What do you guys think? Nice new addition or too little too late?

@CriminalSlang305

2 ай бұрын

It's a nice feature but I would have preferred if they had improved the quality of image generation. Adding an upscaler would have been more beneficial to me.

@LouisGedo

2 ай бұрын

👋 Can you please tell us who won the nVidia GPU? I get literally hundreds of emails per day and I don't even know which day you sent the email notice

@davidbangsdemocracy5455

2 ай бұрын

Too late for what?

@NotThatVinny

2 ай бұрын

​@@davidbangsdemocracy5455There's been a number of image generators with inpainting. Considering their level of quality, it doesn't seem compelling to switch over just for this feature alone.

@AnthonyGoodley

2 ай бұрын

Hopefully they bring back outpainting now that we have inpainting once again. If only we could use the inpainting effectively to correct text. That's a big fail imo.

To tell the truth I expected OpenAI's inpainting to be much better. Failed both at artistic style and text.

@AnthonyGoodley

2 ай бұрын

Yes it really isn't all that good considering this is coming from OpenAI. The ability to fix screwed up text would have been a massive leap forward. Instead they dropped the ball. They can do better.

@kecksbelit3300

2 ай бұрын

Ty

@IntentStore

2 ай бұрын

I’m really curious because the inference time is extremely long for being on H100s so the model must be very large (50B+ parameters) yet still doesn’t perform great. So they may have some architectural work to do.

@AnthonyGoodley

2 ай бұрын

I was just working on a logo where it misspelled a unique username. I highlighted the bad text and the only thing I typed in was the correct version of the text. Dall-E 3 was able to fix the text first attempt. So it can edit text sometimes at least. I was too harsh with my earlier comment. I'm not sure how often it can do this as I haven't used it enough to get a good idea yet.

Dall-E isn't generally my generator of choice but the inpainting is very high quality and the ability to consistently create new elements and remove others is a pretty big deal. Inpainting is generally only consistent at modifying what is there and saying "no butterfly" or "remove butterfly" is asking the AI to not think of a pink elephant.

@IceMetalPunk

2 ай бұрын

I think that probably has to do with it going through GPT first. The prompt that gets to the inpainting model probably doesn't say "remove the butterfly", but something more like "blue sky", because GPT is likely smart enough to know to use a positive prompt.

@brexitgreens

2 ай бұрын

​@@IceMetalPunk Good user image. (As of April 2024.)

@IceMetalPunk

2 ай бұрын

@@brexitgreensThanks. It was pulled from a Stable Diffusion generation.

@TonyRagu_FromAvenueU

Ай бұрын

how do you use it via bing/create? for inpainting

They should add a feature so you can just add the text yourself a little text icon and highlighted, whatever color and type it out and a 2nd button to have the A.I. to stylise it for you

@apache937

2 ай бұрын

That would really work fine for most use cases

Matt as always you are up to speed man :) I really like watching your videos since they are interesting, positive and always valuable!

*Nice video Matt, keep it up!* ❤

Great video, looking good as usual ❤

AMAIIIZIIIING! it was really time, haha!! This was something I missed all the time in DallE-3! Just spotted it some minutes ago, it´s super fresh =D

@avi7278

2 ай бұрын

Wow you are easily impressed

@mirek190

2 ай бұрын

..really ...under offline DSXL + fooocus or a111 inpainting + control net works better

@rossobosso

2 ай бұрын

@@avi7278 so? I love DallE-3 and inpainting was just what I needed! Never worked with stable diffusion, muuuch to complicated and don´t have the hardware for it, working with Midjourney mostly, as it´s just the best image generator out there. And DallE has a wonderful style, MJ can´t offer. So why use SXl and stuff like that. And what´s the problem to be impressed? I don´t get it :D

BTW, thanks for sharing about Ideogram in a previous video and mentioning it in this one. I've been using it a lot. I use it to prototype a complex image and then run that image through Leonardo to make humans look more realistic/photographic. If you have other suggestions, please share!

I did the same thing. Adobe Express? 😳 Checking it out it just generates pre-made templates and takes you to the site to do any further editing. Pretty hokey.

I was using it this morning and was able to correct text no problem

i still enjoy how u explore the product, great vid, gg!

It’s definitely an advancement, with more updates to come that can only lead to more magical ways to create and edit content. Very exciting.

I used Dall-E recently, and I think this update was bigger than we initially thought. I tested the model and noticed that my generations were of much better coherency and overall quality than before, especially with long detailed prompts. I suspected the internal version might have changed from version 3 and tested it this way: In Dall-E GPT, I asked this: "An image of a book that read this is dall-e x! X is the version dall-e is using" And it gave me two images, on one was "This is dall-e 4". I was surprised so I tried again and got the same result. For some result, it seems to believe it's version 4. I don't know what to think, maybe it's a 3.5 version? Anyway, I managed to get some really amazing creations that I don't believe DALL-E 3 was capable before. It's not a complete revolution, true, but a subtantial improvement.

Totally not on topic, but love that built-on pop screen on your mic. :D

I just stumbled across this yesterday

i loved the AI music too

If you think of Dall-E as being inside the computer looking out at you, it did get the peach and lemon on the correct sides -- from its perspective! 😁

I think they are releasing this to try to keep up with Midjourney and Ideogram before launching Dall-E 4 about a year after Dall-E 3's release.

0:51: Nice touch, and a nice generation as well.

That's cool, Pinokio installed on my Pi5, is it useful? No idea but anything that makes it easier to try AI is cool in my book.

Hmm, strange... for me, doing those image edits doesn't change anything in the resulting image - it just generates the same image again, as though I hadn't selected any area. Anyone else having this issue?

You have to cool feature we started using it I think on Friday or Saturday so it's been out for a while but they just didn't mention it until now LOL

Stable-diffusion can do all of it

@brexitgreens

2 ай бұрын

With somewhat lower quality and sophistication.

@GamersBlogX

2 ай бұрын

Yeah, but it does require more effort generally, especially with text. Maybe whenever SD3 drops that will change though.

no way i clicked on this with no views

@CatfoodChronicles6737

2 ай бұрын

Probably just because it hasn’t loaded yet

Help refresh my memory community: What can you use to change the perspective of 2D portrait not previously Ai generated. I have a front on sketch I would like to see in 3/4 profile. I am sure I heard about it somewhere

hey Matt what's the name of the software that allows you to run stable diffusion locally on your own pc?

@starblaiz1986

2 ай бұрын

That's pinokio (with a K). If you Google "Pinokio AI" it will be the top site.

I have the problem with Dall-e 3 in Chat GPT, that nearly all of my "vertical" images i want to create are just widescreen images that are rotated to be vertically. mostly when i try to create "paintings" like oil painting do you have a fix for that?

@goyashy

2 ай бұрын

There's a custom GPT version of DALLE that lets you specify size

You, dive into it!😂

So this feature was live for a day and now its not there...

Is this in open ai gpt or just pilot

how do i use bing create dall-e 3 to inpaint/edit

Not impressed seeing that we all are waiting for GPT 5 to drop. They should just stop playing around, and give it to us already.

@MattVidPro

2 ай бұрын

feel tht

@brexitgreens

2 ай бұрын

DARPA (or was it some other U.S. military research agency) has just revealed in their AI review that OpenAI has not even begun training GPT-5 due to the lack of necessary hardware.

@9educt

2 ай бұрын

@@brexitgreens seems unlikely when Anthropic are ahead on some aspects

Nice !

It doesn't seem to work on mobile or with custom gpts 😔

Dripped out lemon at 9:59

Is anyone else able to get chatGPT GPT3.5 to generate images or is it jist me..?

I don't see any option to edit existing images with this which seems like a big limitation

@vainezaiven6677

2 ай бұрын

It's not possible. That was actually the first thing I tried (uploading an existing image and editing it), and the results were incredibly wild. The AI itself kept trying to tell me it could do it, but it clearly can't. It just makes up something.

@kullenberg

2 ай бұрын

I need Dalle3s prompt coherence with img2img and controlnet@@vainezaiven6677

why my successful rate is much lower, is it that my original prompt before inpainting is too long and complex?

Do you mean CHAT GPT-4 Matt?

I can't edit my own picture, are you able to alter your own picture ?

@kecksbelit3300

2 ай бұрын

I don't think cause that would come with copyright issues

So for at this point Designer don't have any real special features with Dalle 3, But if you create an image in Bing chat is gives you options like styles and landscape format. Witch seems dumb and should be the other way around.

@CanadaBlue85

2 ай бұрын

Those features are now in GPT too as of today.

@reifuTD

2 ай бұрын

@@CanadaBlue85 Too bad the Inpainting and the vertical phone format isn't in Bing chat though, it's weird that they release features one different sites in orders and not the same order every time .

@kullenberg

2 ай бұрын

I don't have those options in Bing chat. I guess they're region locked?

I believe I'm able to do this within Canva pro

The lemon who was wearing Nikes, had two left feet. 😂

Sick……

They most likely removed the signup wall because this would allow them to get more conversational data

How come those AI sites never have a payment option where you DON'T need a credit card?? I would totally get plus for this...

@apache937

2 ай бұрын

What then? paypal?

@markcorrigan9815

2 ай бұрын

@@apache937 Yeah for example

@jasonstatham483

Ай бұрын

@@apache937 cache of course

The final in painting Shih Tzu looked like South park construction paper

Yes, I do WAY better just asking for a completely new image with the changes, like I say that was great how about we make the dog a wizard? and bingo it does it. The only use I have for this is to remove something, but even removing things is not always perfect.

Pika Labs you can do the same thing Matt

Traditional image editing is almost becoming a thing of the past.

I found NSFW image using Calle3, how to do that?

Hmm, Adobe Express Featured + editing reminds of Adobe in-painting + text is a bit hard to get right = hmm.

all I get is "error creating image" POS program.

and honestly: too late, too early, I just don´t car. Not perfect, not this, not that.. nothing in the ai world is still perfect, people have much too high expectations. I´m super happy about every new feature, that´s it. So for me it´s not too late or whatever, it´s just supercool and I´m fine with that =D

CAN U DO THE STABLE AUDIO V2

@MattVidPro

2 ай бұрын

is it good?

I thought they were going to the fix the very poorly (hilariously) spelled banner behind the poodle. "Hepppphyyy"

I just tried it and the results sucked. Not impressive imo.

9:00 It sucks that the AI can’t fix the text that it spelled incorrectly.

Finally, even if it's mediocre... There are a few other insane things they could add which SD has for more than a year now, that DALL-E 3 should had for a couple of months already.

Only in paid version of Chat GPT I guess?

wow, this takes so much longer, than anything i do offline with my 3060...

WOO-HOO! So now anyone can draw. Who needs artists? YAHOOO-HOOOO!

@OnigoroshiZero

2 ай бұрын

I've stopped paying artists for more than 10 months now, for both work and hobbies (character, monster, and location art for a TTRPG campaign I run). I was going to hold on them for longer, but with the stupid meltdown they had when these models first came out (with their bs arguments about theft and copyrights), it was a good chance to rid myself of them much earlier, it did benefit me massively in multiple aspects (at least $5k, dozens of hours communicating with them, and better quality on-demand).

Could add bows in Canva in half the time 😂

What are you talking about? Stable Diffusion can do VERY consistent inpainting.... OpenAI is obviously playing another game, by their own statement. They are trying to build AGI, and they are doing that by building many different models which can do different aspects of intelligence, then tying them together to have one more seamless experience which delivers something approximating general intelligence. Imo it's not actual AGI, but it will be able to do many of the same things, limited by the narrower aspects of each model. If it quacks like a duck...

It is very mediocre for now tho...

You really do not need music !!!

@HardKillaz

2 ай бұрын

Especially if he's going to be talking over it the whole time.

@Bizzybugproductions

2 ай бұрын

😅

Adobe will fear!😂

I think they made it even more censored after adding this feature, which is sad.

at least OAI is not cutting the video and showing us realtime now.......

I'm kinda surprised OpenAI included an inpainting option for DALL-E 3. Inpainting would seem to go against OpenAI's whole ethos regarding AI. They think the AI should be smart enough to give you, the user, what you want, without needing things like "editing" or "inpainting." So I was legitimately not expecting them to ever add inpainting to DALL-E 3.

No, no no no, just tried this, asked it to remove a stray arm in a photo of me and my wife. There's someone to the side and there arm is sticking into the pic. It created a totally different image of a couple similar to me and my wife. That's not "image editing" at all. Edit: I'm presuming my account doesn't have it yet. I can only go to the GPT chat thing.

@kecksbelit3300

2 ай бұрын

I don't think you can edit not generated images cause they would need to check copyright and more.

@bigglyguy8429

2 ай бұрын

@@kecksbelit3300 Yeah I guess, but seems right now I don't have any access to this Dall-E 3 thing. I have the paid version but there's no button or anything. GPT just says it will do it for me but it really messes up. It did try the wifey photo but it literally just took a section of the photo and pasted it over the arm - and the section was part of my wife's face! So literally just bizarre.

@kecksbelit3300

2 ай бұрын

@@bigglyguy8429 just let them generate a image for you click on the image and you should see if you have the button to edit

The edits look like they were poorly photoshopped. They don’t blend in naturally with the rest of the image.

Eat Bacon every day. Bacon is King. 🤴 🥓

Can you do Pi piz

Im not impressed so far. Ive tried it a few times and it didn't give me anything workable

Better - but not redeeming enough. Chat gpt rejects so many benign prompts…. “A karate man facing off against a gorilla”. - was deemed potentially offensive and unsafe. Wow…. Stick with ideogram for 99% of generations needed.

"Because i haven't seen another one be able to do very consistent art styles in their inpainting or editing" well coming from someone that stopped talking about SD and just promotes paid products it doesn't really surprise me.

@MattVidPro

2 ай бұрын

I don't dive too deep into the weeds of SD it's not really my thing... I literally promoted an open source alternative in this video lmfao. And that quote was in reference to CHAT based inpainting

@Hasblock

2 ай бұрын

I generally don’t like paid promotions and I personally use stable diffusion but there really isn’t any amazing open source inpainting or outpainting options. All the open source painting options are extremely rudimentary compared to adobe firefly, which is the king right now in my book. Probably the best open source option right now is Krita with Acly’s inpainting plugin.

@mycelia_ow

2 ай бұрын

@@MattVidPro Well said

@AlbertCunninham

2 ай бұрын

@@MattVidPro While I do think you forgot about SD and it's a shame because a lot of the stuff you promote SD has been able to do for awhile, but you're right and I was kind of rude about it, so sorry about that man.

It is slow as hell.

why the hell you keep chatting with it like its a human? You prompt it and not chat with it, when you type "no text at all" as first words its obvious that it will remove your text no matter what you write later.Also dalle3 is way way ahead of any other image generators, but it was heavily nerfed as we all know.

Dall-E 3 is washed, tell us about Stable Diffusion 3

u darker then th ai, get some sunlight (promt)

It's really nice to see a completely useless image editor in dalle3, can't wait to never use it again.

it does seem like OpenAI are falling behind, Sora gens must have taken ages and definitely cherry picked, this inpainting can't even do text... I mean, i'm not impressed

too late to be impressive. midjourney has had this for months

after a whole year, DALL-E 3 is still trash when compared to midjourney and stable diffusion. what's even the point of using a tool that is this limited? lmao.

I absolutely loathe that particular TTS voice. It's like nails on a chalkboard.

you gotta be kidding me.. is it good? it is pretty bad

Still too boring. Waiting for AGI...

@David-gp3fd

2 ай бұрын

Gonna be waiting a while

That music sucks, horrible.

Garbage lol

Are you completely out of touch? you think inpainting is a rudimentary technology with little success? what are you talking about?

@MattVidPro

2 ай бұрын

Natural Chat Based inpainting not inpainting itself

holy crap. worst image generator yet. this is 2 years stepped back. horrible.

@mycelia_ow

2 ай бұрын

The image gen itself is not bad lmao, just this initial iteration of inpainting. Remember, the quality and coherency of its generated image are determined by the quality and coherency of its prompt. It's better than everything out there besides fine-tuned stable diffusion xl.

Completely useless