First Look at Google's New Imagen 2 & Image FX Interface!
HTML-код
- Опубликовано: 30 сен 2024
- ImageFX is an experimental technology that allows you to generate your own synthetic images. ImageFX is powered by Google’s Imagen 2 and uses Google DeepMind’s novel watermarking technology, SynthID to embed a digital watermark in the outputs.
▼ Link(s) From Today’s Video:
Google Imagen 2/ImageFX: aitestkitchen....
► MattVidPro Discord: / discord
► Follow Me on Twitter: / mattvidpro
-------------------------------------------------
▼ Extra Links of Interest:
✩ AI LINKS MASTER LIST: www.futurepedi...
✩ General AI Playlist: • General MattVidPro AI ...
✩ AI I use to edit videos: www.descript.c...
✩ Instagram: mattvidpro
✩ Tiktok: tiktok.com/@mattvidpro
✩ Second Channel: / @matt_pie
-------------------------------------------------
Thanks for watching Matt Video Productions! I make all sorts of videos here on RUclips! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!
All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.
-------------------------------------------------
► Business Contact: MattVidProSecond@gmail.com
I loved that it scolds you for trying to use its own suggestion 🤣
You had me until the prompts started breaking guidelines.
Open-source now!
Open-source tomorrow!
Open-source FOREVER!
sure, and how much of your time and money are you willing to put in? Zero? Thought so.
I love how every single harmless prompt goes against their guidelines
Yeah. These corpo models just make me roll my eyes. "Oh no, you put ugly or hideous in the prompt, that's illegal!"
@@Klokinator I get it all the time, even without a single word that should trigger it at all
I mean I don't wanna see people with two arms, that's too far.
Open source and locally deployable or gtfo
It's like it's made for children only, no adults allowed.
With all these guidelines I feel like the target audience for the entire tech industry are people who are about to graduate kindergarten.
Pretty sure kids in kindergarten could handle "ugly cat"
@@CoolhandLukeSkywalkrcan you explain please, how to use this font trick?
@@r_faizullin Use the font style called tiny, found from any font changer. It looks like this: ₜₐyₗₒᵣ.
@@r_faizullinLike I mentioned, use the font style named tiny in all your prompts.
@@r_faizullin Did you figure it out? Use tiny font. Lol. Still working.
I tried it and i can hardly generate anything because of the restrictions.
@pizzafication69But that's the thing, It's impossible 🤷♂️
Lol I tried "a clown" and it won't do it.
@@EddieWelch-hk8vs You're just not following them enough, clearly 🤷
@@xdumutlu5869 dude just admit that the censorship sucks lol these generators will never reach its full pointel unless these devs stop with the heavy censorship
@@doseofinformation211 you're not chronically online are you? That was sarcastic 🤷♂
just tried it, wow to many restrictions. hope they loosen up a bit like you said
They need to stop censoring image gen so much
no seriously the censorships destroys these generators why don't these devs see that
Like others have said it's too censored. I was trying to get A man walking down the beach but it tells me it couldn't do that. It's worse than bing. I'll stick to local stable diffusion for now.
We need more open source models.
sure, and how much of your time and money are you willing to put in? Zero? Thought so.
SDXL.
Man with two arms is against the policy😂Hilarious.
This looks good. The lack of seed numbers is the worst thing about DALL-E 3.
Actually DALL-E 3 has seed numbers and even displays them after the image generation if you ask ChatGPT to do so. But then it tells you that you can't use them for regeneration of similar images. Stupid 😅
@@joechip4822 It doesn't even show those details on Bing.
@@joechip4822 Bing Image Creator doesn't show a seed number.
@@joechip4822 On B-i-n-g Image Creator there are no seed numbers shown.
The model itself I can do without, but the interface is nice. Google should just sell this interface. The model is garbage, to be honest. I can do all this with existing models, and that censorship crap discoursed me completely from using it.
You're one of the few voices of reason on here.
Lol nearly every prompt "goes against their policies" lol. Even "man enjoying pizza in nyc". What a joke.
Censorship is so bad in ai I have cancelled every membership I had until they loosen up. It's unusable to me as is.
Hands, hands, hands, do more hands and fingers test
Looks wicked on realism. 💯 agree with you on the strangle hold. If they release it this way, it will crash and burn. The majority of users won't let Google control them in that way.
Mj v6 and Meta Imagine are both far more advanced as far as photorealistic ai goes. Even Dalle3 is better at photorealistic than imagen2 is. Tbh.
@@CoolhandLukeSkywalkr DallE 3 is not better at photorealism I can tell u that 💀
@@Wyn3e Until they fix the steps count on imagefx, dalle3 is actually slightly better on photorealistic. It kind of depends on your prompt, tho. Imagen2 is very plain. All it can do is human faces. It can't even create a guy holding a hotdog, etc.
@@Wyn3e What is your prompt? Try explosions. They are currently far more realistic on DallE3 than imagefx. I can give you many more examples.
@@Wyn3e Imagen2 can't even do a guy holding a guitar without slightly messing up the guitar. Etc.
Figures Google would have the most censored AI service. :/
And it sucks, too, because Imagen 2 actually seems really high-quality. I haven't been able to get Bowsers that accurate with any other model I've tried.
Maybe they should say "may only be used to make pictures of cats".
Reminds me of midjourney v4…not that impressive
Do the comparison with other AI's it would be amazing
I don't think this is Imagen 2. I used the same prompts from the other AI Google Image generator for a few months and copied those prompts into this "new" editor and it looks almost exactly the same. Not even close to the Imagen 2 tease they posted in December.
More lame censoring except on actually google searches. I like the drop the menu I hope a open source Interface like Automatic 1111 or Foooccus will support the word drop down very good feature.
If Google planning to charge for this image creator, they might want to work on this landmine censorship. Otherwise there customers are going to suffer the same fate as those from My Hero Academia from season 2. 😅
Man with 1 arm - okay, Man with 2 arms - violation of policy. What even word "human" means to google?
Policy is way to strict.
Very good for the realistic generation, but I feel google is still playing catch up unfortunately.
🎯 Key Takeaways for quick navigation:
00:00 🌐 *Introduction to Google's Image FX Interface*
- Introduces Image FX by Google in the AI Test Kitchen, highlighting its uniqueness in the AI image generation interface.
- Demonstrates the high-quality and realistic images generated by the model.
- Mentions the potential comparison with other models like Mid Journey and DALL-E 3.
01:09 🎨 *Creative and Exploratory Interface Features*
- Showcases the blue-highlighted words and dropdowns, enabling users to change aspects of the image prompt.
- Emphasizes the novel and exploratory nature of the interface, allowing creative interaction with the model.
- Notes the model's inclination towards photorealism over artistic drawings.
02:17 🧩 *Limited Settings and Seed Locking*
- Describes the limitation in settings, with only the seed being adjustable.
- Illustrates the seed locking feature for exploring prompts over time.
- Highlights the need for more refined settings control for finer detail adjustments.
05:32 🚫 *Restrictive Policies and Challenges with Prompts*
- Addresses the strict prompt policies, causing certain prompts to go against the rules.
- Expresses frustration with the over-restriction of the model, limiting its creative potential.
- Advocates for a more relaxed approach to allow users to explore the model's capabilities.
06:41 🎭 *Model's Strength: Generating Famous Characters*
- Showcases the model's impressive ability to generate famous characters like Sonic, Bowser, and Mario.
- Highlights coherent and realistic images of famous characters engaging in specific activities.
- Suggests that this use case is the most strong suit for the model.
10:32 🌍 *Community-Generated Images and Model Exploration*
- Shares community-generated images and mentions the challenges in getting access to the model.
- Acknowledges the model's potential improvement with more refinement steps.
- Recommends using the model for generating famous characters and realistic images.
12:18 🌐 *Accessing Image FX and Final Thoughts*
- Provides instructions on accessing Image FX through the AI Test Kitchen website.
- Mentions potential access limitations depending on the country.
- Shares personal thoughts on the model, praising its prompting feature and realistic image generation of famous characters.
Made with HARPA AI
I would definitely love an ImagenFX, midjourney & Dall-E 3 comparison with the same prompts
Finally its released
I genuinely think for general purposes, this is the best text-image model i've seen for photorealism and text.
It's awful. The images are extremely low quality. Compared to meta or mj or dall e3, it's a train wreck. I guess it's okay at cats, but it can't do people. Lol. I played around with this text to image generator and never got anything even remotely usable.
@@CoolhandLukeSkywalkr Are you acoustic? This is clearly not at anything above 5 steps lol. Most midjourney or dall-e generations take minutes with upto a hundred steps, while this takes less than 4 seconds. If they up the steps and processing time, it is very easy to see how much better this is than anything we've seen before.
It's clear google will release a version with a higher step count, better gpu etc for a paid plan on their phones and services.
@@RainbowSixIntel Even with higher steps, the actual generations are far behind the competition. You must not even really use these texts to image generators. You don't know what you're talking about. This one is a joke right now. You're a troll or something.
@@RainbowSixIntel If they release an update that includes everything you mentioned, they would have the 5th best text to image platform. They still wouldn't crack the top 4
@@CoolhandLukeSkywalkr I challenge you to get a nearly realistic image as one from imagefx using dall-e lol. keeping in mind its taking exponentially less time to generate.
You are honestly ignorantly annoying, steps are everything when it comes image generation (to a certain point), try generating an sdxl or sd 1.5 image with 5 steps, you'll see what I mean.
Pretty sure it misunderstood the word "arms". Like, "weapons'..?
Which is still absurd that the elites think weapons are "inappropriate" now. Or at least, they want them to be as part of their agenda.
8:42 BTW it's meant to be DALL·E 3, not DALL-E 3.
Sonic eats Chili Dogs, yo.
Works with a VPN here in the UK, selected a USA server :)
Too restrictive but nice quality
Too restricted, quality has a lot of catching up to do if they want to be competitive. Quality is shite.
garbage
In phew I almost forgot I got access to AI Test Kitchen and I can use it
Using this app we created goes against our policies...
No good in Canada ATM, finally getting some real milage outta my VPN..
I've found on most prompts you can hit the generate button a second (or third) time and it will accept the prompt
7:10 This is you can absolutly do in Copilot/Dalle-3 aka ChatGPT, by prompt "Generate an illustration of Sonic eating the chicken nuggets at mcdonalds, photorealistic style, please."
Thank you for your vids. I don’t think I’ll try Google until they loosen their « safe content » policy. This seems to be too annoying, for a quality that is not even « better » than what’s already available…
Matt, Yes We do ! Thank you 🙏😊🙏
Another cat generator.
i dont like it that violence is censored with A.I, violence is our way of life as humans, how are we to make movies with AI without showing violence ? people love action violence, AI should reflect what we are as humans, not tell us what reality we should have
stable diffusion level? 1:40 ?? Stable Diffusion is to day the best tool for image AI. You probably aren't using it the right way if you think its behind MJ or this. LOL
What day is it in 2035 February 5th? Monday or Tuesday.... Copilot and Grok thinks it's a Tuesday. I think its a Monday.
appreciate your review! It's shocking how bad this model is... Literally much worse than a decent fine-tune of SDXL. Reminds me more of Firefly than anything else.
It says "United States" only, but using a VPN set to a place in the US worked fine for me (I'm in Austria, Europe)
Fighting is banned. Like creating the following prompt of Demons fighting medieval clerics in a dungeon, cinematic, 8K
Sonic eating mc nuggets at McDonalds works in Dall-e, but it's creepy... and he is wearing Nike shoes for some reasons.
These guardrails reminds me of when Dall-E used to block all generations that mentioned gorillas. Well intentioned, but cautious to the point of being stupid.
This is why, if you can, making your own image generating AI is better.
There is many on Discord who can help show you how.
I can't get anything drawn. He always says politics, he can't even draw a picture of a girl, what should I do with this artificial intelligence?
ugly is against the policies...battle is against the policies...smelling bad llm here, overpopulated with a lot of filters.
It’s just insane how these tech companies have the luxury to be so strict. Nobody wants to fool with another PG model.
Prompt: Cat wearing a sombrero..
Google: “Prompt violates our policy on racism.”
It has potential but unless it's for simple photo realistic images I won't be using it at this time. Policy filter is totally out of whack plus anything creative isn't at par with other generators
Way too restrictive at this point. Also the graphics are glitchy. Eyes and hands look awful
“Definitely stable diffusion level”. You have no idea what you’re talking about.
Not available in my country…. yes, with a vpn it would be.. but…🥴🤷🏻♂️
Model lockdowns will be the downfall of the popularity of this google ai image generator (and dall e 3)
They said my name Robert Geczi was against theirTOS. lol
Hey Matt 👋 I just tried this out and love it. Did u know there's a create music section as well? I found it by accident lol. Tfs
Maybe it assumed a weapon instead of the plural of arms
oh look, another generator that blocks harmless prompts.
Whatever you do, flollow the policies!!
The restrictions totally turn me off.
google just so much ahead of everyone else.
The wave part at 3:59 reminds me of Boney M.'s album "Oceans of fantasy" where the group is also edited into the wave: ruclips.net/video/3Q7TsFQ3WkM/видео.html
Ridiculous restrictions. No thanks.
hey matt none of your discord links works
Everything is against their guidelines... this is just too much... They could atleast point out the prompt word with the problem, but instead you just have to guess the inapropriate word. rediculous.
Boobs are ok, guns are ok. Ethereal? "How dare you!" But yeah, the quality is amazing and ... I hope the guidelines will be more open in the future.
the fact that one arm is totally fine but 2 arms not 🤣
I tried it myself. It's cool but not the best.
bard is updated now 2024
Each time I try to generate an image it says "we couldn't create what you asked for" EVERY time. What is wrong with this piece of shit?
It never even says that the prompt goes against their policies.
I wonder if arms was blocked because of "arms" as in weapons. Interesting rule. Also really fascinating how it got from Dall E3 to Wall E. Did it prompt itself in the background, similar to Dall E3? Or did the model itself misinterpreted Dall E3 or even took it as a spelling mistake?
omg nooo a man with 2 arms?? how could you
Anything that restrictive is useless.
That dig against SD was uncalled for. SD, especially SDXL, can make much more beautiful (or ugly!) drawings of cats 😅
Not a very good ai still needs work
in Europe we don't even have gemini 😂
Geez❗ Prompt prison.
sadly not available in germany....
You're not missing anything. It's not even in the top 4 of text to image generators. I hear it works with VPN, but it's not worth the trouble.
I think when they release Lumiere it'll probably be with the same prompting template
Wow!
Not available in Japan yet 😢
not available in canada, let us know if you hear of any updates!
Yo matt can i get one of my gens in next time as one of your examples? I wanna feel special 😆
Apparently superhero like characters are two violent to generate
I am not using that
First
compare
Not working in Canada FYI
Works with VPN. Not worth using currently. It has garbage results.
👏👍💕🎆🎇
"Not available in your country." 🇬🇧
Use vpn
@@KH2KG I tried to use Tor to create a Google account 20 years ago. Google disabled the new account almost immediately. That was the last time I tried to use a proxy for Google services - because I learn my lessons. Unlike everyone on RUclips!
@@KH2KG A response was here 🚩. Deleted by RUclips. 🤐
It does music! nice.
😂😂 the Mario it came up with @8:10 kinda looks like you in a way... Lol
Yeah, lol
Not available yet in Turkiye!
call your president
Not allowed to use it in the UK.. 😭
It can be accessed with VPN
Wow thx