You did specify 2015, so before you say the phone looks weird or unrealistic, look up how phones looked like in 2015. There was no such thing as multiple lenses, and most of them had a simple small lense near the top in the middle.
Ideogram and Flux smartphones make no sense for 2015, I think that recraft is actually more accurate there, it does look like an old phone. Also you missed the fact that you can create your own image styles.
I think flux makes sense other than the multi-lens camera. The actual shape and look of the case looks exactly like what you'd see back in 2015. And the image quality looks way more like what my old 2015 photos looked like.
Exactly. I've used recraft a lot, for anime if he just screenshot an anime to use as a custom style it definitely would have generated something better.
i still think from an artistic and cinematic viewpoint is MJ, i just had a go at this and it made boring pics, i came back to MJ and my god the camera shots, the light, the coloring was off the chart. MJ weakness is prompt adherence , but it should be top of that leaderboard.
0:11This is impressive at a glance, but the generator doesn't seem to understand clothing. The green girl's left arm is very baggy compared to right. The brown girls' left and right sleeves are mismatch. The pink guy's right hand is flipped. The brown girl's left hand is in an extremely unnatural position. The list goes on. All 3 are also floating in space, with no consideration for how they jumped. With that said though, I'm still impressed by how far the technology improved compared to 6 months ago.
I chose the far right side selfie due to the single camera on the phone... even with the hallucinated "center cam"... a flip phone in the prompt could have killed it! ;)
14:46 Modern web apps MUST adjust for the mobile phone screens. As 95% of the users will use them from the smartphone. So showing one on mobile does not violate the condition. If we go from the statement of training, even professional UI designers sometimes could not tell if the UI is native or a web app (once displayed on mobile), so when image is tagged "web app" it can be either.
"if you wanted to swap this background in photoshop you're going to have a really painful time (...)" nope... you click on the subject to AI-select it, then you inverse the selection and you click "generate" and you change it to whatever background you need. easy as 1-2-3
It’s still asking a lot for a model to produce flawless output for any prompt on the first take. Most people using AI to generate imagery for real-world applications are using ComfyUI or their own custom coded pipelines to move images through a workflow that includes refinement steps for hands and faces. My own sense is that we’re getting close to having very large models that cannot possibly be hosted on consumer GPU hardware that can automate everything and produce great results. Google’s Imagen 3 model is clearly a gigantic model and produces very consistent results, for instance. But moving images through a few workflow steps with small models runnable on consumer hardware produces exceptional results already. I’m not sure which approach will win out in the end, but likely the big cloud models will eventually take over.
I agree. I also am surprised at what many people think are a "realistic" looking image as well. I have spent a lot of time on various AI generators as an experienced photographer/retoucher and I know that to get real looking results is a refinement - almost an art of using the generator. For example- Mid journey mayn't do so well with a single line prompts, but none of them do very good IMO. But if you work on styles, refine the prompts with good references, and use the newer edit features for fine control - the results are quite amazing. We should be judging these on the absolute best results they can do- not the laziest results.
I'm not sure I've seen him take these simple criticisms into account in future videos. He reports on a lot of stuff, but he leaves in enough small problems that I think it heavily limits his growth.
I have to say, you get the best results at 4MP or higher which they do state. So if you do less than you get bad results. You did do most of the generations at 1024. Can't say if it was a fair assessment. @25:00 If you would have picked the style from the list you may have gotten better results, however it wouldn't have been using their best model. I think for a more fair assessment, you'd have to render them all at close to the same pixel rate. There was also a whole section of 20B styles styles at recraft that you missed. I don't know why they don't automatically put all the styles in the list without selecting the combo box first. Great job on showing how to do the layers. I kept only getting recraft option and was getting annoyed with it. I didn't realize that you had to merge down the layers. Overall great job!
In 2015, when I was no longer a teenager, I couldn't afford better than a Kazam Trooper: I invite you to google this name, you will find photos of a smartphone no less "ugly" than the one presented by Recraft
Recraft makes odd-looking head and face, look at the Yoga-poses. I bought access to the model, and I find it does this a lot for my generations. Also, when asking it to do fantasy/fictional, it tends to often look like cosplay instead of FLUX which makes it looks stylized. YMMV.
Must note that most of the teens in that era (especially the hairstyle) did look like the output of recraft. The teens in other 2 images look like today's teens, or young adults. Even the phone looks like its actually from 2014-15.
Recraft has the most pleasing results, regarding small details it messes up , such as shape of the smartphone one can simply give a little more detailed prompt about the phone he wants to see
try packing it, i just got an 'aerial /overhead/ from above' shot... Keep dropping these descriptions into the prompt, dont write proper english, write like a child. - overhead shot of a man in a hallway, i can see the top of his head from this overhead shot, hes walking forward one step at a time as seen in this angle from above . mention things that you would see from above - i can see the persons top of their head,
@@Betonbent try add the words, - symmetry, well balanced, perfectly aligned, and eyes looking straight forward. staring directly at you i sometimes use mugshot or passport photo. If you ever stuck try find artists or photographers who specialise in that pose... wes anderson? Stanley Kubrick? david bailey. The AI might be confused with "camera"? Where is the camera? I'd try, - I am holding a camera, i look thru the lense and looking back is... a woman, head on, perfectly aligned shot
How about jenna ortega makes a handstand on a tree stump taken with a alpha 7 on a rainy day in her garden? Or a sword master cuts a apple falling down with his long sword while using a skateboard? Or a young german man with a body like bruce lee trains with nunchacs while wearing a suite from loose cotton?
From the perspective of the dog and the cat, the dog is on the right, by the viewer yes it's wrong, if you said from the perspective of the viewer it might do it correctly. But it's not 100% wrong.😉
Can the recraft generate an image like "a man is sitting in an aeroplane seat; the seat is very lowered; the woman sitting behind him is very annoyed"? I mean more precise images, not just very, very general.
09:36 You asked for a 2015 snapchat mirror selfie. In 2015 phones didn't have clusters of multiple cameras like in the Ideogram image (look at IPhone 6S Plus or Samsung Galaxy Note 5 as references for top 2015 phones) so this image is a fail. I would say Recraft actually got this one most period correct. Flux is somewhere in the middle, as Flux shows a poorly defined multiple lens cluster on the camera.
5:59 Flux Pro 1.1 ! I prefer much more the Flux Pro 1.1 image... The two others have big flaws 8:09 Recraft v3 and Flux Pro 1.1 ! Both are excellent (the ideogram image seems lacking of precision) 9:30 Recraft v3 ! In 2015, the smartphones were closer to the Recraft v3 version... Flux Pro 1.1 display a poor background and the Smartphones weren't like this in 2015... Ideogram v2 displays a poor background as well, even if it is a more detailed one than the Flux Pro 1.1 11:18 Flux Pro 1.1 ! You did not notice the problem with the double finger on Flux Pro 1.1, though this is my favourite image (very precise and realistic) as Recraft v3 lacks of precision (the face is blured) and Recraft v3 and Ideogram v2 lack of aesthetics 12:59 Ideogram v2 ! Not really Ideogram v2, except for the prompt adherence... Flux Pro 1.1 forgot Will smith as we don't find him anywhere, Recraft v3 displays characters that we can recognize but they're still not like the persons in the real life 16:12 Recraft v3, Flux Pro 1.1 and Ideogram v2 ! (all are ok... You did not precise in your prompt if you wanted two hands from two different persons or two hands from one, so Flux Pro 1.1 did not make a mistake, and neither the two others did) 19:19 Ideogram v2 ! I did not have the merge you had with Flux Pro 1.1, and I used the exact same prompt you used (exactly, with the comma, the points, word for word, scrupulously identical in all respects), but I did not get the soles either... I tried several times I never got that merge but I never got the soles either, I only got the palms ... They were perfect, but that was not the prompt, so there is still a problem in the prompt adherence Recraft v3 lack of realism in the soles 20:56 Flux Pro 1.1 and Ideogram ! Aesthetically, I would choose without any hesitation Flux Pro 1.1, but there is here again a prompt adherence issue (-). On that point (and not the aesthetical one), Ideogram v2 is better, the worst is Recraft v3 22:29 Flux Pro 1.1 ! Flux Pro 1.1 and Ideogram v2 are the most realistic ones but Ideogram v2 has a prompt adherence issue (your prompt said "a red sphere", and Ideogram v2 displayed two spheres one on the other, one blue and the second one red... Definitely not Recraft v3 : lack of realism (see the cat head) 24:07 Ideogram v2 ! Recraft v3, Flux Pro 1.1 and Ideogram v2 are aesthetically good and thehe three respect the prompt if we take quickly a look, but if with a more careful watch, recraft v3 dispay an unrealistic helmet, Flux Pro 1.1 seems not to have seen a snail in a long time (the snail's head is not realistic at all!), the best of the three for that image is then Ideogram v2 25:14 Flux Pro 1.1 ! If the three models display a whale, only two of them (Recraft v3 and Flux Pro 1.1 display a whale in the sky, and only one of the three (Flux Pro 1.1) display a watercolour painting of a whale in the sky 26:39 Flux Pro 1.1 ! Recraft v3 completely messed up the details and the aesthetics and Ideogram v2 messed up the aesthetics 27:43 Flux Pro 1.1 and Ideogram v2 The three models (Recraft v3, Flux Pro 1.1 and Videogram v2 matched a very aesthetical level, but Recraft v3 cruelly lacks, of precision, of details, on the head of the animal SCORES : Recraft v3 => 3 Flux Pro 1.1 => 9 Ideogram v2 => 6 The winner is... Flux Pro 1.1 !!!!!
prob just confuse the AI. You'd end up with cats eating a deer, but it would look more like a nature shot. If you found some real gory movies or directors and placed them in the prompt you would get better visuals
would be nice to see more good ai generators that can create animestyle. Or maybe some ai tool that can colorize manga and comics or uncensor. I am asking for a friend.
Generating people is better on mid journey, because Recraft does not do eyes very well and the random person generated is often the same looking person.
Yoga pose goes to Flux by a mile. You’re reaching when you say recraft was better. Recraft wins the phone shot. Recraft loses tea with the queen. Loses the web landing page. A toss up on heart hands. Loses the snail challenge. So how is it “the best’?
that's what i'm hoping to convey in this video: there's no clear winner. it really depends on the content. recraft is only 'the best' according to the leaderboard
I don't care how well AI can do images if I can't do nsfw with it. As far as I'm concerned stable diffusion is still kind if you know how to prompt. I just need better coherence.
8:58 "This doesn't look like a real phone" Come one. It looks like Nexus 5. 9:07 looks like iPhone 6. You should take a look at actual phones from 2015. 9:37 That is "legit looking phone" from the future.
The flux phone looks closer to a phone you would see in 2015. The ideogram photo looks like a modern phone, and the image quality looks like a modern photo, just with an older 2015 style Instagram filter on it. The flux image actually looks like the quality phone photos were taking back in 2015
Wonder how it's even no 1 on the leaderboard seems most of these AI are curating/cherry picking best results to beat the benchmarks yet the performance is average
Are you for real? Did you even live here in 2015 to experience what cheap phones looked like back then? Ideogram got the worst image there cause it's an iPhone. The rest are Android phones... and is thus more prompt coherent.
For me recraft is total fail it try to use buttons and settings. what my envision is buttons, slide bar, settings, etc must be replaced by chat like interface or even natural human AI Conversation interface
Thanks to Mureka for sponsoring this video, a versatile AI platform to create music! bit.ly/4ehrdaq
👋 hi
ama-rei-ka home of the hamburger
“This might be the best AI generator” everytime you say that for a new video I’m going to do a shot.
clickbaiting at its finest
💯
*What type of shot are you talking about* 😯
@@DjHazardous Backshot
haha yes, and I'm surprised it's not "insane" and "game changing"
You did specify 2015, so before you say the phone looks weird or unrealistic, look up how phones looked like in 2015.
There was no such thing as multiple lenses, and most of them had a simple small lense near the top in the middle.
i thought the same thing. Either this dude is younger than 20 or this video was created by AI 🤣😂
Ideogram and Flux smartphones make no sense for 2015, I think that recraft is actually more accurate there, it does look like an old phone.
Also you missed the fact that you can create your own image styles.
Exactly. That ideogram iPhone must have time traveled. Recraft got it right.
up
I think flux makes sense other than the multi-lens camera. The actual shape and look of the case looks exactly like what you'd see back in 2015. And the image quality looks way more like what my old 2015 photos looked like.
Exactly. I've used recraft a lot, for anime if he just screenshot an anime to use as a custom style it definitely would have generated something better.
Nice vid ! Flux with loras still wining the war so far
True. I prefer Flux over any other text-to-image model so far.
@@DmitryRomashovit's open source and does not censor images.
i still think from an artistic and cinematic viewpoint is MJ, i just had a go at this and it made boring pics, i came back to MJ and my god the camera shots, the light, the coloring was off the chart. MJ weakness is prompt adherence , but it should be top of that leaderboard.
@armondtanz it's cuz it's not open source. No one talks about MJ. It's not bad but not worth the price.
I hate when the 'Public' option is mandatory!
Thank you buddy for your effort :)
True. They should remove the free option.
I got baited by feet
Same, he’s leveraging foot fetish to get clicks.
Or maybe he’s a foot fetishist himself.
0:11This is impressive at a glance, but the generator doesn't seem to understand clothing. The green girl's left arm is very baggy compared to right. The brown girls' left and right sleeves are mismatch. The pink guy's right hand is flipped. The brown girl's left hand is in an extremely unnatural position. The list goes on. All 3 are also floating in space, with no consideration for how they jumped. With that said though, I'm still impressed by how far the technology improved compared to 6 months ago.
Ideogram looking the best to me
I think Recraft is probably a great tool for editing existing images using a reference image, future contender as a free replacement for photoshop
I chose the far right side selfie due to the single camera on the phone... even with the hallucinated "center cam"... a flip phone in the prompt could have killed it! ;)
14:46 Modern web apps MUST adjust for the mobile phone screens. As 95% of the users will use them from the smartphone. So showing one on mobile does not violate the condition. If we go from the statement of training, even professional UI designers sometimes could not tell if the UI is native or a web app (once displayed on mobile), so when image is tagged "web app" it can be either.
"if you wanted to swap this background in photoshop you're going to have a really painful time (...)"
nope...
you click on the subject to AI-select it, then you inverse the selection and you click "generate" and you change it to whatever background you need. easy as 1-2-3
It’s still asking a lot for a model to produce flawless output for any prompt on the first take. Most people using AI to generate imagery for real-world applications are using ComfyUI or their own custom coded pipelines to move images through a workflow that includes refinement steps for hands and faces. My own sense is that we’re getting close to having very large models that cannot possibly be hosted on consumer GPU hardware that can automate everything and produce great results. Google’s Imagen 3 model is clearly a gigantic model and produces very consistent results, for instance. But moving images through a few workflow steps with small models runnable on consumer hardware produces exceptional results already. I’m not sure which approach will win out in the end, but likely the big cloud models will eventually take over.
I agree. I also am surprised at what many people think are a "realistic" looking image as well. I have spent a lot of time on various AI generators as an experienced photographer/retoucher and I know that to get real looking results is a refinement - almost an art of using the generator. For example- Mid journey mayn't do so well with a single line prompts, but none of them do very good IMO. But if you work on styles, refine the prompts with good references, and use the newer edit features for fine control - the results are quite amazing. We should be judging these on the absolute best results they can do- not the laziest results.
To me the Recraft image of the young man with the phone was probably the most accurate for a 2015 phone
Am I blind or where are the links in the description to these topics??? Wtf!
there ain't i've been a subscriber of his and there isn't
I'm not sure I've seen him take these simple criticisms into account in future videos. He reports on a lot of stuff, but he leaves in enough small problems that I think it heavily limits his growth.
man Im hoping someone does a open source and UNCENSORED tired of censorship garbage... nothing good today I'll wait next year..
the dog in recraft IS actually on the right as prompted (and the cat IS on the left), if you consider their position from their POV :)
🤯
Imagen 3 does feet and hands THE BEST by far! I dont know what fhey did, but everything pales into conparison with Imagen 3 with human anatomy
9:49 Recrafts image is the only one where the phone really looks like a 2015 phone and not a modern iphone
I have to say, you get the best results at 4MP or higher which they do state. So if you do less than you get bad results. You did do most of the generations at 1024. Can't say if it was a fair assessment. @25:00 If you would have picked the style from the list you may have gotten better results, however it wouldn't have been using their best model. I think for a more fair assessment, you'd have to render them all at close to the same pixel rate. There was also a whole section of 20B styles styles at recraft that you missed. I don't know why they don't automatically put all the styles in the list without selecting the combo box first. Great job on showing how to do the layers. I kept only getting recraft option and was getting annoyed with it. I didn't realize that you had to merge down the layers. Overall great job!
Why is AI so fast it’s faster than a trend
In 2015, when I was no longer a teenager, I couldn't afford better than a Kazam Trooper: I invite you to google this name, you will find photos of a smartphone no less "ugly" than the one presented by Recraft
Recraft makes odd-looking head and face, look at the Yoga-poses. I bought access to the model, and I find it does this a lot for my generations. Also, when asking it to do fantasy/fictional, it tends to often look like cosplay instead of FLUX which makes it looks stylized. YMMV.
Must note that most of the teens in that era (especially the hairstyle) did look like the output of recraft. The teens in other 2 images look like today's teens, or young adults. Even the phone looks like its actually from 2014-15.
I'll still stick with SD1.5 with controlnet because of the NSFW 😈😈😈
😃😃😃 sdxl is also pretty good
SD3 is gonna be good soon...
For the Snapchat image, recraft did great. Remember, the iPhone 15 was not released in 2015!
Some models intentionally do not include celebrities or intellectual property from movie studios.
Recraft has the most pleasing results, regarding small details it messes up , such as shape of the smartphone one can simply give a little more detailed prompt about the phone he wants to see
15:45 oh my goodness!! What is this..!!??🤣🤣🤣 I feel u bro
One thing I noticed is that Recraft can't really generate front view shots. Anyone else noticed that?
yes thats my experience as well
try packing it, i just got an 'aerial /overhead/ from above' shot... Keep dropping these descriptions into the prompt,
dont write proper english, write like a child. - overhead shot of a man in a hallway, i can see the top of his head from this overhead shot, hes walking forward one step at a time as seen in this angle from above .
mention things that you would see from above - i can see the persons top of their head,
@@armondtanz i tried frontal view of a woman looking directly into the camera, and i still get this odd side pose shot lol, still has a long way to go
@@Betonbent
try add the words, - symmetry, well balanced, perfectly aligned, and eyes looking straight forward. staring directly at you
i sometimes use mugshot or passport photo.
If you ever stuck try find artists or photographers who specialise in that pose... wes anderson? Stanley Kubrick? david bailey.
The AI might be confused with "camera"? Where is the camera? I'd try, - I am holding a camera, i look thru the lense and looking back is... a woman, head on, perfectly aligned shot
Incredible review like always!
Thanks!
Ideogram definitely looks the best to me
what i learn... ideogram v2 is a beast o.0 , I almost cant tell the difference from real photo
How about jenna ortega makes a handstand on a tree stump taken with a alpha 7 on a rainy day in her garden? Or a sword master cuts a apple falling down with his long sword while using a skateboard? Or a young german man with a body like bruce lee trains with nunchacs while wearing a suite from loose cotton?
9:33 A phone camera from 2015 wouldn’t look like that.
Testing text in ai creation is just bonkers, it is just the most easy to change per hand with PS or his clones.
flux all the way . impressive
In 2015 there were no phones with cameras like that
ideogram looks better
Ranked #1 by whom???? Clickbait
From the perspective of the dog and the cat, the dog is on the right, by the viewer yes it's wrong, if you said from the perspective of the viewer it might do it correctly. But it's not 100% wrong.😉
So Ideogram is the best!!! 😅
Can the recraft generate an image like "a man is sitting in an aeroplane seat; the seat is very lowered; the woman sitting behind him is very annoyed"? I mean more precise images, not just very, very general.
no, it only generates image with weird man and woman faces
@@Rilex037 Thanks.
Not the dreaded feet pick test 😭😭😭
Does this man not know what a phone looked like in 2015?
09:36 You asked for a 2015 snapchat mirror selfie. In 2015 phones didn't have clusters of multiple cameras like in the Ideogram image (look at IPhone 6S Plus or Samsung Galaxy Note 5 as references for top 2015 phones) so this image is a fail. I would say Recraft actually got this one most period correct. Flux is somewhere in the middle, as Flux shows a poorly defined multiple lens cluster on the camera.
Unrealistic llama sunglasses, they need to be at more of an angle as llamas have a pointy face.
We know that Flux can produce more realistic images with finetuning the settings. I am surprised that by default it produces these "plastic" images.
29:18 jackie legs!!!!
NOT THE FEET IN THE THUMBNAIL BRO 😭😭😭😭
Whats the best for anime style AI image generation? I used to use SD last year
5:59 Flux Pro 1.1 !
I prefer much more the Flux Pro 1.1 image... The two others have big flaws
8:09 Recraft v3 and Flux Pro 1.1 !
Both are excellent (the ideogram image seems lacking of precision)
9:30 Recraft v3 !
In 2015, the smartphones were closer to the Recraft v3 version... Flux Pro 1.1 display a poor background and the Smartphones weren't like this in 2015... Ideogram v2 displays a poor background as well, even if it is a more detailed one than the Flux Pro 1.1
11:18 Flux Pro 1.1 !
You did not notice the problem with the double finger on Flux Pro 1.1, though this is my favourite image (very precise and realistic) as Recraft v3 lacks of precision (the face is blured) and Recraft v3 and Ideogram v2 lack of aesthetics
12:59 Ideogram v2 !
Not really Ideogram v2, except for the prompt adherence... Flux Pro 1.1 forgot Will smith as we don't find him anywhere, Recraft v3 displays characters that we can recognize but they're still not like the persons in the real life
16:12 Recraft v3, Flux Pro 1.1 and Ideogram v2 !
(all are ok... You did not precise in your prompt if you wanted two hands from two different persons or two hands from one, so Flux Pro 1.1 did not make a mistake, and neither the two others did)
19:19 Ideogram v2 !
I did not have the merge you had with Flux Pro 1.1, and I used the exact same prompt you used (exactly, with the comma, the points, word for word, scrupulously identical in all respects), but I did not get the soles either... I tried several times I never got that merge but I never got the soles either, I only got the palms ... They were perfect, but that was not the prompt, so there is still a problem in the prompt adherence
Recraft v3 lack of realism in the soles
20:56 Flux Pro 1.1 and Ideogram !
Aesthetically, I would choose without any hesitation Flux Pro 1.1, but there is here again a prompt adherence issue (-). On that point (and not the aesthetical one), Ideogram v2 is better, the worst is Recraft v3
22:29 Flux Pro 1.1 !
Flux Pro 1.1 and Ideogram v2 are the most realistic ones but Ideogram v2 has a prompt adherence issue (your prompt said "a red sphere", and Ideogram v2 displayed two spheres one on the other, one blue and the second one red... Definitely not Recraft v3 : lack of realism (see the cat head)
24:07 Ideogram v2 !
Recraft v3, Flux Pro 1.1 and Ideogram v2 are aesthetically good and thehe three respect the prompt if we take quickly a look, but if with a more careful watch, recraft v3 dispay an unrealistic helmet, Flux Pro 1.1 seems not to have seen a snail in a long time (the snail's head is not realistic at all!), the best of the three for that image is then Ideogram v2
25:14 Flux Pro 1.1 !
If the three models display a whale, only two of them (Recraft v3 and Flux Pro 1.1 display a whale in the sky, and only one of the three (Flux Pro 1.1) display a watercolour painting of a whale in the sky
26:39 Flux Pro 1.1 !
Recraft v3 completely messed up the details and the aesthetics and Ideogram v2 messed up the aesthetics
27:43 Flux Pro 1.1 and Ideogram v2
The three models (Recraft v3, Flux Pro 1.1 and Videogram v2 matched a very aesthetical level, but Recraft v3 cruelly lacks, of precision, of details, on the head of the animal
SCORES :
Recraft v3 => 3
Flux Pro 1.1 => 9
Ideogram v2 => 6
The winner is... Flux Pro 1.1 !!!!!
How come people who make these A.I. generator videos never do weird prompts like "starving stray feral cats eat roadkill maggot infested deer carcass"
i don't think they wanna see that
prob just confuse the AI. You'd end up with cats eating a deer, but it would look more like a nature shot. If you found some real gory movies or directors and placed them in the prompt you would get better visuals
RUclips TOS
Also the AI companies' TOS.
Lots of ai companies have TOS so strict that you couldn't even recreate most movies.
would be nice to see more good ai generators that can create animestyle. Or maybe some ai tool that can colorize manga and comics or uncensor. I am asking for a friend.
Generating people is better on mid journey, because Recraft does not do eyes very well and the random person generated is often the same looking person.
Yoga pose goes to Flux by a mile. You’re reaching when you say recraft was better. Recraft wins the phone shot. Recraft loses tea with the queen. Loses the web landing page. A toss up on heart hands. Loses the snail challenge. So how is it “the best’?
that's what i'm hoping to convey in this video: there's no clear winner. it really depends on the content. recraft is only 'the best' according to the leaderboard
flux with loras and controlnets smokes all of them
Cant wait to 3d print my future AI wifey 🤑
9:30 i think recraft generated old phone as compared to quality of picture ,
It not created apple device with 3 cameras
That iphone is definitely not something from 2015 🤣
I don't care how well AI can do images if I can't do nsfw with it. As far as I'm concerned stable diffusion is still kind if you know how to prompt.
I just need better coherence.
"no no no, you're better than that, watch you profanity please" ... no profanity was used. stable diffusion still king.
Which AI image generator is the best at creating hands and feet?
8:58 "This doesn't look like a real phone" Come one. It looks like Nexus 5. 9:07 looks like iPhone 6. You should take a look at actual phones from 2015. 9:37 That is "legit looking phone" from the future.
PonyXL >>>
honestly im sick of closed-source garbage. It's like selling umbrellas to passing cars because it's raining. Things are moving to fast for this.
I’m surprised that people still pay for image generation. Unless they have low end PC I understand. I’m happy with SDXL.
Would you try to make a tri-plane with an ai-image-generator. That something where they have a lot of trouble with.
22:10 incorrect. That's a pyramid.
HE CAN'T KEEP GETTING AWAY WITH THIS
But recraft exist for a long time now
Sir , why ai does not sleep ?
I just ran all of these peompts through Imagen 3 and it beat most of these tests over all these other models. Why arent you testing that instead?
yes imagen is very underrated. i only did a video on that w the same prompts ruclips.net/video/3t91qWofuuk/видео.html
My reaction on 8:52 having the exact same phone while he calls it ugly and not looking real 😞
The flux phone looks closer to a phone you would see in 2015. The ideogram photo looks like a modern phone, and the image quality looks like a modern photo, just with an older 2015 style Instagram filter on it.
The flux image actually looks like the quality phone photos were taking back in 2015
26:12 🤨
How..cool..is..THAT?
Wonder how it's even no 1 on the leaderboard seems most of these AI are curating/cherry picking best results to beat the benchmarks yet the performance is average
Unable to access. Ad blockers and vpn are off, keep getting an aw snap page
Amazing!
i seen this exact video before on another channel
but i need offline solutions. online ones are way costly to try and learn. :(
F E E T
2.7 on the App Store, yeah no thanks
Actually, Imagen 3 is the best overall image generator, and nobody is using it.
ay you post this in all of these videos i tried using it and they require you to pay to even make humans lol that's wild.
Lol ikr! Imagen 3 is garbage! @@brockhampton3078
Ai Jesus Facebook accounts: 🤑🤑🤑🤑
love it
i like ideogram more
Are you for real? Did you even live here in 2015 to experience what cheap phones looked like back then? Ideogram got the worst image there cause it's an iPhone. The rest are Android phones... and is thus more prompt coherent.
I tried it and I'm very much not impressed
Recraft gets the award for worst user interface at least.
it's the best
this video is more like a bot made sponsor video, If you can show 2 pics from a new thing why not others? it's purely misleading
Is it censored?
If it's not open source, I don't give a...
But it's not open-source so I won't use it.
Still looks bad. We are not there yet.
For me recraft is total fail it try to use buttons and settings. what my envision is buttons, slide bar, settings, etc must be replaced by chat like interface or even natural human AI Conversation interface
Bro why you always put feet on the thumbnail 😭