AI News - ComfyUI Segmoe and Stable Video Diffusion 1.1 - Phụ đề song ngữ
Hey guys,
this is Kevin for Pixovert.com and in this video we're going to be taking a look at all the important news in Generative AI in this week of February 2024 starting off with
a special offer for the Creative Cloud.
Adobe's Creative Cloud is currently 40% off and this one is a really nice offer.
It's 60% students and I think it covers the entire first year.
So is not one of those ones where you get it at a discount for two months and then you have to pay for the full amount for the rest of the year.
You've got a lot of generative AI tools inside of Photoshop and it
Illustrator which spoken about those on this channel and the Illustrator one the Adobe Firefly I think is particularly
good There are also a whole ton of generative AI features in Premiere Pro
Some of those are very good
But you do need a powerful graphics card to take advantage of those ones and the news that makes this really pertinent is that
there is now a new model for Adobe Firefly.
Now there are now two models and Adobe demonstrate these two models that you can use on their website.
So we've got model one and model two and still I think the best place to use the Uranative AI features for Adobe is on
the Adobe 5.
like website where you can create images like these ones here.
There's a whole bunch of new features like changing camera angles and other options that you can see down here.
I think this one is still free.
I'm not a hundred percent sure but I think there may be some costs if you want to do a lot of images.
You have got the option here to choose.
choose the different models.
And they say here is that Firefly Image 2,
the new image model, can generate higher quality images of people, improve text alignment, and enjoys better style support.
But then they go on to say Firefly Image 2 model is only available on FireflyAdobe.com.
So you want to give this a try, you're going to have to go over to the website.
And are quite a lot of galleries that show you some spectacular results that people have managed to get.
Lynna Kahn, who the chair of the Federal Trade Commission, has started an inquiry.
happening with the AI companies that produce some of the most important and popular models.
Now, if you have been following this channel, you know, I actually had a survey where I was
asking people about OpenAI, which had quite a lot of problems last year.
And some of us thought OpenAI was going to add
die or become some sort of zombie company when Sam Oldman was fired from the company.
However, it was later taken over, more or less taken over by Microsoft and Satya Nadella.
Now this whole development seems to be what triggered the FTC to launch an inquiry.
into these generative AI partnerships.
And the FTC says is history shows us that new technology,
new technologies can create new markets and healthy competition as companies race to develop in a and monetize AI,
we must guard against tactics that foreclose this opportunity.
And basically, she's saying, Lena Carnot.
saying he wants to ensure that the public gets the best out of this new technology and none
of the companies can monopolize the new technology.
The FTC says is that the orders issued today were sent to companies involved in three separate multi-billion dollar investments.
Microsoft and OpenAI, Amazon, and Amazon.
anthropic and Google and anthropic.
What she wants to know, what Lina Khan wants to know is what the hell is going on with these companies.
And I think that's what we all want to know.
The companies are going to have to disclose what kind of strategic partnerships they have.
They're going to have to hand over documents and they've got 45 days in which to do this.
Now one company which It mentioned here is stability AI and they have a special relationship with Intel,
I think we've covered that in a previous video,
Intel are investing quite a bit of money into stability,
but they are still relatively free and independent and they have some interesting news actually where they have come up with a new version of this.
stable video diffusion.
And can see here some of the example images, some of the example videos that can be created with that model.
Now the new model is stable video diffusion 1.1.
And they say that it's been fine-tuned so that with certain parameters, it actually produces better quality.
outputs.
Now, stability have been gradually developing a membership scheme, which allows you to use some of these models commercially.
Now, you may remember with stable video diffusion, they initially were not commercial.
You not use them commercially initially, but then with stable video diffusion.
diffusion they have now got a commercial model that you can use is a $20 a month fee.
So is going to be for companies that need to use this type of technology and it is going to be
for companies with less than a million in annual revenue or institutional funding.
So this is one way which stability is trying to finance itself,
and I think this is important because if they've got these professional partners,
they're going to put a little bit of pressure on the companies to produce high quality models,
and at the same time, the money that's coming in is going to help to maintain the stability and sustainability of the organization.
In a previous video,
this was week,
I spoke about SEGMO, which a new technique in stable diffusion developed by third party and it's now got its own comfy UI custom nodes.
This is actually a third party custom load from a Chinese developer that allows you to use the new models.
This one here, the Stable Diffusion 1.5 one is a, I think, an 8GB download or something like that.
The other ones are huge downloads as well,
but they show you what you can do with this kind of combined method for creating models inside of Comfy UI.
very bare bones approach and there's a lot more power that you have actually with segmo.
The new custom node comes from this developer Joe Joe Joe and he has quite a lot of comfy UI custom nodes and a lot of it is in Chinese but it's basically a
very very basic implementation of the same technique that you'll find on the official segmo page.
Now I did find this new custom node allowed me to work much much quicker than using the integrated development environment which was extremely slow,
but it does use exactly the same.
kind of features as the as the official segment so it basically plugs into the pipeline from
the official repository and it has the same advantages and disadvantages but it doesn't
actually pull in all of the features that are available with the official segment.
So, segmo is very powerful, but we've only seen so far in the custom load, just a tiny
fraction of what it's actually capable of doing.
Huggingface have actually created a blog on a github, which explains a little bit more about segmo, and specifically the segmo mixture of experts.
but also about the philosophy behind this particular technique.
As can see here we've got some of these very bright, very vibrant images.
These ones were created by SDXL times 4 models.
Here we have the SDXL times 2 models and then we have the same models that I was using which is the stable.
fusion 1.5.
They also have a page from December last year which describes the whole mixture of experts methodology,
the history behind it and it goes into a huge amount of detail.
If want to learn about this, the strengths and the weaknesses.
So this is pretty interesting,
very interesting and hopefully I don't know I'm looking forward to seeing something more along
these lines it's a really fascinating idea and you should be able theoretically to choose whatever models you want from stable diffusion,
CIVID-AI models or hugging face models and just combine that.
them very quickly.
And themselves have a research roadmap for this.
And the same what they're looking for is extra speed for the Segmental and optimized memory usage.
Memory usage is going to be huge with these particular models because they're literally just combining four models.
models, two models, they want to add Laura support, support for more models and also support for training.
Now in the field of AI video runway held a small competition where they received numbers of entries,
hundreds of entries like this one here, which is called red energy.
And this one covers a story about someone whose planet is dying,
the stars are dying, and is a short movie, one and a half minutes and is really fascinating using the runway models for video.
Some of are really fascinating tiny little stories which tell fascinating,
I know, it's kind of like dramatic stories about stuff happening that you can only put together using AI.
So this one, Red Energy, was one that caught my eye.
Another one that really caught my eye is this one here which is called Come and
Get It and it goes into some really fascinating story about this sort of
Freudian galactic story about this sort of crazy empress who dominates over this empire and it's very surreal and it's of the technology.
Now the technology is still fairly new but with good audio, good storytelling, you can have some really interesting outputs.
This one here,
which called I think SUNY and Holdra,
is one which really makes good use of the technology and it tells a story within the limits of the technology.
The audio is good,
storytelling is about these spirits in the forest and it brings these spirits to life using the technology,
the ability it has to create more thing and to create unreal images and it's a really interesting way of using the technology.
story whilst working within the limits of the technology and it's really good to see
people putting together ideas in storytelling using the video but also using very good vocals,
very very, very good sound as well which makes everything come together really nicely.
Now, as to which is better stable video diffusion 1.1 or runways videos, I think stable diffusion
might actually have a slight advantage in terms of the image quality.
And I've been following some tweets from Imad Mosdak, who's the big man over at Dustability AI.
He's the CEO.
And he's been posting these really fascinating little videos, or at least reposting them, which apparently are using Stability AI's stable video diffusion.
And can see some of these are really, really nice quality.
Nvidia announced the launch of GeForce RTX 3050.
6GB.
This a new card which has got 2GB less than the previous RTX 3050 and it's one which is going to come in under $200.
There were rumors about these ones and I think I might have mentioned this in a previous video or I might not have because we weren't
sure whether it was.
definitely coming and to be honest with you we're still not a hundred percent sure when it's
going to actually hit the shops if it's going to hit the shops should be under two hundred dollars
if it does come but this might be limited to countries like Russia India some of those countries
where the market is a little bit different who knows this very little information coming from Nvidia.
The main thing that this saying here is that the new card is going to have a TGP of 70 watts.
Now you can get 75 watts from the PCIe lane inside of any computer, in any modern computer.
And what this means is that you will be able to power this without any additional cable.
just need a decent power supply.
It's provide all the power that you need.
So it should basically be plug and play in practically all computers.
I think for someone who's using stable diffusion with comfort UI with other versions of stable diffusion,
that 6 gigabytes is gonna be just enough to get started.
You
Ngôn ngữ dịch
Chọn
Mở khóa nhiều tính năng hơn
Cài đặt tiện ích Trancy để mở khóa nhiều tính năng hơn, bao gồm phụ đề AI, định nghĩa từ AI, phân tích ngữ pháp AI, nói chuyện AI, v.v.
Tương thích với các nền tảng video chính
Trancy không chỉ cung cấp hỗ trợ phụ đề song ngữ cho các nền tảng như YouTube, Netflix, Udemy, Disney+, TED, edX, Kehan, Coursera, mà còn cung cấp dịch từ/câu bằng AI, dịch toàn văn sâu sắc và các tính năng khác cho các trang web thông thường. Đây là một trợ lý học ngôn ngữ đa năng thực sự.
Trình duyệt trên tất cả các nền tảng
Trancy hỗ trợ tất cả các trình duyệt trên tất cả các nền tảng, bao gồm tiện ích trình duyệt iOS Safari.
Nhiều chế độ xem
Hỗ trợ chế độ xem rạp, đọc, kết hợp và các chế độ xem khác để có trải nghiệm song ngữ toàn diện.
Nhiều chế độ luyện tập
Hỗ trợ luyện viết câu, đánh giá nói, trắc nghiệm nhiều lựa chọn, viết theo mẫu và các chế độ luyện tập khác.
Tóm tắt video AI
Sử dụng OpenAI để tóm tắt video và nắm bắt nhanh nội dung chính.
Phụ đề AI
Tạo phụ đề AI chính xác và nhanh chóng trên YouTube chỉ trong 3-5 phút.
Định nghĩa từ AI
Chạm vào từ trong phụ đề để tra cứu định nghĩa, với định nghĩa dựa trên AI.
Phân tích ngữ pháp AI
Phân tích ngữ pháp câu để nhanh chóng hiểu ý nghĩa câu và nắm vững các điểm ngữ pháp khó.
Nhiều tính năng web khác
Ngoài phụ đề song ngữ cho video, Trancy còn cung cấp dịch từ và dịch toàn văn cho các trang web.