Pika 1.5 Explodes, Kling Talks, & Minimax Sees!
HTML-код
- Опубликовано: 2 окт 2024
- Big AI Video Updates! Pika just dropped their 1.5 update, and we’re diving into everything you need to know, from new generation models to hilarious new features like squishify and cakeify! Plus, AI Lipsync updates from Kling, and a new platform on the horizon-- AND, I’m giving you an early look at an exciting Minimax feature that’s is due to drop!
🔔 Chapter Breaks:
00:00 Intro -
00:22 Pika’s Free Tier is BACK! (And the Hardware Struggles)
01:11 Image to Video: The Good, The Bad & The Skater
02:17 Hand vs. Teeth - AI’s Latest Struggles
02:48 Camera Controls: What’s Missing?
03:20 Silly Effects: Squish, Explode, and Melt Away!
04:38 Why Not Upgrade to Pika’s Paid Plans (Yet)
05:22 Kling’s New Lip Sync Feature
7:07 - Community Kling Outputs
07:58 Lip Dub: How it’s Breaking Barriers in AI Video!
8:51 - LipDub w/ Live Action Footage is Amazing
09:56 Minimax News - English Site & Image-to-Video Tease
🔗 Links Mentioned:
• Pika 1.5 pika.art/
• Minimax (English): hailuoai.video/
• Explore Kling 1.5: klingai.com
• Lipdub Beta: www.lipdub.ai/
📢 Subscribe for more AI video breakdowns, tips, and exciting updates from the world of AI!
Thanks for featuring my video, Tim. ;) Yeah, i have my Pika page full of videos still waiting to be generated...for about 14 hours. :) But hey, we waited so much for Pika come back, i think we can wait a little while more. :)
Oh, great to see you in the comments! I'll admit, I was like: Eh, it's fun-- but I'm not sure what I'm goi---OH, Alex just nailed it. That's what you can use it for!
Great work!!
@@TheoreticallyMedia Thanks, Tim. Really looking forward to have some more fun with Pika 1.5 as soon as the problems with the generation times are over. ;)
As a french man myself, I'd say that hesounds like a guy from Montreal. Strong Quebec accent.
EXACTLY
exactly! (and yeps, watch The Bear) (in whatever language)
So, in a way he's right.
SO good! I know Season 3 got some grief for being slow, but man-- still really powerful!
Yes French accent from North America, i.e. Québécois
Thanks for including my MiniMax clip, Tim!
Great overview and examples of Pika 1.5 and examples using Lipdub. I signed up on the waitlist after seeing Ryan's video yesterday😂
Awesome! And stellar work, Heather-- for a split second, I thought it was a loop video-- and that gives me an idea!
Bradhuntz here, appreciate you featuring my video, love your content tim, longtime follower here
For just generating and hoping for correct mouth movements, you did pretty good. That sounds like a nightmare to do.
Oh, it was pretty insane. There was also a lot of speed ramping and chopping there as well. Luckily, at that time, I was one of the few in the US who had access to Kling, so I could generate quickly like a madman. I'd have like 5 going at once!
@@TheoreticallyMedia 😂
I wouldn't say the lip sync is "Photo Realistic." It's as good as a lot of game CGI. An incredible improvement though
Yeah, agreed-- I think in the original Dead Sea (Pirate movie) someone said around Playstation 2 level? Maybe the tail end of that generation-- Getting there for sure, and at this point: I think it's safe to say, we can so cinematic scenes...maybe not the highest level of it, but enough that if you've got a good story?
The accent is pretty similar to Canada's / Québec french accent
That’s really interesting. I wonder if that was intentional? Wife actually got to go to Quebec on a work trip recently- I was jealous as I’ve always wanted to visit.
…but, when I say “recently” i mean February, and when I saw the temperature I wasn’t jealous anymore!
@@TheoreticallyMedia It's French canadian accent, although some words he said like "toujours" really sounds a person with an english accent, not Québécois.
You should come in December during christmas time. It's magical up there especially in Quebec City :) .
Kling is still King but MiniMax looks like it could end them all
The accent on the French was heavyer than the VRAM draw on my Graphics card.
For the the algo ✊ Great round up Tim, thanks for the feature this month will be 🔥
I'm not going to sleep at all this month! Oh, and I'm going to Central Europe this month? I'm REALLY not going to sleep...ugh.
Kling AI is still the best. I'm still waiting for a feature that allows us to input a middle-frame or multiple frames (like Tooncrafter's Sparse sketch guidance feature) so we can have more control over the output. So far, Kling's end-frame and motion brush feature is the best but we need more control for img2vid to be production ready
Funny side-effect from the lip sync: from 6:55 onward the back of the pirate to the left looks like a face.
Still not as disturbing as those squishing images though, or the Indiana Jones: Raiders of the lost Ark melting process...
Haha, I didn't notice that! I love AI Ghosts!
It’s kind of a Québécois from the Montreal area how has spent a lot of time speaking only English and now speaks French in this scene. But it shows that he had spoken French for a very long time.
Great rundown, thanks for all the insights and links. Much appreciated 🤘 Kling and Minimax in particular look super-interesting 📽
It's Canadian French (Quebecois). But pretty good dub, better than the Americans do on foreign films for sure (have yet to see a foreign to US EN well made dub, saylike EU countries do on English movies)
It"s french with a Quebecois accent :)
Thank you for making such good content ! 🙏
Thank you so much! And thank you for chiming in! I knew I could count on you guys!
@@TheoreticallyMedia Thanks again from Belgium !
🔔 Chapter Breaks:
00:00 Intro -
00:22 Pika’s Free Tier is BACK! (And the Hardware Struggles)
01:11 Image to Video: The Good, The Bad & The Skater
02:17 Hand vs. Teeth - AI’s Latest Struggles
02:48 Camera Controls: What’s Missing?
03:20 Silly Effects: Squish, Explode, and Melt Away!
04:38 Why Not Upgrade to Pika’s Paid Plans (Yet)
05:22 Kling’s New Lip Sync Feature
7:07 - Community Kling Outputs
07:58 Lip Dub: How it’s Breaking Barriers in AI Video!
8:51 - LipDub w/ Live Action Footage is Amazing
09:56 Minimax News - English Site & Image-to-Video Tease
The french at shifted to French Canadian (my native) to an american english trying to speak french. It's the first thing I noticed, right before you mentioned it.
how the heck do you release these videos so fast. Even if you have a team, this is seriously impressive
Haha. The “team” is me and the dog. The dog isn’t very helpful!
I actually think it’s because I work solo- so, I don’t have a round trip of sending anything out to an editor. Basically, wake up, see what’s happening, learn/play with the tool, start shooting and editing, thumbnail and post!
L(A)IP Sync really seems to be the flavour of the week following on from all the podcast stuff huh! The next Frontier. Great vid as always! :)
As a French speaker, it's pretty good but still room there for improvement (i dnt have the English clip as ref though so thats just to a blind ear!)
100%! For sure a big breakthrough in a few of the models. Whenever I see a pretty major advancement (like Emotalker months and months ago), I'm always on the lookout for a version that really nails it a few months later. That seems to be the trend: Something kind of works, and then a few reverse engineer and improve, launch with a big splash-- followed by a LOT of iterations on it.
Happy to have it though: this has been much needed!
Sora is still the cheapest and the one with less customer complaints. 🤣😂
Haha- You got me for a second there!
is sora is available now?
@@mdmafoz8513 No, he was joking.
did you get some framerate drop when using lip sync? my video looks very chopped before lipsync
I always run stuff through Topaz for final-- but yeah, you see a little stuttering here and there. A good old Topaz wash tends to clean that up right away.
Makes me wonder, by what metric will we dub AI Video 3rd gen? Progress is occurring really quickly at the moment, will we only see incremental gains from here on out, or will there be another model that will reset the bar?
great video!! and many many thanks for the shoutout :)
10000% Tech!! And keep crushing it! (or blowing it up...or whatever else Pika has put in there! haha)
@@TheoreticallyMedia "hair it" would work for me :D
Mini max about to clear the room 💯🔥
yes, looks good, but I dont know Pika I left using before, as I never got good results
hi. is pika the current best image to video service available to the public?
No, kling is the best currently.
Yer the servers at PIKA got hit hard :)
They’re taking on water to be sure, Cap’in!
I tried to use kling and bought the credits and it still wouldn't let me generate anything. I even provided screenshots and emailed the company.
Every time I try generating it was just provided me errors. Hoping I could get fixed soon
brilliant as usual Tim
I tested Pika 1.5 all the generations. 10 hours and not a single one is done
Yeah, I think that's everyone right now. Even on the paid tier. I'd say give it a few days to a week. I imagine there is a smoking crater where the server used to be!
9:22 French-Canadian with an American accent... Why all these AIs have a French-Canadian French is beyond me!
Great video! and it's a Canadian french Accent! 🥰
But what will happen when internet is just full of AI generated video, image and text. What do they train AI with then? I've read that it will cause a detoriating in the AI quality, on each time it's trained on it's own outputs. Even if they were flawless in the beginning, it will gradually worsen. Sounds to me like people who scraped Internet first have the advantage. But even they will run quickly out of data, and internet will be, in many ways, dead.
I heard a podcast where an AI expert said that by the end of 2025, around 90% of internet content will be AI generated. If it happens quite so quickly, not sure, but eventually it will happen. For example Google Image Search results are heavily AI polluted already, people with extra fingers etc. And the ratio is worsening each day.
Do we get a situation where the AI peaks pretty quickly, and then hits a wall, because the internet (it's main training data source) becomes polluted with AI content, and unusable as training data. Eventually we won't have either AI or internet, and have to go back to basics and doing art, music and video manually ourselves. (And promoting it offline)
I mean, you're underestimating that people will continue to make videos. I think we'll run through a period of junky AI stuff (some argue that we're here already)-- but, eyes will eventually adjust and folks will spot it as low effort.
There will be filmmakers who use this to create amazing stories (keyword: stories) and those will rise to the top. People will still make traditional films, and those will be awesome as well.
As far as training data: I mean-- I think the move will start toward more of a 3d model anyhow. Have you seen what games are looking like these days? Think that, but AI.
Less training on movement/characters-- more (near) realtime 3d model rendering.
They need to deliver something with quality or the interest will drop fast.
@@TheoreticallyMedia Yes I'm fully expecting 3D to be next. But overall I don't mean just video, but the overall quality of the Internet training data. The reseachers did an experiment where they trained an AI with wikipedia data about one specific subject (I think it was history of church architecture or something like that). At first the AI produced flawless replies, when asked about the topic. But after it was trained on it's own answers a couple a times, it's replies quickly detoriated into full gibberish that didn't make any sense. This happened only after a few training rounds (always feeding it it's own previous version outputs).
Of course if companies can feed their AIs good data, and make an isolated AI that isn't trained on internet data, that will always work. But somebody has to create that good quality input first (human labor)
@@shredd5705 it won't need an endless stream of novel data
@@tuckerbugeater Perhaps not. But the internet will be polluted. They will have to use the scraped data from 2022 (or so) until the end of time. And try to make better use of it. Or use local/handpicked data, that isn't randomly scraped from the internet
The french dub was def someone english speaking french
Nah, it is from Quebec
it's french canadian not french. But cool
Ahhh, Merci! Which is about as far as I get in both Canada and France!
French is a Quebec accent
Not french, definitely an accent from Québec ;)
The guy had a Quebecois accent, but spoke like an anglophone speaking french... and it a bit gibberish at the end.. does not make sense...
Ahhh- good to know! See, I knew I could count on you guys!! Me, I'm like: Yup, sounds french!
Thanks for the video, Tim. AI video needs to deliver something useful. Today it's just an expensive toy. I subscribed to Runway and asked for a refund an hour later. Full of inconsistencies and weirdness. Totally useless to make something serious. But developers are making fast cash with the hype.
I've played with the Runway free version. Gen-3 turbo is impressive BUT you never quite know what you get. And sometimes it's just nonsense. And the credits are expensive as hell. So I never wanted to pay. Overall I don't like AI but I'm trying to get on with the times, and admittedly on some level it's fascinating, mostly just scary though
@@shredd5705 I want to make my series. I already have 3 episodes done with 3D. I'll keep using this technology, maybe using 2.5D AI scenarios. There's no free lunch or free movie.
I would consider a different thumbnail. It doesn’t match the content
Yeah, I'll probably futz with it later-- I was in a rush to pick up the kids from Field Hockey practice. Haha, life on RUclips! Good note though!
Awesome. Maybe Instead of doing an English speaking show into French. Do a French speaking show into English... Then we'll get a true idea.
oh, that's super valid. When I get access I will do that for sure! Oh, y'know what-- I might try German with DARK. That show was amazing, and too few people watched it because they hate subtitles or dubs.
But man...missing out.
@@TheoreticallyMedia I'll keep an eye out for it. The tech is super promising for opening up shows like that. I still haven't seen Squid Games ha.
It’s not proper French but Québécois with an American accent
Not First!
winner of the chicken pot pie!
Can't tell if these videos are sponsored or you are just blatantly ignoring Kling's inability to consistently make a video instead of just spinning for literally days.
I always disclose if sponsored, so not here. I’ve heard about people having problems with Kling, but haven’t really run into it. Granted, I am on a paid plan?
paid plans and you wont get stuck 99%
i go with Tim on that - i'm on paid plan in Kling too, and never had this problem. With the Luma free tier though, one time for a 3 seconds looped video it took 3 days to generate!!!
2:31 The overt bigotry of the brits and our aching dentures!!!, i think we need to have a word with our prime minister... revoke 1776!!! we want our tea taxes! 300 yrs backpay!
Nope! Me first!
Sorry, Fizzy beat you by about 30 seconds! NO CHICKEN DINNER!!
first.2 ;)
first
The record states that you are the winner for the chicken dinner!
4th
Their logo is a Pika - a small animal - and probably pronounced the same: Pie-ka