers to games that had none, that's about it. Modern games will probably not make use of it aside from maybe randomly generating chatter that will be then censored/curated by devs and placed in game as they do right now but with more variety. If you think anyone is going to put something like chatgtp in game so you can talk with npcs it's not gonna happen with current hardware not mention it had to be very limited to topics only related to that game world resulting in pretty much same thing we have right now, tree of few choices. Tech demos and projects made for fun do not reflect realities of commercial products.
The king in my book is and will continue to be PixAI. Its free, uncensored, has a ton of img gen models, and they're still updating the site. with the release of AnythingV5, I dont think anything can top it. These other img gen model devs heavily censor their shit to the point they arent worth using imo.
@@deathorb Its entirely free. You have to mess with it, but it is free. Make sure to have it on Low priority and only generating 1 img at a time. Mess around with sampling sizes, etc. The Credits are ONLY for generating faster. You cant buy them, and you get 10k free daily.
Just keep in mind it is unlikely to stay free long-term, just because of the high costs involved in running any of these sites. But we're definitely getting a lot of options lately. Mage is another that is pretty open about what can be generated, though it does cost to use anything but the default SD models.
@@ShawnFumo Thats why I have a $25 NovelAI sub as a backup. Its not near as good for images, but still uncensored and you can tune it to produce some amazing images, just not as consistently. And for the price, its very much worth it. Not only do you get unlimited img generation, but also unlimited story generation, and the chatbot they use is fairly good. Im surprised PixAI has been around this long while being free, right now it has no monetization model. But, given how their updating it, we might see one sometime soon.
All these websites will pale in comparison with what Adobe has up its sleeve. I'm talking about Adobe Firefly. It will do to AI art generation, what Photoshop did for the digital art industry back in the day. Can hardly wait
Hey Matt, thanks for this but unfortunately using huggingface only produces small very blurry pictures...the whole page doesn't look at all like what is presented in your video. Is there a different link that you forgot to include? I followed what you put the description. I wish I could show you a screenshot of wat I see.
Amazing! This is going to be great. I appreciate your finding and showing all this. Slow down on the scrolling past everything, though! We're trying to see the prompts, too! 🙂
I am v.excited about an open version of Midjourney that will quickly become superior. I'm waiting for all AI/AGI to run on mobile phones but I am considering getting a laptop 💻 just so that i can play with AI art. Any suggestions of something affordable are welcome 😅😊
A recent model of iPhone or iPad will run "Draw Things" which is based on the open source Stable Diffusion model. It runs VERY fast on these handheld powerhouses due to Apple's machine learning cores built into recent years of Apple Silicon. PC fanboys have no idea how equal these tiny platforms are to their very expensive Nvidia GPUs in comparison. ~!
Once again, VRAM is the barrier. I started to feel the limitation of VRAM when I started using Blender and how GPU companies are basically robbing consumers with VRAM. But that was when there were only few consumers who needed a lot of VRAM. I nope this growing A.I. image popularity will make more consumers force GPU companies to add more VRAM by boycotting GPU's with smaller VRAM.
If only there was a company who said "Hmm, maybe if we make cheap GPUs with tons of vram we'll sell a lot!" I'd buy a gpu with 24gb of ram that can't game in a heartbeat. Same with AV1 encoding. They don't sell these things on their own for a reason though. They want you to pony up for a high end gpu instead.
Deep Floyd has better accuracy - isn't necessarily as nice looking as MJ5... Since MJ5 uses human feedback to get closer to what humans will think is cool. Since it's opensource, and since Midjourney has its own datasets, they may well be able to finetune this to produce MJ style images.
Deepfloyd IF might win against Midjourney. I still want to do further testing. I prefer an image model that incorporates all aspects of the prompt first and foremost, while MJ might be clearer or more aesthetic, it often ignores parts of my prompt entirely.
Yes let's do a compare!
if it is open source, why a user has to identify itself for just joining the webpage to download and use it?
The agreement says "Non-Commercial Use", does this mean generated imagery may NOT be used in anyway for business purposes?
@@jams2u786 yes, probably commercial license will be paid. They do need to make money somehow tho we could argue that ai generated art is not owned by anyone as it was created by tool.
Except it's a total pain to use. It's not totally local like A1111. What is this crazy 'notebook" stuff? It's all tangled up in Huggingface. It's like half-software, some coding required.
i love that open source is not getting left behind
We ALL need to support open source. That's going to be imperative for the "normal people" to have ai as powerful as these corporations.
Absolutely
And so that the Artist Can't sue it into Collapse.
@@spinninglink yup
this is fake opensource, tying to attract free programmers that will test their sht for free. Check the license you cannot even publicaly use the images. Check the restrictions of the license. Same as facebook fake open source and openAI fake opensource and microsoft fake open source. Only google offers true unrestricted open source as a giant.
I can't believe this just came out after I spent 12 hours in front of a computer trying to keep up with ai tools. Insane progress. Thanks for the video!
:p True, cherry-picked, like what debunked religious and debunked material atheists do. Lol. I use that term pretty often.
:p Awesome! Yes, words have sucked so bad with regular AI, and I've tried several sites, and several apps. Hehe. This'll be fantastic for selling them, and should be good for consistent images that you want similae for children's books (what I have written by AI, but no images for it due to inconsistency), and other things.
Great Job keep up with the good work.
I just tested it on Huggingface and i already noticed some other major difference, and that is generate images of cars. It does a way better job from the start then SD. It looks way more realistic there to generate a car in terms of design.
SD had very little car pictures fed to it, you can tell it very quickly but if you use correct Lora for model of car you want it works very well.
Absolutely, it's impressive to say the least, a huge milestone for the open source community.
no it is not, it is a milestone for the company that made it. Check the license
@@dik9091 yes it is.
@@WeirdSmellyMan how is it open when you have to sign with name and email? That is by definition not open, how can you see that differently? My email and name have value that I have to trade in for a crappy license, uh no.
@@dik9091 I didn't have to sign in at all.
@@WeirdSmellyMan the license prohibits commercial use, copying and creating derivatives, three core requirements of being open source
Remember your vid on Floyd way back, happy to see it release. Open-source AI is really the way to go!
Midjourney will just implement Deepfloyd to their code and charge for it.
They cannot Deepflyod is not really opensource license, just a shitty one for non-commercial pupose/research with tons of restrictions
Midjourney uses their own codebase, that's why v5 is so much better than SD
@@damien2198 it's still opensource. The menacing of opensource is, that the sourcode is public and can be read by anybody, the licence used dint chnsge anything in the fact that the code is opensource. Also who controls if you sell the AI images produced by this AI? There isn't any true indicator so if someone ask you, you can just say it's made with stable diffusion and so you are allowed to sell it.
@@electricz3045 you can read it but cannot do shit with it unless for research purpose with lot of limitation. not opensource
@@electricz3045 you cannot do anything with it, even run if not for research purposes, and you are limited on what do can even run it for. not opensource.
You are such a great source of AI news Matt, thank you
:3 True, cherry-picked, like what debunked religious and debunked material atheists do. Lol. I use that term pretty often.
:3 Awesome! Yes, words have sucked so bad with regular AI, and I've tried several sites, and several apps. Hehe. This'll be fantastic for selling them, and should be good for consistent images that you want similae for children's books (what I have written by AI, but no images for it due to inconsistency), and other things.
Hey Matt, hope you are great. Keep it coming, you put out absolutely awesome videos. I feel like I will never get to where I want to be, before I'm 3/4's in the gave, LOL.
Oh shit, I sent the first comment to ask you why I can't get into DeepFloyd AI? Thank you Matt😅
Unlike Stable Diffusion this model is only non-commercial use so definitely won’t have the same dev excitement
Oooh, good catch.
Noob here, does this mean that you can't take the pictures and sell them or using them as book covers or on print on demand things?
@@seemlessartcreations yea
@@anonymous49125 no as of now. you can sell your images. Laws are not put into place rn
@@aarohanyt7374 prove it
The text output is definitely very impressive, and some of those output images aren't bad either. I think consistently Midjourney will still provide images that are on average better. Although, these new AI art generators will be better at a specific thing over Midjourney i.e. text, inpainting etc. Very keen to get started with this!
Many many thanks for the heads up on this! I just tried a few prompts on hugging face. Indeed, there is nothing like it in terms of handling text. However, the imagery it produces is way way behind midjourney or SD. In producing text, it often misses letters or misplaces them, but still that is way more than others can do. Let's give it time...
I love how AI news is happening so fast that by the time you finished the video there's already new updates to the topic lol Thx for doing what you do, you're the best !!
The boot image with that text is a reference to Venus in Furs by The Velvet Underground which came out in 1967.
Uh, thanks.....
@@Tony_Baloney_69420 You’re welcome.
That image stood put for me too. I suppose it's heavily influenced by floyed throughout.
This is looking amazing. I can’t wait to see what the future holds for AI.
World domination 😈
This is great, thanks for keeping us all updated! Hope you’re feeling better!
3:04 They're called meerkats. They're related to mongooses.
I've been playing with the HuggingFaces demo and... I'm not really impressed? It's better at text than other models, for sure, but... maybe there's some settings that aren't right, but the upscaled images are often distorted. Especially of human faces... even with things like "ugly, distorted, monster" in the negative prompt, the resulting faces range from mildly awkward to nightmare fuel...
..."with things like 'ugly, distorted, monster' in the negaqtive prompt" 😂😂😂 That was fking hilarious--and I totally agree. I'm playing with it for creating mockups for my design business (yes, I know, no commercial use yet, I'm just testing), and my prompt was "a 24 year old girl with blonde hair wearing a Bella Canvas tshirt with a boho background". It was still kind of impressive, but those god damned eyes will eat away at my soul.
This is IMPRESSIVE, this can literally help us makes T-shirts, Logos, and much much much more! Such a great news from you, matt! TYSM
For non commercial use only by the looks of things. I wonder how that might apply to making a logo/tshirt design for your own business?
@@Fontgod wondering how they could actually inforce that since images can't be copyrighted?
This is cool but even more of a let down for the general user that the vram is so high when Nvidia (your only option for windows users) refuses to give any reasonable amount for non-ludicrous prices.
Claiming it beats MJ is a massive stretch. I tried it and honestly it feels like MJ version 1 or 2 at best. But again, it's just the beginning and I don't expect anyone to launch something out of the gate that can compete with MJ at this time.
You can tell he barely has a clue about MJ. That or he's getting kickbacks with all those affiliate links, clear embellishment, and out of place enthusiasm.
With that being said, I hope this tool improves down the road because I love that it's open-source. In the meantime it's a candle to a flame
I like how AI made an accidental sarcasm there by putting a gun inside the cover of heart shaped box even it's not specified on that short prompt in 13:54
It can spell, but MJ's quality and flexibility with V5 is pretty amazing.
Maybe Dreambooth us what takes DF there
Wait, MJ has flexibility? It literally has no inpainting, outpainting, training, embeddings and many more basic features
@@jopansmark That is true, but it's model is far more capable.
For example, try to go make a well designed and very aesthetically pleasing horror monster in Stable Diffusion. It will take you quite a long time, if at all, unless someone comes up with a new trained model which contains relevant information.
MJ's V5 can handle a lot of these monsters fairly easily, and is flexible enough to mix it well with other concepts.
Inpainting, embeddings and other tools are powerful extensions of the base model, but are still limited to the initial training.
Thanks Matt! I can’t wait to check this out!
I'm willing to bet controlnet is the solution for near perfect text generation like almost on every prompt
Cant wait to see someone apply the nvidia video ai method with this
Ok now I'm just waiting for someone to incorporate it into A1111
I am so glad that some open source program has defeated a big company program that takes a lot of money for its project
I came here expecting something similar to bluewillow AI, but you have blown my mind out the waters again matt.
This is very nice. But I really want an AI that will allow me to work with my own original characters and pose them together. I wish I could use it to make a comic. So, consistent characters & backgrounds, as well as posing, are all very important features for an AI to have.
Maybe we could see that in the upcoming months. But yeah, consistency is something creators want
@@rabanal_josh64 Hopefully, we will see this feature soon. Consistent characters, as well as posing of our consistent characters, would be a very important feature.
You can actually use the hyperlink to a preferred image in midjourney as a reference for any prompts you afterwards, essentially making a character you previously generated as the referenced character in future prompts, u just need to include the hyperlink in prompt and - - the image stock number
3:25 but the main thing, is the hands and how well done they are.
DeepFloyd IF generated a real Lamborghini for me, brought my ex wife back to life and cured my cancer. Thank you DeepFloyd 🎉
I'm excited AF for DeepFloyd IF.
Darn, that is really good for a base model. Mostly solved text. The other major weakness of these models is getting multiple characters/things to interact with each other in a believable way. Is this model any better at that than stable diffusion?
Better, yes, by quite a bit. Probably better at that than MJ.
There's node editor for StableDiffusion which allows manual composition using multiple generations, Z-buffer, vector poses and the rest of crazy tooling. But if you wantt something super complex from just one prompt, then no, it doesn't exist yet. I suspect it may be possible to train a model to create setups for this node editor, but I don't know whether anyone has tried.
@@Athari-P You're talking about ComfyUI
Cool, especially with text. Steep VRAM requirements though.
We're going to be able upscale old videos to 4k on the fly very soon. All videos will be high resolution.
Topaz Video Enhance AI seems to remain the best video upscaler at the moment, and I haven't heard about Topaz making a good progress for quite some time. And Diffussion models are too slow and inconsistent for video, so the current progress in generative models doesn't improve the state of video upscaling.
This is literally the case of the first stable diffusion model, its expensive to run, but look at it now. thanks to open source!
Amazing!
But can it draw hands?
This BEATS Midjourney ? we can't tell if we can't use it :)
You CAN use it! Link below!
OMG. The freaking Doc boot with the Venus In Furs lyrics #PunksNotDead lol
Peak of science! That's beautiful! Imagine paying 30$/month when people get these images for free(almost, at least without this subscription based business model(I hate it))
First to do text well, but midjourneys image quality and creativity is better imo
Never say never lol I have heard that Midjourney is working on lettering. DF is pretty incredible but I still give MJ the crown as far as quality. Based on what I have seen of DF so far.
I honestly believe that its clear the ai padora box has been opened and cant be closed. We better get on as a community and build our ai or its gonna be over for use. We got two future ahead of us, under the boot of corporate ai's, or we balance the field and live in megaman battle network world. Shiiieeet, imma start working on my megaman.exe
My favorite is still Bing create, it has the best colors, best composition, best ability to understand prompts, I think it's the best
It has a LOT of restrictions and limited usage
Love your channel, I tell others about it. One tip though, can you warm up your video portions of yourself? It’s so bright on our TV’s, it washes you out.
I personally think BlueWillow is far better than others free AI. BW is 100% free tool. Though BlueWillow is so early stage. I experimented with BW, and I'm really amazed by the results
I use BW a fair bit and for free it is excellent. I would just hope they include more features like Midjourney. If Midjourney opened up for free it would runaway with the lead out of the two, I have just cancelled my subscription to MJ because there are other AI doing a too similar job for free, Midjourney will die behind a paywall! I think of Blue Willow and Midjourney as a sort of VHS - Betamax battle, I just not sure which one is which right now!
Nah, it's not open source so I don't trust it. Also, is that bots? Like I seen two exact same comment about Blue Willow from two accounts. Really shady
"No I don't have a gun" is a line from "Come as you are'' by Nirvana. Misha and I are both fans, it seems!
So far, it definitely beats it when it comes to cohesiveness, but I don't believe DeepFloyd can beat Midjourney when it comes to making characters, yet.
It beats it when you need to generate Xi Jinping, lol
bro, you are pumping out video after video after video, you are probably an ai yourself ;D love your videos, love your enthusiasm, love your excitiment and your energy. Do you mind me asking if you use dark mode on Twitter or everywhere possible? Your videos are so bright^^
I will have to try dark mode… thanks for the kind words!
Either I'm doing something wrong or I'm expecting something different. It does well at words but that's about it. People are severely distorted with messed up faces, arms coming out of the middle of their chest, etc. Typing any prompt with a known person like "Walter White" produces half of the results matching some random person and the ones that look like WW are distorted. Other than text, this looks like a very early AI text-to-image generator. Like in the realm of the old craiyon.
It's to be expected. Videos like this are always baseless hype. "I've used it before and it's amazing." Okay where's the proof? And where's the prompts to go with it? They always lie and show cherry picked results.
Easy way to know if a video is clickbait: check if it has Midjourney in the title.
The quality is incredible. I am going to use it for future thumbnails on my channel!
From what I could understand of their license agreement, you can't do any commercial use with anything made in it. That means, can't use it as a book or album cover, etc. That sucks :(
I honestly am interested in the possibility of incorporating this new gen image generation models with the exist A1111 and comfyUI tools.
Do you know how many times I've heard "this beats midjourney" in the last couple of months?
I’m serious about this. I don’t toss that around
Well, DeepFloyd definitely beats Midjourney is some aspects and in some prompts. The same is true for StableDiffusion, especially specialized fine-tunes. The only problem is that Midjourney still looks better in 80% of cases. :) And doesn't look like DeepFloyd would be able to beat Midjourney with its heavy reliance on upscaling (4x + 4x), as textures are ruined. But it's "modular", so the IF-I-XL is a huge progress either way.
The most impressive thing aside from the text is that image of the 5 meerkats with different colored sweaters.
Finally I can ask the AI for some photscans of the uncensored Roswell documents.
Oh no! If the community really gets into this I will have to buy a 4090. My eyes have grown hypersensitive to the typical SD coherence problems that have never been properly solved, and I just don't find it very fun unless I can run my models locally. But man! I am super impressed that something this advanced can run on consumer hardware at all. It looks extremely promising!
I happen to already have a 4090 in my PC, so if that's what it takes to run AI models like this locally with good performance then I may as well go ahead and do it. How convenient.
Or you could buy a Mac with Apple Silicon with the requisite amount of ram. The Video ram is shared with regular ram so much bigger than you'd get on Intel video cards by default. M1&M2 also have ML cores that contribute to FAST rendering for some models. I've got an M1 Pro with 16gb ram which can do many open source calculations. The M2 Max would be the one to consider today if 24+GB video ram is required.
This looks like a really good model but it looks like it has inconsistent lighting across some of the generated images. Looks kinda like photoshop jobs with the ability to add text where as midjourney csnnot do text but still looks like it is producing more consistency and realism in its generation. Very powerful tool but so far ill stick to midjourney v5. Also missing aspect ratio options and such
Thank you for all the great information, thanks for keeping us all updated
Looks like proper prompt generations but with a poor aesthetics. It should be great as a starting point for text2img in PF and then move it in to SD or MJ for img2img
Awesome! Yes, words have sucked so bad with regular AI, and I've tried several sites, and several apps. Hehe. This'll be fantastic for selling them, and should be good for consistent images that you want similae for children's books (what I have written by AI, but no images for it due to inconsistency), and other things.
Of all the image generators I have used. I think Bing Image Creator generates the best images so far.
Can the images be used for promoting Social media content? I heard that you cannot use the images for commercial use.
FINALLY!!! IVE WAITED MONTHS FOR THIS!!!
You could have used an Automatic1111 extension for SD Text generation
Wow! This almost looks too good to be true !
My RTX 3090 has 24gb VRAM. Let's get back to this when it's integrated to A1111
From what I understand from the agreement is that all there images are subject to copywrite and your not allowed to change them either. sounds like a lot of future litigations.
I have been waiting for this for so long that I was seriously considering that it might never come out
This looks very exciting! Is Deepfloyd able to be run on a Mac as well as PC with dedicated graphics card? Technically the M1 Max should be capable enough, but has anyone tried setting this up on a Mac yet?
Looks great but I wouldn’t switch until this is more user friendly then I’d try it out.
Can it be downloaded to the computer to use it with A1111 or do you need another webui ? And if so, which file should be downloaded ?
I wait for the day strong open source multimodal models become bidirectional 😍 - so you can input text, getting a rendering of your prompt (including written text), and input an image, getting a description (including the text written in the image)...
the AI powered NPCs are the thing I'm most excited about. it's really going to change gaming dramatically
Nah, it will be used mainly to add voiceovers to games that had none, that's about it. Modern games will probably not make use of it aside from maybe randomly generating chatter that will be then censored/curated by devs and placed in game as they do right now but with more variety. If you think anyone is going to put something like chatgtp in game so you can talk with npcs it's not gonna happen with current hardware not mention it had to be very limited to topics only related to that game world resulting in pretty much same thing we have right now, tree of few choices. Tech demos and projects made for fun do not reflect realities of commercial products.
Yoo Matt.....thank you so MUCH for all this cool incredible info...you da best!!! Phil
The king in my book is and will continue to be PixAI. Its free, uncensored, has a ton of img gen models, and they're still updating the site. with the release of AnythingV5, I dont think anything can top it. These other img gen model devs heavily censor their shit to the point they arent worth using imo.
Is this one censored?(which includes not generating famous people photos)
thanks for that!!!
Not quite free really but good
@@deathorb Its entirely free. You have to mess with it, but it is free. Make sure to have it on Low priority and only generating 1 img at a time. Mess around with sampling sizes, etc. The Credits are ONLY for generating faster. You cant buy them, and you get 10k free daily.
Just keep in mind it is unlikely to stay free long-term, just because of the high costs involved in running any of these sites. But we're definitely getting a lot of options lately. Mage is another that is pretty open about what can be generated, though it does cost to use anything but the default SD models.
@@ShawnFumo Thats why I have a $25 NovelAI sub as a backup. Its not near as good for images, but still uncensored and you can tune it to produce some amazing images, just not as consistently. And for the price, its very much worth it. Not only do you get unlimited img generation, but also unlimited story generation, and the chatbot they use is fairly good. Im surprised PixAI has been around this long while being free, right now it has no monetization model. But, given how their updating it, we might see one sometime soon.
Super excited until you mention 16GB GPU. Can it reduce down to 8GB using half accuracy?
Yes but can it do accurate celebrity likenesses or consistant character design unlike Mid journey
If i woulda known this when getting a video card, i would have opted for one with more vram lol
All these websites will pale in comparison with what Adobe has up its sleeve. I'm talking about Adobe Firefly. It will do to AI art generation, what Photoshop did for the digital art industry back in the day. Can hardly wait
Hey Matt, thanks for this but unfortunately using huggingface only produces small very blurry pictures...the whole page doesn't look at all like what is presented in your video. Is there a different link that you forgot to include? I followed what you put the description. I wish I could show you a screenshot of wat I see.
Amazing! This is going to be great. I appreciate your finding and showing all this. Slow down on the scrolling past everything, though! We're trying to see the prompts, too! 🙂
I am v.excited about an open version of Midjourney that will quickly become superior. I'm waiting for all AI/AGI to run on mobile phones but I am considering getting a laptop 💻 just so that i can play with AI art. Any suggestions of something affordable are welcome 😅😊
You need a decent gpu to run AI locally, so a laptop isn't going to cut it. Better to just get a cheap used computer and plop in a gpu like the 3060.
A recent model of iPhone or iPad will run "Draw Things" which is based on the open source Stable Diffusion model. It runs VERY fast on these handheld powerhouses due to Apple's machine learning cores built into recent years of Apple Silicon. PC fanboys have no idea how equal these tiny platforms are to their very expensive Nvidia GPUs in comparison.
~!
@@razoraz So you're saying iPhones and iPads have a reputation for being affordable?
first ai image generator that can generate words without mistakes
I am super interested in the upscaling. But confused asf on how to do it.
Excited!
unfortunutly you can use the tool for reasearch use only, no comercial use allowed
bro, midjourney is light years ahead of this...
Once again, VRAM is the barrier. I started to feel the limitation of VRAM when I started using Blender and how GPU companies are basically robbing consumers with VRAM. But that was when there were only few consumers who needed a lot of VRAM. I nope this growing A.I. image popularity will make more consumers force GPU companies to add more VRAM by boycotting GPU's with smaller VRAM.
If only there was a company who said "Hmm, maybe if we make cheap GPUs with tons of vram we'll sell a lot!"
I'd buy a gpu with 24gb of ram that can't game in a heartbeat. Same with AV1 encoding. They don't sell these things on their own for a reason though. They want you to pony up for a high end gpu instead.
If it is open source Midjourney will soon implement it just like they did stable diffusion.
A wise man once said👆
It's not open source
Deep Floyd has better accuracy - isn't necessarily as nice looking as MJ5... Since MJ5 uses human feedback to get closer to what humans will think is cool.
Since it's opensource, and since Midjourney has its own datasets, they may well be able to finetune this to produce MJ style images.
The Venus in Furs reference. Nice!
AWESOMEY - except if you see the letters IF and it's not I.F. - it's probably "if" not I-F.
Looks really cool! Will it beat sd though?... The darth vader prompt looked really promising
The meme potential.
I can’t get an image to load . Just a blank square with an X in it
3:30 it‘s more about being able to draw hands and even hands drawing hands, than someone drawing a drawing, I‘d say.
YES!! This is going to be extremely useful 🧙♂
Please make a video about the Vits AI (where you can make songs using artists voice 👀)
"IF" by Rudyard Kipling? "If you can keep your head, when all around you are losing theirs...."
Very cool, but will it remove unwanted text?
MJ gives me a headache when it throws garbage text all over an image where it doesn't belong.
Try clip drop by Stability Ai
DALL·E 2 is the sapling. The sapling grew.