Text To Image AIs Just Got Supercharged!

[@kenb1]

More *pineapple* on my pizza? What a time to be alive!

[@mercerwing1458]

mama mia

[@stevemartin9396]

Secretly the best pizza is ham and pineapple!

[@marceelino]

@@stevemartin9396 no.

[@Uhfgood]

One of those odd things is you either love it or hate it. I typically don't like sweet with savory, however, there are some things I do love, and pineapple on my pizza is one. (Sweet&Sour Pork, and Teriyaki chicken are a couple of others). I imagine there will be half the people in the comments that hate that pineapple on pizza, and the other half to love it.

[@kristoferkrus]

Have you played the game Pineapple on Pizza?

[@Dan-hw9iu]

I thought "two minute papers" meant quick summaries, not "a new, groundbreaking paper every two minutes." The pace of AI progress is dizzying.

[@Barquevious_Jackson]

What are the breakthroughs on this one? This looks like something that was already possible.

[@kamikeserpentail3778]

Isn't that basically what the singularity is about? We may not be in it, but it sure feels like we're getting close fast, and this is before we have true AGI much less ASI.

[@siddheshdhomane7642]

soooo true 🥲

[@matthewdignam7381]

Love how Google just broke the ai industry when you commented this 😂

[@funnyperson4016]

What a time to be alive! 🎉

[@acostin]

Károly thanks a lot for your review. This is Alexandru from the Firefly team, big fan of your channel and I have to say that some of your episodes that I watch with my family helped me get educated on the progress happening in this space and motivated to drive the team and Adobe towards embracing this. THANK YOU!

[@tommyxev]

Keep up this incredible work!!!

[@Chemson1989]

Looking forward for an AI tool to generate 26 characters of new font from only one character

[@kashaid]

These tools will make photoshop redundant soon, however, I think Adobe has learnt from Blockbusters - better to be in the game rather than be substituted for another

[@JackCrossSama]

Competition is going to get crazy.

[@hombacom]

Without control and a good user interface you cant do consistent work. When the hype fade away it become harder and harder to do an AI that fit all needs.

[@kashaid]

@@hombacom these are the initial version, they will only get better

[@hombacom]

@@kashaid what do you mean, this is nothing new and more or less all software get better. Selected highlights has little to do with daily work where you run into all kind of oddities.

[@kashaid]

@@hombacom the ability for you to say what you want from an image and get what asked for rather then spend hours fiddling around with different combination of features and settings to get an image to 'look right' would be boon for most individuals, who don't have the time or inclination to learn the intricacies of tool like Photoshop.

[@AMA14700]

We need a video about Google's new Gemini AI

[@yudoball]

Footnote function is insane

[@JorgetePanete]

The rich text prompt is interesting, but using it is as insane as the password game with all kinds of things on a prompt

[@anirudhnarla4711]

"My goodness" has to be dr karoly's official catchphrase

[@guilhermehx7159]

Actually its "oh yes..."

[@nadavvvv]

What a time to be alive

[@nikamak1617]

AI is developing so fast!

[@BorisEysbroek]

Nah, it's developing too slow. We still haven't hit singularity.

[@lonewitness]

@@BorisEysbroek Silly goose the corpos would never allow such a thing

[@BorisEysbroek]

@@sfjlfkjsdlfkjds I like your energy. Lets hope for the best.

[@florisr9]

@@sfjlfkjsdlfkjds Wise remark about the greens and lefties

[@OperationDarkside]

Released the minute Google Gemini is showcased. What a crazy time to be alive.

[@ThatBidsh]

what we need now is to go from text to image -> sensor data to image + audio + pressure + heat + smell + taste + proprioception + etc... where all those things are integrated into a single data structure, and where changing the values of different aspects of that structure allow you to move through a sort of latent space of all possible world states and self states for an agent to experience... and then we need to give that being a gun, ya know... for science.

[@kristoferkrus]

The comparisons between Firefly 1 and Firefly 2 looks very much like the comparisons made by FreeU between a normal U-net and FreeU, for example kzbin.info/www/bejne/Y3S9ZqiNrduLaZI and kzbin.info/www/bejne/oIXGeI2BoM6ZqaM . Basically, they added a Fourier-based convolution in the skip-connections in U-net which made it perform better. I guess what they did in order to get so similar images for the "old" and the "new" method in their comparison was take the (trained) U-net used by the old method, insert the convolution operations to make it a FreeU network and finetune it once those were inserted. It's likely that Firefly used the same trick to enhance their models.

[@t2t117]

Im afraid of the day he calls something a big AI

[@anatolydyatlov963]

Big Pharma, Big Tech, Big Brother, Big Data... yeah, Big AI is next

[@iusearch]

I'm afraid of the day he tells us to throw out papers because AIs write them now

[@TLguitar]

​@@iusearchI'm pretty sure AI is what ultimately ruins civilizations (referring to the hypothesis that advanced alien civilizations never contact us because there is a threshold of technological advancement that results in self destruction, although there is probably a physical limit to space travel regardless). The person running this channel gets off over whatever advancement in AI is demonstrated because it is technologically impressive, but him effectively ignoring (or at most diminishing the importance of) the inevitable grave consequences for all of these giant leaps humanity is taking, sprinting through a marathon with its eyes blindfolded, is borderline ignorance.

[@MikeyDavis]

@@iusearch I’m still holding on to my papers for dear life.

[@MikeyDavis]

@@TLguitar You must accept unchangeable inevitabilities and enjoy life the best you can until then. There is so much destructive potential on this planet that has been more than capable of destroying us all, yet we’re still here. Just assume that’ll continue this time too.

[@avenuex3731]

5:10 ANCHOVIES

[@Lucas-po6mn]

1:35 did you put that audio clip by accident ?

[@siddheshdhomane7642]

please stop giving us devs existential crisis 🥺...

[@DemonDogs-z4d]

True man

[@ShrutiLagad]

lol some AI did predict long ago that AGI would be achieved by 2028...

[@udittlamba]

6:30 that's matrix in the making. 7:40 that's neo learning kung fu.

[@Max-ji4wk]

We are starting to ride the exponential slope 🚀🚀🚀 What a time to be ALIVE 🪨

[@fynnjackson2298]

Since about when ChatGPT came out, since then the progress seems to have really started speeding up

[@Max-ji4wk]

True. It just feels it is picking up more speed this week :) @@fynnjackson2298

[@21EC]

Judging based on the current speed of things that are developing in the AI world I think that its a type of AI singularity of sorts or an initial pre-singularity of the AI field since it develops from week to week and within months the progress is like years worth of progress compared to other fields of tech, its beyond incredible at this point how fast it all develops on all fronts of AI

[@kolosso305]

"What a time to be a live" Is becoming more and more wild these days.

[@xl000]

* alive

[@anatolydyatlov963]

Meh, I don't see anything revolutionary in Firefly 2, considering that we've had Stable Diffusion models capable of much more for quite some time. It's akin to Google releasing a proprietary text editor similar to VSCodium, which is already open source. Why get excited about it? What's the point?

[@Nathanza]

I think the difference is that millions already use Photoshop. Stable Diffusion and even Midjourney is not nearly on the scale of Adobe in terms of users. You're right that the AI itself is not that impressive with what we know is possible, but the amount of users having access to it is notable.

[@ipodtouchiscoollol]

if these demos are actually anywhere close to the actual thing id say its a big step up over SD, last i used SD you cant change hair colors without the entire image changing as well also SD suffers quite alot from color bleeding but then again maybe there are new controlnets or whatnot that can fix these issues

[@anatolydyatlov963]

@@Nathanza Well, you have a point, but on the other hand, it's pretty interesting that people are not only willing to spend over a year waiting for a proprietary system, which is worse than the already existing open-source counterpart, but they're also ready to pay for it. This reminds me of how Apple fanboys get excited about "new" features in iPhones that have been present in Android phones for ages. But yes, you're absolutely right: people accustomed to a particular ecosystem often hesitate to step even slightly outside of it, and it's definitely more convenient to generate pictures directly in PS than using a separate tool for this purpose.

[@anatolydyatlov963]

@@ipodtouchiscoollol Indeed, you can easily fix those issues with various plugins, controlnets and especially in-painting, that allow you to achieve results much better than what is shown in those demos, not only in terms of pure quality, but also the amount of control that you have over each aspect.

[@altberg__]

amazing. footnote is sick! but. fact is that they took our jobs. tuka jooo. dey duka doe. tuka dweee!

[@Zen_Power]

Tsukyomi training. When 1 second = hours

[@davidwipperfurth8465]

The only rich text feature that seems even remotely useful is the hyperlink feature, assuming it supports multiple links so that you could link every major concept to an example of the concept you were thinking of. An image is worth a thousand words. Plus, while it's possible to combine multiple images into a single generation currently, it's not as easy to associate each with a different concept in the prompt. All the other things are easily doable by adding an extra word or two to the prompt. If I were to try to find a use case for all those other rich text features, the best I could come up with is auto-generating images for existing wikipedia text, or other existing wiki/blog text.

[@Yourname942]

I'm so grateful for your videos. I love seeing the progress!

[@jamesscourtos3583]

Love the font alternating interface, I hope that becomes a standard in future AI image generators.

[@kinsley7777]

LOVE your energy !

[@jaredjones6570]

Confirmed: Two Minute Papers uses AI voice generation in the videos. How can I tell? Just listen to the time segment between 1:13 and 1:18. No human would ever say "Can you?" that abruptly and unnaturally. Still, great videos! I do enjoy watching them. The speech patterns have just always been a question at the back of my mind...

[@synaestesia-bg3ew]

Who can afford $600 annually for Adobe?

[@Wrzl]

00:06 Adobe's Firefly 2 AI is a huge improvement 01:21 Text to Image AIs have improved significantly and are being widely used. 02:21 Advancements in text to image AI technology 03:26 Using rich text effects to make prompts and images more expressive. 04:26 Text to Image AIs can now be customized with links and text size. 05:31 AI can create virtual copies of real-world scenes 06:36 Text to image AIs have been supercharged! 07:47 Simulating real-world environments for AI training

[@JohnSmith-gt3be]

I’m just waiting for the day that AGI is spoken about on this channel

[@-Evil-Genius-]

🎯 Key Takeaways for quick navigation: 00:00 🌌 *Introduction to Three Key Developments in AI* - Adobe introduces Firefly 2, a significant improvement in text-to-image AI. - Comparison of Firefly 2's capabilities with its predecessor. - Emphasis on the widespread availability of Firefly 2 in Adobe tools. 01:21 🎨 *New Features in Photoshop for Creative Text Effects* - Overview of new text effects in Photoshop, including generative fill and recolor. - Introduction of text effects like generative recolor and generative fill for vector images. - Critical analysis of the overall effectiveness of the new Photoshop features. 02:59 📜 *Enhancing Text Prompts for Image Generation* - Discussion on using rich text effects for creating more expressive image prompts. - Introduction of font styles, footnotes, links, and text size adjustments in text prompts. - Exploration of the potential creative applications of enhanced text prompts. 05:16 🌐 *Transformative Capabilities: Altering the Real World with AI* - Explanation of a technique to capture real-world scenes, estimate geometry, and simulate virtual objects. - Illustration of how the virtual objects interact realistically with lighting, reflections, and shadows. - Discussion on the potential applications of altering the real world through simulated training scenarios for AI. Made with HARPA AI

[@markmuller7962]

These real time art makers looks very funny to use

[@DamianReloaded]

I've become so numb, I can't feel you there Become so tired, so much more aware I'm becoming this, all I want to do Is be more like me and be less like you

[@alexdavies7112]

Cheer up mate

[@DamianReloaded]

It's the AGI @@alexdavies7112

[@gustavodutra3633]

I love the lyrics of this song

[@AlanMitchellAustralia]

Incredibly cool tech. But part of me wonders what the next major real-word application of generative imagery is, beyond largely novelty? Gaming? Entertainment? Personalised education? Personalised fashion?

[@GhostBranchFPV]

I'm still relatively unimpressed by Adobe Firefly for text-2-image generation. The new stuff they're doing is just a quirky version of features other text-2-image models have had for a while. It is however incredibly useful inside of Photoshop for generative filling. If only it were as good as Midjourney or offered as much flexibility as Stable Diffusion with Control Net, that would be such a big leap for Photoshop. As for the rest of the video: Wow, just wow! People have been saying it will take at least another ten years before self-driving cars become perfect, but now these ten years could pass in days. It's like time travel.

[@tomaszkarwik6357]

1:12 just look at the finely tuned sdxl over at cvitai

[@shdon]

Pineapple on pizza? That is playing with fire! You may have gone too far in this video 😂

[@nskies2001]

Белеет парус одинокой В тумане моря голубом!.. Что ищет он в стране далекой? Что кинул он в краю родном?.. Играют волны - ветер свищет, И мачта гнется и скрыпит… Увы! он счастия не ищет И не от счастия бежит! Под ним струя светлей лазури, Над ним луч солнца золотой… А он, мятежный, просит бури, Как будто в бурях есть покой!

[@konrad129]

Is there any AI you would recommend for generating 3d models? Or mb some image to 3d model?

[@P.viridis]

1:13 your voice is AI generated

[@encabulator99]

handed out is a strong way to put it.

[@doctorartin]

GTA VII perhaps won't take 10 years then 😁

[@WifeWantsAWizard]

Excellent video!!! (3:18) The answer isn't rich text. That's still looking at the problem like a flat canvas that must be painted from left-to-right in one unending wave. Instead, the answer is "grouped subjects". There can be no true AI art unless users can describe the subject, the "middle ground", the background, etc. a proposed generative image independently in different prompt boxes so there is no adjective drift. (3:52) That's nice except "Ukiyo" was invented by a man named David Kerkhoff. How does it know not to turn the ocean into a kitchen table?

[@aron.mp4]

Incredible! Fantastic work sir!

[@triplea657aaa]

Gonna need to do a video on all the Gemini stuff Google just dropped

[@beansnrice321]

Ung, Ai art is soul crushing. (

[@dibbidydoo4318]

They open-sourced font styles for Stable Diffusion.

[@rajendrameena150]

So they can track stable diffusion also and increase their font use in open-source tools.

[@privileguan9127]

Sounds mostly like advertising a different text2img model. The font stuff is mostly useless and unpredictable instead of just weighting. The integration of objects in virtual spaces seem to progress well, though.

[@Chris.Davies]

I pay for Adobe Suite every month. How do I access this, please? WHAT A TIME TO BE ALIVE!

[@RichardPeterShon]

This is absolutely bonkers?!?!

[@lolgreenwood7038]

There are multiple giant discoveries every day now

[@Famouslastplace]

@TwoMinutePapers are you using AI for your video audio now? Their is this weird hard gating effect between syllabus.

[@ibie27]

is it free?

[@alexanderwalter4595]

6:03 Seems like a glitch -- the orb casts a shadow, but the table has no shadow.

[@Bartskol]

That's been for ages. Adobe firefly 2, nearly two months ago.

[@bcgames92]

I’m confused, is this in the current version of Photoshop? Because I am using generative fill and it is still awful.

[@PythonPlusPlus]

How are they able to compare the exact same images between model versions?

[@ratatouillepg3135]

We are actually going to create the matrix, WHAT A TIME TO BE ALIVE!

[@meijiishin5650]

Isn’t text size already available in stable diffusion? You can change the weight of words in the prompt.

[@brianorca]

I might be a little behind, but I thought the weights in SD are using text mark up [3] instead of rich texts. The footnote idea looks new.

[@Sound_Stable]

You lost me at adding more mushrooms

[@FreedomRoseStein]

Hey random question, did you use Ai to do the voice over for this one?

[@avenuex3731]

Waiting for, say, a Gibli movie recreated with dead realistic rendering

[@ajs1998]

The font stuff seems like such a gimmick. We already have adjectives with definitions we all agree on. Why would we want to replace adjectives with something more ambiguous like font family, color, and size?

[@alan2here]

The 2nd one is completely goofy, I mean 🤷🏻‍♂️

[@KittyAnimates]

what a dim time to be an artist...

[@syntaxed2]

Thank god, I missed the AI news - We need more news about AI, more more!

[@64kernel]

Since everyone is an artist now, prices dropped to the floor for vfx artists. Adobe needs to rethink its business model.

[@lobabobloblaw]

Stay focused, fellow scholars! When overwhelmed, retreat to your philosophical studies and take a moment to breathe.

[@blackurco]

Another AI video, what a surprise!

[@SHAINON117]

More love ❤❤❤❤ ai is just oozing love everywhere hehe 😋

[@woodsofchaos]

how is this better than midjourney 5 or dalle 3 ?

[@willbrand77]

Getting closer to the matrix every day

[@Atriace]

This news would be so much more relevant if Adobe actually sold software. Until then, the fact that they're doing anything with this blackmail money is borderline insulting. The sooner people stop paying for their gestapo subscription model, the better the industry will be.

[@SumUnicus]

That Brickhead looks good, but still it can't handle real Lego parts. When will we get an AI that uses real Lego parts? Can we make one Doctor? I know u like Lego 😁

[@gawni1612]

tfw when youre hoping hell say as a light transport researcher by trade!

[@bzhmaddog]

What freak would put pineapple on Pizzas. This is madness !

[@thomasgoodwin2648]

2 more papers down the line, the AI's writing them will optimize to 1/2 paper down the line. 🖖🎅👍

[@TeddyLeppard]

Graphic designers and photographers are now looking for new careers.

[@riyan_prof]

I think AI just for scientific and development usage only, not for mass production... For calculating and prediction and analyzing. Or people will loss the job...

[@Falkenbergo]

Why do you show something you already have shown? This firefly 2 is not new?

[@Isaac-wr8et]

I dreamed of a future where AI made mundane jobs unnecessary so that we could all focus on our art, not a future where AI took all the art jobs, leaving everybody else with mundane jobs 😅

[@Ortagonation]

Seems like it is great for fantasy like thing. For Something real, the old is better...

[@Uhfgood]

Looks great, except I don't use adobe products, and probably won't ever.

[@아카만미카사]

make a vid about google gemini pls

[@alexdavies7112]

Are we actually just unlocking the same technologies that are used to generate / simulate our current universe?

[@onyx5902]

I really love your videos, but I preferred it when you made videos that were closer to 2 minutes

[@deadcreatives1033]

supercharged my a$$

[@gfdggdfgdgf]

Not being able to run locally basically equals not very useful.

[@GifCoDigital]

wow they added in SDXL. This is very not impressive

[@eSKAone-]

crazy 💟

[@Lucasbrlvk]

😮

[@matthewromig8748]

Me when I whatever is the newer more betterer gaussian splatting with firefly and make a decievingly realistic VR game

[@JOHN.Z999]

Wooow 😱😱😱😱😱

[@SaltyJuan]

I feel like you’re voice is AI now or atleast sometimes

[@OmegaGlops]

I can't help but notice the unusually high amount of furry characters in Adobe's examples... 🤨 Do they know? 🤔

[@Sarimae23]

adobe is censoring my own owned photoshop ...no...adobe sucks.

[@ronnetgrazer362]

Adobe isn't paying you, so you can be truthful. So we shouldn't trust your endorsements, got it.

[@ico-theredstonesurgeon4380]

Tisca tusca