The Wait is Over! Gen-3 is OUT! - First Testing & Impressions
ฝัง
- เผยแพร่เมื่อ 8 ก.ค. 2024
- In this video, MattVidPro AI explores the newly released Runway Gen-3 Alpha, a next-gen AI video generator. Despite minor hiccups with complex prompts, the host demonstrates its powerful capabilities, generating various AI-created videos from simple to complex prompts. The video includes real-time generation, prompt enhancements, and community reactions, showcasing the potential and limitations of AI in video creation. Viewers are also invited to join a live stream for more interactive AI video generation.
▼ Link(s) From Today’s Video:
Gen 3: runwayml.com/blog/introducing...
Gen 3 Prompt Guide: help.runwayml.com/hc/en-us/ar...
Emad's Post: / 1807839613014159700
Dor Bros Post: / 1807782470164590825
Gen 3 Text: / 1807584743069790650
Always Generating: / 1807629033028407754
► MattVidPro Discord: / discord
► Follow Me on Twitter: / mattvidpro
► Buy me a Coffee! buymeacoffee.com/mattvidpro
-------------------------------------------------
▼ Extra Links of Interest:
AI LINKS MASTER LIST: www.futurepedia.io/
General AI Playlist: • General MattVidPro AI ...
AI I use to edit videos: www.descript.com/?lmref=nA4fDg
Instagram: mattvidpro
Tiktok: tiktok.com/@mattvidpro
Second Channel: / @matt_pie
Let's work together!
- For brand & sponsorship inquiries: tally.so/r/3xdz4E
- For all other business inquiries: mattvidpro@smoothmedia.co
Thanks for watching Matt Video Productions! I make all sorts of videos here on TH-cam! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!
All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.
Timestamps:
00:00 Introduction to Runway's Gen 3 Alpha
00:41 Exploring the Gen 3 Alpha Interface
01:00 First Video Generation Attempt
02:14 Complex Prompts and Their Results
06:29 Using AI to Enhance Prompts
09:06 New Prompt Ideas and Generations
14:15 Settings and Custom Presets
18:45 Community Reactions and Comparisons
21:27 Conclusion and Upcoming Live Stream - วิทยาศาสตร์และเทคโนโลยี
![[LIVE] : ONE ลุมพินี 69 | คู่เอก "กุหลาบดำ vs นาบิล"](http://i.ytimg.com/vi/PNM6U8e_HMo/mqdefault.jpg)
![กินอาหาร ถูก vs แพง!! จานละ 500,000 vs จานละ 50!! [Ver. 2024]](http://i.ytimg.com/vi/FF_HppKg4vs/mqdefault.jpg)
JOIN ME FOR A LIVESTREAM TOMORROW WHERE I GEN YOUR PROMPTS! th-cam.com/video/mksEAr6MRc0/w-d-xo.html&ab_channel=MattVidProAI
👋
Will if I remember! :3
Hollywood artists breathe a sigh of relief. Safe for another day
AI Tools will never take jobs.
Cope more. 😂😂😂
yes, gen 3 is pure trash lol
🤣🤣🤣🤣
The state AI is in right now, nobody has to worry. McDonalds recently dropped the AI from their ordering. LLMs hallucinate way too much, and the text to image and text to video stuff isn't even at the quality it would need to be. Also, most of the time, they don't follow directions exactly. But all those unskilled chuds on Reddit who thought they would be taking jobs and all of that learned a hard lesson. "We are the movie producers now!", "Artists will become irrelevant!". Yeah. They should just get a life.
Come on Matt, it's utter shit and you know it
Haha 🥳🎉
pure trash
I never sub to this weirdo I don't know why I'm getting his notifications his videos suck
Really disappointing. Yeah the quality has improved but the prompt understanding is still absolute crap, it ignores like 90% of my instructions. And you cannot import images so .. yeah .. you can make fun little clips by accident or whatever but I would not know how to be able to get what I have in my mind with such a bad prompt interpreter.
Luma is *slightly* better imo.
@@RockyBergen kling is the king
Yeah but I think it'll improve with the upcoming versions its too early to judge imo
Try using Gemini to refine the prompt.
@@dhruvmehta2377Is that not our hope with EVERY release of EVERY model. They just want to monetize so they put out junk before it’s ready like tooncrafter, and the rest of them.
conclusion : let's wait for GEN-4
nah, runway will always be behind everyone else, lets wait for sora 2 or kling 2 or midjourney video model
@@pragmata7997bro it's in alpha version wait fro the to release full version
its like runway is running away from being competition
It's useless, and its not free
Real question. What am I supposed to do with this? The footage is unusable except to look at and say "wow, neat". I'm not going to pay $15 for 60 seconds of semi-coherent video that isn't even what I asked for.
I am not paying just to just to test out the alpha version as well
We are getting so used about impressive things that we can’t see the potential of this things. This is a peace of engineering art. We should appreciate it a little more
@@matiascoco1999then be an investor.. but consumers have a right to point out the flaws. So for now I’ll appreciate from the sidelines
1.Music Videos 2.Sci-Fi Documentaries i.e UFOs, Paranormal, Ancient History 3.Educational for kids 4.Shits and giggles
@@matiascoco1999 "Potential" is great if you're not paying. I get that it's alpha but I paid $25 and didn't get a single gen that followed my short prompt.
"Imagine being able to play a video game where what you see is generated in real time". This is exactly what the CEO of Nvidia says is the future. Pixel by pixel generation processed by a GPU, no game assets, pre-recorded audio / video. Purely generated and I assume also stored in the background in case you wanted to "play" that particularly generated game, given each new generation would likely be wildly different and any story line would be also wildly different.
Luma has a huge advantage, being able to import midjourney/dalle images. Both are extremely expensive.
its not an advantage...runway jus hasint released it yet as u can see its in gen2 that img2img in gen 3 will be insane. if their text2img is lookign this good.
Then ill just have to wait the exorbitant prices until that feature is released! Then I can do comparisons of animated images. Would be an excellent test.
kling > all these shitty shits
Luma just added uploadable start and end image frame too. It interpolates from A->B. I've tested it a bit today and it's quite interesting.
I love your channel. I have been following you for years, and the thing I love most is that you are always quick to post news and always put links in the video description.
12 dollars a month for 62 seconds of video? laughable
I don't like any service where you can use up all your credits in a day and tell you to come back next month. I want daily credits with stuff like this
@@reifuTD seem like a bad business model to me.
62 seconds of pure trash and inconherence
12 dollars a month if you pay for a whole year upfront. Monthly it's 15
@@reifuTD Exactly, there's no way we should be paying that much for only 10 videos a month to generate that aren't even longer than 10 seconds. I need the offline local designers to step up and make their own Gen 3 and Sora AI for us, so we don't have to pay for anything.
This is very disappointing when you see the ad for it
Especially when it's so damn expensive
I honestly do not agree that this is the best we have access to. I feel like we got in general alot more coherent and usable shots with Luma from what I have seen so far.
It seems you can get some really good shots that look better than Luma, but those seems so incredible rare that there is no way you could do that without a unlimited plan and alot of rerolls. What I hope is that Kling releases their model to the world soon, as that one actually seems to be alot better than both of these video models.
I've been getting some great generations but the pricing is rough, nearly $15 for a minute of footage if everything you get is usable which is unlikely unless you get a more expensive plan or subscribe for a year.
get a more expensive plan which allows unlimited generations
@@businessmanager7670 Or wait for it to get cheaper, or if you wait long enough, maybe in a few years you might be able to run it on personal hardware like you can with image gens.
Yeah, for serious aspiring AI filmmakers, the unlimited plan isn't bad but at the cost of multiple Adobe subscriptions, it's not really approachable for someone not in the industry.
Seems like another let down vs the teaser videos they've been putting out
cherry picked examples for sure...
For sure
Can't forget the "the first thing you're probably wondering about is safety" thing in that reveal trailer💀
yup it's trash i cancelled my sub already, not only did almost 100% of my video get falsely flagged, the few that did work couldnt of been farther from the prompt. Im aware how to use these tools as I have over 100k generations in midjourney alone and know how to prompt Ai properly.
Smh you wont be missed just like anyone else who cries or troll over everything. Just let me know when you create such a device. I will cry beyond science..@@iamisobe
I think it would be interesting to use a suite of a.i tools.
Train a GPT on 3D objects, and use another GPT trained on skins and textures and use another GPT trained on animation. Put them all together and automate anything.
Instructions - if only it was able to follow them…
Thank you so much for the shoutout Matt! 🙌
waiting for open souce community to build up something to beat it
This is really exciting!!!
I've not only been testing gen-3 but as a producer and artist have been using it for real world purpose and content. My most recent music video I made cost me $30 and I think for the first time ever in ai video creation it's finally worth it. I'll be using it more if anyone wants to follow along on my channel
Thanks matt for not making the screen move while you hover your mouse. Thanks for listening. You can still zoom in just don't make it dance according to your mouse movement :)
:)
It's a trend to release semi-usable things, probably out of fear of the implications. Note that no closed or open source model (Stable video) has released the video to video function, although it is by far the most useful thing they could do.
Video-to-video would allow you to basically input short scenes of an actual movie into the model and recreate a full movie in another style. Like a CGI version of a live action TV show. That could be a copyright issue.
Yeah that's an interesting point. You thinking things like instant deep-fakes etc.?
@@maxington26 Sure. Furthermore, a video is the most versatile guide with fewer variables to compute, but looking at the examples I suspect that not even OpenAI AI has given common testers access to this option.
The left and right hand shaking XD
You need to upgrade to try it unfortunatelly
Yeah it's bullshit tbh not to mention I heard its not even that good and most of the videos were cherry picked
It's worth the $15
@@jmd489didn't look worth 15$
@@jmd489 ion know it seems like cherry picked the videos so idk if it's worth the $15
@@aiartcreation223 its pretty bad actually
Gen 3 is great if you're happy with whatever it makes. I ripped through a whole month PLUS paid-for added gens. Not one gen followed my prompt. I put 'treehouse' in the prompt 3 times and one only appeared in about 1/10 of my gens. I put in low-angle camera and it sometimes did that, sometimes did birds-eye. $25 and I didn't get a single gen that showed a treehouse at a low angle. I think it would help if they used a form rather than plain English. I tried LTX studio and it was 10x better. Photoshop-like fill and erase, automatic updating of the image, tons of free gens. The resolution is pretty low but then so is Runway.
It seems they couldn't have enough GPU power for a wide audience to use at the same time and get great results, like the ones they showed before.
LOL Matt it's so funny when you mentioned generating video game frames in real time because we already have that with RTX DLSS xD
DLSS is not generating frames it's upscaling frames.
@@vommnah, it's generating extra frames between the already existing ones.
But its not really generating a new frame from scratch to produce the next thing seen in scene, its just interpolating between 2 frames. Very different. One has no limits to what it could generate, the other is just filling in gaps. @@SW-fh7he
frame interpolation/upscaling isn't the same as real-time video generation
There is competition, You seem to be forgetting Luma Dream Machine. That seems to produce very similar quality videos
I personally have found luma to be worse in most cases
Quality seems a bit less in the best examples of cherry picked generations they showed from runway, but overall in real world usage the coherency of Luma seems alot better. Even with the free 30 credits i got at least a few good generations that did what I asked it to generate.
gen 3 is trash man
@@MattVidPro It is easy to say when you have unlimited credits as a partner. The average joe gets only 12 5 seconds generations per month for the paid plan.
With luma i get more than double that for free and with image 2 video as well as being able to extend videos further than 10 seconds max.
You showed off more generations than we get to play with using the standard plan and I did not see a single clip that was good enough to be used for anything and you did not even show us the worst generations.
@@bloxyman22 I showed you every generation that I made. I PAID for the mid tier plan.
I will wait the image to video feature
I wonder if using Gemini 1.5 Pro can fix the prompts to get better results?
🍋 bout to watch rn 🍿
The prompting coherency is awful, but the quality is pretty good
Kling seems to be better!
Every MAttVidPro video summarized: "Hey guys, this is one of the most important releases ever, this is gonna change the world forever!" *spends 15 min demonstrating how absolutely horrible the AI software is*
Still useful to watch
Yeah this is not what we thought it would be.
Is there any "video model" available for download and local execution, something like Ollama but for videos, to download models like Llama3 (Sora), Mistral (Runway), etc.?
I think the ideal solution would be an animation app that uses these models to create motion interpolation and provide more control.
It really looks like this is mostly trained on stock videos and struggles to extrapolate beyond that.
The alien handshake a perfect example. It's like it has never or rarely seen a handshake that isn't an extreme close up of two business men making a deal with a handshake, so all it can do is sort of overlay that video, or at least if there are 2 close up hands in the scene already, it cannot resist making them do the handshake.
its good to generate story
This requires camera movement skill
There should be a series of standard tests like walking, playing soccer, jumping, fighting, catching a butterfly, etc
Can't wait for Sora
lol... Ilya left OpenAI
@@manonamission2000 Ilya work in the the safety department
If ‘AI’ simply pulls stock footage and puts it together everything would look better.
I am so happy, that I've subscribed, and I can't wait to try it, Thanks for testing it and reviewing it for us !! I tried twice, and guess what? No more honey because Gen 3 Alpha is extremely expensive to subscribers, fair warning ! and I had to upgrade to Pro, to have a taste of it, and, summarizing, prepare your Wallet.
No image to video testing yet?
I hope you test some anime style generations tomorrow, So far I've seen a few Gen 3 attempts at anime and it seems to have a style that different from Luma.
what do you think of the monoliths when you consider AI? And other news things, like... other things...
The undeniable Emperor of Emperors with new clothes .
You can iteratively rate each generation for Claude to give better prompts.
If you use a workaround, you can actually use Kling as of today. I believe Kling was the top Sora competitor. Best part is, as of now, you can generate for free.
Only drawback is that it's mobile only and you have to use a translator if you don't understand Chinese.
Much better than Gen 2, but comparing quality, and what is available, Luma Labs looks better to me.
"A machine here!"
can use it for as many credits as you have tho.... contemplating the unlimited option
It seems like the video quality keeps improving but the AI's ability to generate what we ask for is stuck in a rut. I'm waiting for Runway to add image to video with Gen-3 but I'm sure the experience will be as frustrating as it is with Luma Dream Machine. At least Runway has an Unlimited plan (for $$$).
the future of indie horror videos, games and other things is legit.
I tried it, it’s far too restrictive
Which generators have people used to make the stuff I've seen all over TH-cam? The Family Guy shit and all the trippy AI fever dream videos? Pooky Park?
Whenever I generate an AI video, it's not that great. Which website or tool?
In the alien generation the forest looks great.
It requires Upgrade. I can only use Gen-2 for free, but 3 isn't free. Is this the case for everyone ?
BTW, LUMA does better job by generating videos and it is free right now.
LUMA is free until your burn through your credits.
Runway 3 is paid access only atm, for anyone, yep.
With the current parameter scale, if you want to achieve the best results, definitely need to try so many times.
I'll bet sora comes out real soon. Or they wait to hear a little feedback on Runway G3
Hold your breath and see how soon Sora comes out.
If open AI releases Sora , and you can have access to generate videos at fair price, Gen 3 will lose strength!
LETS GOOOOO TIME FOR GEN 3 🙌🏾
Not sure you are right about the pricing. There are several AI companies starting to crackle - because its far too expensive running models versus their income (Stable Diffusion for ex). If anything it may actually become more expensive - as newer models will be far more expensive to train and that's not accounting for how much it costs to render each shot. We shall see.
I respectfully disagree that there's no competition in this space. While Runway Gen-3 has its strengths, it may have limitations for certain creative purposes, especially without image or video input.
Kling offers some interesting features, including image-to-video capabilities, and it's publicly available at no cost. Based on the footage I've seen, I'd suggest that Kling might even have some advantages over Sora, particularly in terms of reducing visual artifacts like morphing effects. And as we know, Sora is still not available.
Is it really better than Luma Dream Machine tho? I think it's on par.
They are about the same. People say Luma is the best but you might get 1 okay generation out of every 10 generations you generate on Luma. Luma's pricing is also a mess.
@@qwax I think the image-to-video feature really improves it. Runway Gen-3 needs that.
@@ThomasJDavis Nah. Luma's image to video feature is mediocre. Believe me. I've run a ton of gens. I haven't tried Kling AI. I'm sure IMG to VID is coming. They probably released gen 3 to counter Luma because people are acting like Luma is this perfect thing when people who have used it will tell you, you get more bad generations than you do decent ones.
@@qwax I've used Luma as well. Image-to-video creates much more consistent and coherent animations as long as there aren't thin, long objects it has to animate. Same goes for Haiper and Runway Gen-2.
do luma vs runway
I haven't try the paid version of luma, luma or gen 3?
The "paid" version of Luma is exactly the same as the "free version" which is just credits they give new people. The problem with Luma is their pricing is ridiculous. If you aren't on a paid plan you can only generate 5 creations a day due to demand, other then that, no difference between paid and the credits they give you. The people complaining probably use a VPN and just keep generating trial accounts on Luma.
@@qwax You say the pricing of Luma is ridiculous, because the free plan is slow? That doesn't make any sense!
You get over 10 mins of gens in Luma for $30. You get just over 60 seconds in Runway 3 for $15. You do the math(s)
@@maxington26 Bruh, with Runway you can just pay the yearly or monthly unlimited and just make as much as you want. Do the math bruh.
10:54 The left hand has 8 fingers, no less.
$144 per yr. Well worth it! Wish you got more gen seconds tho😢
That’s so realistic. But I can’t use Gen 3 alpha because it wants me to buy some plans but I don’t have my OWN money for this. That’s cool!
Gen 3 disappoints already, who wouldve have thought
Another few more months just wair
@@kevinsm2039 sora will be the number one
In what way? explain
@@BroskiPlays nah, sora is outdate already we need something actually good, maybe a midjourney video model or pika 2.0 idk, but lets wait
I make video art for over a decade and have the unlimited plan. I was able to prompt mind blowing video art styles in live visuals and vjing all day, insane visuals that I’m really into personally. I am also a music artist working on a new music video and I was able to generate the scenes that I needed for my music video.
lol random handshake
I like how the lady of the Sora demo became a reference, while Sora may be a horrible AI video generator as it hasn't been released yet.
I wonder how do people who know how good Sora is are feeling. Of course I don't know, and Sora might be great, but I'm just thinking...
does not seem like the Sora kids need to be worried about ANYTHING from Runway
The Sora kids that will never have access to Sora? You know they probably cherry picked the very best videos and Sora probably isn't that great. OpenAI loves to hype stuff up.
ilya left OpenAI... you really think Sora will see the light of day?
@@manonamission2000 I mean, one only has to look at a random Stable Diffusion discord to see what the majority of those chuds are using it for. I wouldn't release it either. These are the people are giving AI as a whole a bad rap. AI will never be praised for all the good things its doing but you sure here about all the bad things people are doing with it.
From what I see is the videos look good but Gen 3 has a real lack of understands what's it's doing so it has a hard time giving what's asked. Also they don't' seem to have Image to Video up yet?
Oh well, I guess we'll all just need to wait a few decades for Sora to be public instead
uh... it aint free to use
That stinks you have to have an upgraded account to try Gen 3 😢
this thing is trash at listening to your prompts. looks pretty high quality though
It's too much money for what you get. How many credits does each request take?
Try prompting it, alien shakes hands with human. Let's see if it shakes it's own hand. : )
How can they charge for that? It's like going to a dentist and asking for a root canal and waking up with a boob job and dentures stuck to your forehead.
Pls go live and test pls
Luma doesn’t look so crap now. Runway also crap. 😂
very expensive
It is pretty impressive and disappointing. It is really incapable of understanding a bit more complex prompts.
Pretty poor results imo
I've just made a music video on my channel using Gen3, time lapse flowers, which worked really well. Please check it out. A lot of my ideas using people really weren't working, but this simple idea worked well.
That Minecraft generation makes me want a Minecraft-like video game that's real-looking. There kind of are some, but Rust is apparently filled with dickheads, and Rising World is almost impossible to play once you find a tiger.
Bro, these comments have me dead 😂😂☠️💀
really? 1st time? and Luma?
I just tried about 10 different generations and I am really not happy with my results. If it had image to video, maybe it would be better but the marketing results where HUGELY cherry picked based on what I am seeing pop out.
Wait, but it is not free, it is under a paid subscription plan. 😅How it is then better than Sora?
toooooooo expensive!!!!
I do like it but their business model is awful, as this is quite expensive. For those of us who generate hundreds and maybe thousands of images a month know that it takes a lot of generations to get this AI stuff right. To only generate 10 images a month under the most basic plan is just not going to cut it. Not even close for us! The videos look amazing, still really buggy and the prompts are inaccurate, and they have no image-to-video like Luma does. I think we're at the point now where the community should get to it on a local offline version like Fooocus and Stable Diffusion to help us generate these videos.
It sucks
does anyone realize how much we will miss early ai and their thousand finger hands and other naturally sacary things? you will miss the fear.
Not very impressed. I would never pay for something that doesn't generate accurately.. Don't understand your excitement.
Bruh why is it not free
It's dreadful. Matt Wolfe is saying the same thing over on his channel - that's its incredible(!)
Runway must be sponsoring them, there's no other explanation