Funny how all AI websites have anual plans, the landscape changes so fast than in weeks or months you might just have a better open source alternative.

Buying yearly subscription when things move that fast in AI it should be 90% off lol

with how fast things change i dont see myself subscribing annually to anything.

It's crazy how I can check your channel almost every day, and there is something new and exciting!!

Stability AI should give us an open source upscaler like that :)

I don't get the point of upscaling if it alters the image as substantially as this does. What I want is an AI upscaler that produces an image that looks indistinguishable from the original when downscaled.

Thanks a lot for all your videos, Matt! Just a note according to Magnific: it only takes 5 credits for a 2x upscale of a standard e.g. midjourney image size (round about 2000 px > magnified to 4000 px), so my monthly 2,500 credits in the basic subscription go a long way with 500 simple 2x upscales. And if I combine it with the cost free upscayl app for the final bigger upscales, it is a perfect combination for a lot of use cases, and I often prefer the Magnific upscales to the midjourney ones. And special thanks for introducing Upscayl in one of your last videos :*

The chameleon doesn't have a little extra eyeball... the upscale gave it a nostril!

Will it also be an upscaler model one can use locally/offline?

Its not an upscaler at all. Its just another Image generator. If you look at all the images that its upscaling. It's just a regenerates the image from the first image. Its just a Image to Image generator. Every image generated is different to the original. So if you have an image that wants to be upscaled, and it needs to be a perfect 1:1 image, this will not do that at all. It will just generate another image of what you have, and things will be different from the original. Dont get me wrong, it's really amazing, I love it myself. But I use Topaz Photo AI. and it upscales and does not change the image.

Nice work Matt keep those new gadgets coming our way.

Just 10 seconds in for single images of a video, and you're in for a visual rollercoaster - objects leap and morph unpredictably, making every moment an unexpected thrill.

Loving the content! Keep it up!!

Nice reveal of the new tool looks very slick can't wait to try it out on my cat photos.

Absolutely excellent info as always. Congrats on a quarter million!

Love it! I upscale photos before printing with great results. I use open source free Upscayl and add my own models

Who else checks Matt every couple of hours for cool AI news like these 😂

Nice ! Thank you Matt. 😁😄😊

this is not upscaling… this is stylizing

I would also add that Leonardo has some pretty cool upscaling options, although said upscales are modest. Thanks for the video!

9:44 It looks like a real cameleon nose, not an eyeball

Matt can you speak about commercial use with Open Art though? It says on their website attribution in required. How's one to do that on a tshirt design? Kinda makes it a deal breaker for me

I’m looking forward to a local upscale that provides almost the same result.

i wonder if it would be possible to have upscalers improve their consistency by having them trained both to produce a bigger image, but also to keep the bigger image looking similar to the smaller image when downscaled to the same size; or even downscale both to something like half or a quarter of the size of the original to give some room for creativity.

Hey Matt! Check this out Theres this thing, Air LLM If i understood it correctly, it seems to allow 70b models yo run on 4gb hardware Here is a long wall of text by perplexity Understanding AirLLM and Its Significance AirLLM is a groundbreaking technique that facilitates running a 70-billion-parameter large language model (LLM) on a single 4GB GPU, overcoming traditional hardware limitations. Traditionally, such large models would require more powerful and expensive hardware, restricting accessibility. **What is AirLLM?** AirLLM stands for a method allowing large language models to operate on smaller, less powerful GPUs. It employs two main techniques: 1. **Layer-wise Inference:** Breaks down the model into individual layers, loading only the necessary layers into memory during inference, reducing the overall memory footprint. 2. **Flash Attention:** An optimization within layer-wise inference, focusing on loading and executing only one layer at a time, further minimizing memory requirements. **Why is AirLLM Significant?** The significance lies in democratizing the use of large language models, making it accessible for personal projects, educational purposes, and small businesses. Practical implications include the ability to run advanced AI models, like chatbots, on smaller hardware, fostering innovation in various fields. **Running 70B LLM on a 4GB GPU with AirLLM Technique** **What it is:** AirLLM is a groundbreaking technique enabling the operation of a 70-billion-parameter large language model (LLM) on a single 4GB GPU, overcoming traditional hardware limitations. **How it works:** The technique employs layer-wise inference, dividing the model into layers and loading only the necessary parts into memory, drastically reducing memory requirements. **Key Features:** - **Layer-wise Inference:** Splits the model into layers, significantly cutting down on GPU memory usage. - **Flash Attention:** Optimizes single-layer execution, further reducing memory needs per layer. **Benefits:** Allows for efficient and effective inference on a 4GB GPU without compromising model performance, bypassing the need for high-end GPUs or excessive RAM. **Applications:** Ideal for scenarios where hardware resources are limited, such as personal projects or low-budget research. This technique represents a significant leap in making advanced AI models more accessible, enabling users with modest hardware to leverage the power of large LLMs for various applications.

It’s a good day when Matt uploads

Is there any good uspcaler for VHS footage?

Great video, thanks

I started playing around with Pinokio and Stable Diffusion on my PC. Are there open source image upscalers we can install locally?

2:40 It's not 3D "render", it's "rendering" ...paint vs a painting = render vs a rendering. render is not a noun.

Released??? Where can I try SD3? I'm on the waiting list.

This upscaler thing is way over my head but looks cool!

👋 Is this new upscaler the one that comes packaged with a ClipDrop subscription?

Will you do a Suno v3 video?

It's reality touch it put on all objects increases its “real” look and more difficult to distinguish between an actual photo and AI generated, which is one of the biggest tests for me on how good a model is. Well done!

When will we see this and SD3 on Clipdrop?!??!?

Doesn't Krea have the creative upscale? Also note, the Krea free version is very limited to the paid plans. (But are much less expensive than Magnific)

Love your vids

6:30. This is really sweet. The lemon was meh.

It did introduce an error though. Go big or go home apple. The G in big has a super comma

Great video. Test it on ps1 screenshots.

7:53 the O in "OR" is messed up

suno v3 is out now for pro and premium members

Kinda cool, but shouldn't they offer an open-source option if they're calling themselves OpenArt? Or are they open in an OpenAI way ? (i.e. closed)

Still has some issues upscaling hand digits and the flag of the astronaut on the pig...... I'm still not quite convinced....

This is more like inpainting than upscaling; adding a bunch of details that weren't even implied in the original image.

Lets hope Open Art provides early access to Sora just like they did with Dalle 2

Cool video! Guys, does anyone know of a model, checkpoint, or specific library for generative AI image models, suitable for UI/UX, web design, and user interface prototyping? The models I've found are very expensive in terms of ultra-realistic images, beautiful illustrations, etc. If anyone has an idea, it would be a huge help, it's for my college work

I gave it a picture of myself nude. and a big warning came up saying, 'no AI system on earth has the compute power to make that look good'

Denoising strength = Creativity Slider 😊

Push for open source upscaling✊

krea ai also has a creative upscaler

Subscribed to basic plan and tried creative upscale, but after wasting hours after trying I find that the image upload process is endlessly running. It seems the website is not functioning properly and the helpline is not even replying to my SOS mail. Thinking about asking for a refund.

начало тут 10:46 если устроят условия , можно смотреть ролик целиком. start at 10:46 if you're happy with the conditions, you can watch the whole clip.

Why is your camera so shaky

I thought there was something cool here. It's basically a basic thing I do by hand. It's not a one-button solution, but it's completely buttonless and under my control.

Matt camera could do with some stability

07:56 "take about perfection", the text was clearly messed up... 08:56 "the pig feet look awesome", it has two extra feet...

Their AI really makes average photos shine.

So... it seems like its generating an entirely new image with a higher resolution. Not adding details and upscaling. Pretty much the same as using controlnet in img2img.

Hi Matt! I didn't know about openart, thank you for the enlightenment. Hope one can train LoRA there, if so, that's what I'm looking for 💪🏻

i can’t even use sd3….

Where the hell has the upscaling option gone on open art? I've just generated four images and they look terrible and I cannot find it anywhere. If they have removed it then I'm cancelling my membership.

8:01 - That's not perfect since Open Art AI changed the O into a G and that's not good, it ruined the text, but regardless it's pretty amazing how it made it look real like a photo in comparison to the SD 3.0 original image

The astronaut pig reminded me of Muppets. Piiiiigsss innnn spaaaace!

Fix your camera on the wall, when you touch the table you make it wobble and so the image wabbles too...

Cool.

That image enhance of Yeshua is really good

5 a day.... lol

It worries me that I'm seeing fewer and fewer examples of what these new image generators and upscalers can do when it comes to people. Hands. Fingers. Eyes. I worry that we've gotten so concerned with what people are going to do to Taylor Swift's saintly image that the guardrails are being brought up around human images in general. Somebody please prove me wrong.

It rate limited me for 24 hrs after 5 upscales. I paid $12 for THIS?

No link in description, Matt? You're usually pretty good for that.

This is great news!

His hairline is not receding, I saw his giga pixel video earlier and noticed that his hair looks the same

Open Art ❤️

coo

are you generations private?

Thanks John Oliver

no free upscale trial

Nice

I really hope newer models get better at upscaling than SDXL so we can avoid using these payed services.

I went on Openart but I am not even allowed to test the upscaler without upgrading to a premium account (that means paying !)... Sorry Open art, but I don't pay for something I personally did not test before

Midjourney's paid plan isnt worth it anymore.

THat's not upscaling! That's reimagined high quality image 2 image ai. I've yet to see a proper upscaler

5 images per day XD wtf

taking so far over 2 min to make one image...

Hello, I have watched many of your videos and I always wish you could speak in your normal voice, also you never seem to have anything negative to say about AI.

I really hate the fact this is being marketed as an "upscaler". At 7:03 you can clearly see the result isn't being upscaled but instead regenerated, the lighting, color and shape of the apple has completely changed and mind the fact that the text in the background has too.

Check AI Animation Music Video for some dope HipHop Track .Peace Matt

Local Stable Diffusion does creative upscales. This tool is for people who don't run SD. The upscaler botched the text in the apple image.

nowadays upscale is changing to redraw.

Umm 1:37 up-scaled Asian person = white person? I hope that transformation was due to the prompt and not a bias in the model 😅

Don't mislead with the thumbnail!

Ummm can I just say something here, and other techies of the same age chime in... so things move fast in tech, been in IT now 40 years... its a rollercoaster. But this time, with AI tech... Its moving too fast, people can't apply it cos by the time they come up with a use case somebody makes that idea obsolete. Its not sustainable for industry at this speed of advance.

dude! they wont even let you try it once for free. I mean seriously.

Tried this but was eh.

It sucks for two things, 1-it's not free and 2-it doesn't allow nsfw

1:36 An Asian Girl Became Non-Asian 😐

Nah. I’m good. Open source is the way. This kind of stuff needs to be available for everyone for free. There are a lot of creative people with ideas that can’t pay for stuff like this.

I need to upscale Matt's hairdo.