The Easiest Design Tool is also the Most POWERFUL. (thanks to AI)

[@MattVidPro]

Playground has built an AWESOME tool that is both powerful and great for beginners. While it has limited FREE gens, it still has a subscription. Am I the only one here who is getting pretty sick of that model? Everyone does it, but I can't afford endless subscriptions!!

[@LouisGedo]

👋 hi

[@MrArrmageddon]

This is not open source right?

[@Liminal_Ascetic]

It makes no sense. Subscriptions are minimum $10 per month. Just using a few things, which is necessary for serious work gets quite expensive. And if we lose our income or are just starting out, we can't pay. Games started losing players with this model too.

[@robertolanzone]

Perhaps I shouldn't say this but I feel like it was hinted in this video as well so I'll go ahead... you can always make multiple accounts to leverage the free gens 😅 and hey, at the end of the day, I'm sure the developers knew perfectly well one could do it, it's probably still advantageous for them to let users try for free, be convinced and pay for a plan. It is, after all, an inconvenience, having to switch accounts every time 🤷

[@harikot.vert.7777]

Its not better than Flux ! PG pass the image to chat gpt ... you get similar or better result with FLUX tht is an open model, playground is a closed model they trained, nothing special there DONT wait your money on this !

[@BryanLomotan]

I personally like the old version better. It was so much more customizable and able to "generate" pretty much how I prompt it. Now, I'm still trying to learn this new UI but damn, I've been on it for hours and I just can't seem to find any pros for this over the old one.

[@JNachosl2]

Personally, old version was more useful. This version in just for create stickers based on templates.

[@lin-m80]

I totally agree!!!!!

[@SchusterRainer]

Playground are killing their product!

[@lin-m80]

Yep, I could not have said it any better. I am going to see if Wayback Machine will help me out.

[@electric_bot]

I don't understand why they are killing Board, Canvas and the community of the site. Instead of a full-featured tool that could be used to do a lot of things, they just made a form with two buttons.

[@ECho-q5h]

знаешь какой-нибудь аналог старой версии сайта?

[@electric_bot]

@@ECho-q5h Какой старой версии?

[@grinrobot1562]

I need previos version!!!!!!!!!!!!!!!!!!! Pleeeeeeeease!!!!!!

[@_._._._._._._._._._._._._.]

Too bad they removed the canvas feature from playground

[@jirazjezvec]

WHERE ARE MY PICTURES ON MY CANVASES!!!!!!! I lost all my work.

[@VittorioDaVinci]

I have no words! Are you sane? Bring back the old version!

[@Akshatgiri]

I've been making custom models ( training loras ) for clients recently. The templates are most likely flux loras trained on a specific style. The the way they have the natural prompt iteration set up is most likely an LLM pipeline that takes the original prompt and the input prompt of the user, combines them to make a new prompt and passes the original image as a reference image to the image gen model. This is exactly the pipeline I've been using recently because clients don't really know how to prompt ai gen models, so I translate their general prompts into more specific ones using llms.

[@zedxxx9]

How difficult would it be for a new developer to set up AI models that function in a similar way to this but with a whole bunch of unique starting points and stylistic images? Thanks for your comment. I just found the "The Ultimate Stable Diffusion LoRA Guide" so I can investigate this further. If you hadn't mentioned "training loras" I wouldn't have known what to look for. Very interesting. I will also be looking into similar training for music.

[@apache937]

@@zedxxx9 very doable, the basics on training a lora are out there, just need some cash to rent gpus for training and inference, make the ui etc. still going to take some time . ask yourself if its worth it

[@justtiredthings]

To keep the style and composition so similar, I think it must be doing even a bit more than that on top. Like seperating the elements into layers and applying IP adapters for consistency. Then again, Flux doesn't seem to work well with IP adapters, and SDXL isn't as good at producing great images in one go like this, so honestly idk 🤔

[@justtiredthings]

unless the LoRAs are hypeeeer specific

[@InsideYouTubeMinds]

13:11 "welcome to my 2nd account" 😂

[@armondtanz]

ooooh the dastard!

[@PiotrGzhymajlo]

Hey where is Canvas? aaaaaaaa!

[@joeyhandles]

wasn't onboard until 6:57 this is wild

[@sherlockjerry]

WHERE THE FUCK IS CANVAS!??!?!?

[@zedxxx9]

Playground is NOT GENEROUS. I tried out a few edits, and finally got around to uploading an image to see how it could deal with that, which was thing I was MOST interested in exploring to decide whether a subscription was a must-have... and the minute it uploaded my image it wouldn't let me change it until I started a subscription. What did I get? A total of 10 trial edits? They were just throwaway initial experiments and then I was denied the opportunity to test something I really wanted to see. What a crappy setup they have. If they really want to attract users to start a subscription I think we need more than 10 trys before we are locked out. I will NOT be subscribing. Not right now anyway. Anyone else going on there to try this... remember to use your 10 prompts wisely.

[@Jack122official]

Mat would you like song ai dubbing idea

[@shamz_ai]

Honestly all current/ future generators should operate like this

[@roybec]

I really hate the new version

[@conrifor]

This is such a huge leap in the ai image space that we haven’t had in awhile! I might actually drop some money on this for the KZbin thumbnail potential.

[@AmandaFessler]

Panther-headed centipede isn't real, it can't hurt you. Panther-headed centipede: All joking aside, this app is wild. Thinking of dropping in a month for sub, but I'm tight right now.

[@tigercubmedia7103]

Thanks Man, yeah it is awesome and I would probably use this in combination with Photoshop to grab elements and then add them to something else if required. Got to say the "text" thing they are getting so good with that

[@armondtanz]

Adobe putting together a 5 billion bid for it, then charging everyone more $$$ to use it!

[@DataJuggler]

7:19 Without spell check I couldn't tell if centipede is spelled write or not.

[@RMCanimationOFFICAL]

Even playground Ai has language translations is where to put your image, generate it, And Translate any different language (Not every single like Korean Japanese or Hindi when I tried)

[@youngbroadway]

Yea bro playgrounds upgrade is DOPE

[@kuromiLayfe]

think it indeed uses a pre-prompter with the templates (and a model like flux for the generation).. especially if on cloudservers flux can create insanely good almost img2img level replica’s if used in conjunction with img to prompts nodes.

[@mrrfyW]

13:12 is literally me/everyone else who uses AI in mass amounts 😂

[@mov2947]

這樣改變完全和以前想做的無關，還有以前做過的東西都沒有了！ 現在這樣方式做出來都和選擇了的太像，到時大家做出來的東西都差不多一個樣子了， 完全不好。 沒有創意了。

[@johnny1tap]

Here's how I think it works: 1. Segments the template and creates mask areas that are linked to their verbal descriptions. 2. uses llm to understand what you mean and what that means in relation to the verbal representation of each masked area 3. inpaints needed masked areas using a style fine-tuning (control net?) based on the template and prompts that are created from your prompt

[@andrewsilber]

It would be great if they could train a model such that you specify what “sliders” you want and it gives you an image with sliders for whatever characteristics you specify. “Draw me a lion and let me tweak size and scariness”. And then in real time you could manipulate the sliders and see the image change. Full disclosure I’m not involved with the latest in diffusion models so I’m not sure how realistic that would be. It might need to emit not an image but a mini model of an image generator itself. So you’d need to train the model to emit sparse models. Anyway, that would be a groovy workflow to any of you folks working on frontier image generation…wink wink…nudge nudge…

[@justtiredthings]

slider loras are a thing if you're working with stable diffusion locally. I haven't tried one yet, though, and my impression is that they're not very consistent and that even when they work they often change a lot of other details that you weren't trying to change. The moment someone really nails that, it will completely change not only image gen but video gen, too

[@AI-Art-Explorer]

imagine being able to tweak every detail of an image in real time, that would be wild

[@justtiredthings]

@@AI-Art-Explorer apparently you can now, for expressions anyway, with Advanced Live Portrait

[@TheRealUsername]

Remember GPT-4o omnimodal capabilities? Well it can edit exactly like that, fortunately this startup deployed their own version before OpenAI

[@RafaTV-ai]

Thanks so much again, Matt. You are the go to, bro! I am totally going to use this one for my designs. Stay Kind

[@mkaczmarcik]

Meh... it's boring. Very little creativity involved- just pressing buttons with simple designs. Also, they say on the site to basically take any image of choice from the internet- and run it.. which is very problematic. Playground Canvas is/was better but they are targeting a different user type now. I want the knobs and buttons. Bye, Playground.

[@1242elena]

Working on a crew AI ideogram workflow to do just that! 😎

[@iqmarketers]

The crazy thing is that the Amiga computer had this back in the '90s crazy. Get a chance to get a hold of one it could do everything pretty much the playgrounds doing.

[@JayJay3D]

How could it do that? I had an Amiga and it couldn't?

[@Futurist_05]

Is there any option to change the position of image once ita designed? To left, right or even center?

[@Dmvillan]

where is the prompt ? i am relying on it to change something, also the negative prompt

[@IcyLucario]

A step closer. What I want is this to be handled more with smart masking and inpainting, not just a closely matching image. Smarter and more seamless SD inpaint, basically.

[@skr_8489]

Any offline alternatives like this? Good templates, good text handling, simplicity.

[@Crovea]

It seems to me for the graphic design that they are using a layered approach, which clearly has its advantages, but I bet you would have a hard time integrating e.g. text into the environment/objects. Like a tilted sign at the side of the road reading something speciifc

[@totempow]

I gotta say, I enjoy Playground a bit, but if you have Ideogram... there really isn't much of a point in using it considering the Remix option. If you can't create your own, you can always remix someone else's till it suits you enough so that it isn't the same. I mean if you think about it people are going to be doing the same thing with Playground it's just a matter of who has the template and if you're using Ideogram, you have so many more to change so much more. Just a thought.

[@you2449]

I love Matt's cheery enthusiasm, for something that will eventually destroy us. 🎉

[@bookaitv]

Is the existing function gone? I would like to do the existing method at the same time. I'm doing something appropriate to the existing method.

[@moshuphq9057]

how do you get the old playground interface? i hate the new one

[@digigoliath]

Click on your Profile (bottom left) & "Go to Board". Was surprised and lost too.

[@moshuphq9057]

@@digigoliath thanks...ive actually had a play on the new interface and features now...hate to say it, i guess cos i like to think i am actually doing stuff when i play with these new toys, but it is pretty badboy to be fair

[@IceMetalPunk]

Feels like just the latest iteration of instruction-tuned image-to-image models, like the ones SD has had for awhile. Plus an LLM layer in between to write a better prompt. It's cool, but I don't know how novel it really is.

[@garyrichardson475]

Ideogram is the only one I've tried that spells things correctly. I followed you prompt, "letter text that spells".

[@雪鷹魚英語培訓的領航]

I got some nice results! Definitely a good one to add to the arsenal.

[@TMtheScratcher]

This looks nice for the shown use cases, but the old version was much better to create illustrations from scratch, since you had more control over it. I really hate the new version for that

[@willbrand77]

good for making start and end video frames?

[@shehabdentalclinic857]

Very interesting idea 🤩

[@juninhomello100]

I just need to know why the function of searching for a specific niche and displaying several images related to the topic is no longer available in the paid plan. I am very unhappy with this and will stop subscribing to the tool.

[@DJTechnosapien]

Nice, way too easy! Will try to use this for my album covers and music video thumbnails

[@I-Dophler]

A ban on alcohol would be like banning rain to stop flooding-both unrealistic and rather counterproductive. We can dream, but let’s stay grounded in reality!

[@andydataguy]

This looks awesome!! 💜

[@coloryvr]

Yes! Amazing! Big FANX!

[@AS-ih2dk]

Can it do 4k pixel size or upscale, OR PRODUCT MOCKUPS like the Flux lora?

[@bloodywolftr]

The original prompt is kept within the metadata of the picture.So when user prompts it can do a variation by mixing the original prompt at some weight vs the new prompt.These prompts can be compared by using an llm(even a vision llm can be used) behind the scenes.Also there maybe a mechanism about measuring that how much of an difference in user prompt vs the original prompt about determining the mixture amount(prompt weighting). By using the same seed with a little bit of variance it can be done. If loras are needed or used they can be added as well. However, neither this is a necessary information for the end user who is looking for simplicity, nor for the people who will read these lines to this point :D. A ComfyUI workflow can be done. Edit : I forgot to mention that when the resolution changes,all the magic breaks apart which you have shown in the video and that is the expected behavior.

[@taomaster2486]

So matt you didn't do a good comparison here. When you put outside image into GPT it will avoid copying it due to copyright. To test this change feature you will need to generate image and then change it inside the conversation where you can literally edit

[@fullnfttv]

I don't know why I change the interface often. Please don't change it for those who are used to it. Give me a link to access my existing design! It is really annoying. The previously designed ones have disappeared. Where can I find it? My Design is blank. Please open a link to access the existing version for anyone who is used to the existing one! I can't work on it.

[@bookaitv]

Is the existing function gone? I would like to do the existing method at the same time. I'm doing something appropriate to the existing method.

[@Sandra-d5q3g]

Alguém pode me dizer porque o Playground não está aceitando mais minhas pesquisas de prompts?

[@juninhomello100]

I have the same problem, I paid for the tool and I can't do any more searches

[@NoahtheGameplayer]

Con: the Image to Image does not work technecally, because it has to Generate a Image from your Custom image to order to make it work, which is honestly sucks because of it not working as you would expected, Pro: But i do believe that in other senerios in General Stuff is Fine as well incase if you ever Like this Font you can change the font whatever you like, or Maintain a certain style of the Reference Image

[@kevnar]

Playground seems to fall apart if you add a whole sentence. A few words are fine, but it falls back into the same old AI text flaws with longer phrases. Suddenly, you get alien hieroglyphs.

[@apatsa_basiteni]

Kind of "Instant LORA"

[@southcoastinventors6583]

You just showed training loras for flux this just a more advanced version with integrated llm that could be done using stable swarm or comfyui they already have modules for running llm with image generators still need a powerful pc atm.

[@siddhubhai2508]

Who loves matt's eyes ❤

[@flutterflowexpert]

2 mins after the video is posted and people saying its amazing video :D

[@FSK1138]

voice to txt would be the best input for this

[@gameon2000]

What about the similarly designed Pixverse? It also has Text to Video AND Image to Video!

[@JonnyFootman-k5t]

Very useful tool, i think midjourney has had prompt based variations for two years now

[@gerardolopez9368]

I'm highly convinced your coming out with a Lemon Game or Dev already 😅

[@juninhomello100]

I'm leaving this tool, not even with the paid plan can I search for images in the image bank anymore.

[@peacetoall1858]

Playground is amazing

[@ventonthorn3455]

It took almost ten minutes to do something with lemons? You're slipping. 😏

[@t2p5g4]

This is great. The problem is that the original playground is still a memory hog that doesn't properly free up memory.

[@MusicalGeniusBar]

But no thumbnail feature 😢

[@ZenchantLive]

It obviously uses segmentation like Sam or Sam 2

[@tombradford7035]

It has a lack of control - you upload a pic and it changes it completely(!)

[@juninhomello100]

I'm leaving this tool, not even with the paid plan can I search for images in the image bank anymore.

[@FRareDom]

yo this is pretty cool

[@AI-Chill]

If you are going to choose a template it starts to feel like Canva.

[@juninhomello100]

I'm leaving this tool, not even with the paid plan can I search for images in the image bank anymore.

[@juleslincredule]

You are great!

[@ECho-q5h]

что вы сделали с сайтом? там раньше можно было рисовать крутые рисунки, а теперь это бесполезный сайт

[@Pattayaman999]

He looks like Harry Potter

[@jasonshere]

Cool tool. Poor prompting and spelling, but thanks Matt!

[@claudiac.knauer3410]

It's such a shame that they've changed the program so radically. You can no longer create or design anything yourself. It's like going from expert level to toddler level.

They find good prompts and keep the same seed (then save a preprompt as a "style"), when you replace something in "pic", it actually just does a text replacement in prompt before regenerating image. It's not the AI, even simple javascript.replace(old,new) would work for presubmitting. ALthough, I'm sure there's some prompt magic going on with an LLM.

[@kyouninja]

put a /create button omg

[@magnacarta620]

While you found it impressive, compared to its previous version, in short, it has become useless and they have also deleted all our previous creations!

[@grouchiegrumbles2080]

On computer, maybe it works in IOS, the inpainting does not work. If you just want to modify the text, and not the image, it does not pay attention.. It will redraw the image too. I made the mistake of paying for a month, but this is seriously not worth the money at this point... I tried using the prompts you said, and it still redraws the image.. It's funny how they link to your video for training on how to use their app...

[@R123Productions]

that editing is 🔥

[@garyrichardson475]

We are all going to get "democratize" out of our jobs soon.

[@lin-m80]

The old version was far better, this one sucks, and why remove the daily theme page ? That was the best part, always had to have a look at that every morning.

[@nedo68]

what happens with the shy guy? ;D

[@digitalazorro]

wow! and it's understands swedish:)

[@gerardolopez9368]

Who got here because of his add posted on your channel 👀👀👍👍👍🍋

[@dnero6911]

I think you should switch to grapefruit ... it's time for a change :P jk

[@MisterDivineAdVenture]

Dude - buy some poly Vee Tees or better shirts. Those cotton rags are seriously under-class. Playground seriously DUMPED their original fan base and a lot of excellent work on their Canvas editing toolkit. Without that expert-level editing ability - we're seriously set-back a year.

[@PixelsVerwisselaar]

Playground aint working for me, slow generate, most logo templates are paid (just show me the free ones)

[@MattVidPro]

it would def be nice to sort them

[@harikot.vert.7777]

Just copy the pro image and load it in start design from image !

[@PixelsVerwisselaar]

@@harikot.vert.7777 Will try it later thanks

[@FusionDeveloper]

Cool

[@Azynaya]

Beautiful, but uncomfortable!!! This is a terrible update! There is no possibility of fine tuning, that's why I loved this neural network! Return it as it was!

[@qaswaraaa]

When miniMix video

[@FarazKhanYusufzai]

OMG!!!!!!

[@zabist2938]

This tool is just using their last model v3 under the hood. Nothing funcy. It just a good as adherence prompt and text, no lora no nothing. The templates are just pre made prompts and what it does under the hood when changing things its using "describe image" featrure they have and then reajust the prompt with your new words. That's why when you change or add anything you the image is never the same , it changes always. So if you like one image, and you want to change just one part it will change everything. And if you use their filters, you will loose your original image. A fun tool to play as a free tier and get ref images not for business for fure.

[@nattydg74]

What a shame that palyground has made these changes, the old version was amazing.