Regarding the camera rotation, I've found "cinematic orbit shot" works quite well. That being said, while the camera movements are definitely impressive and so useful for actual projects, the control is still a bit limited, as you say. I gave Luma that as feedback because it's critical when you don't just want a bunch of random shots and a more fluid sequence of clips.
Ah, I was wondering about that! You got some GREAT Orbits going in your film! Did I see some tricks with zooming in on a still image before letting Luma take over as well? Agreed on controls for movement! Would love to keep it simple still though. Part of me was really relieved by the simplicity of just a prompt box!
@@TheoreticallyMedia I don't really do a lot of pure text-to-video because it's useless for most of the stuff I need to work on where character consistency is key. But I've noticed that you get a lot less of the random takeovers from the model if you use an input image and then use fairly simple prompts. Kinda like old school "partial phrases" like we used to do in MJ. PS: Another interesting keyword to play with is "dolly shot" or "tracking shot"....of course, assuming that's what you actually want to do. PPS: Those "tricks" you mentioned are actually 2 different generations of the same input image, but one reversed. If you stick them together... that's the effect you get 😉
I have tested several mainstream AI video generators, and most of them still have a long way to go, especially when it comes to character details, particularly facial features. Although Dream Machine retains most details at the beginning, it still loses some when the character starts moving. However, the overall effect is really impressive. I'm truly grateful that you could share such a great tool. I've been waiting for this for so long.
Thanks! As a young comic book/Stephen King fan, obviously Among The Living hit pretty hard for me! Scott was SUPER nice as well. The guy is the real deal! On the Dream Theater front: I have a video today that's going to make you pretty happy!
I experienced the same issue with trying to have the camera rotate around the subject, zoom in or out works really well as does pan slowly to one side or the other
Just so everyone is aware: anything that’s actually released Is automatically better than Sora. Sora doesn’t exist until it’s actually released as far as I’m Concerned.
Reminder that Pika Labs is still a lot of fun. Does cost money for the website version, but they do have some unlimited use paid plans (if these models end up charging per use instead of unlimited when they go pay). Waiting to see what Midjourney does for video.
Pika is still a TON of fun-- and I'm sure will be releasing an update that is similar to this as well. Once they drop, usually the rest aren't far behind! As for MJ...yup, can't WAIT to see that as well!
@@TheoreticallyMedia right now, I'm working on reconstructions of old, deleted TV programs (Doctor Who) where the originals are lost, but audio and stills exist... Dream Factory might be really useful here - as right now I'm using everything from animatediff to viggle, to deepfakes and lipsync tools)
I always get so excited when I see you post a new video. You always have the best information! I've made a ton of videos from your Haiper AI recommendation and am off to try Dream Machine now. Thank you so much for your great videos. 😁
Thanks, just tried it and it's probably the best image-to-video platform right now. Amazing job with face and anatomy consistency! However, I'm also looking for prompt consistency. Looking forward to creating the real dream machine.
I consider this a big first step toward getting there. I think we’ll see some massive advancements by the end of the year. November last year had some pretty big announcements, and I expect to see the same cycle this year.
Tried it out immediately. Starting from a picture it works like a charm, even keeping art style consistency (when the style is obviously anime). Now that it is day in Europe, and everybody got the news the queues are clocking up, and I don't get any more pictures...
Interesting. They must have pushed something out. I noticed yesterday if I had enhance on, the results were super insane. I think they might be dialing things back a bit,
I've got the best results where I've two photos taken soon after each other. You can supply both of those, and it will create a fill between those two scenes. A lot of the time, you get just a zoom. Sometimes a pan, and with a group of people maybe one or two seem to be alive while the rest stand still.
Ha! That camera control can do some weird stuff! Eventually we’ll have better control over them. But in the meantime, I’ve found some pretty good results by toggling the enhance prompt on and off.
Excellent video, thanks. I just tried Luna and I'm not super excited about the results to be honest. It response to parts of my prompts and not others. I tried starting with existing photos and very detailed prompts and all I got were fades and pans, not real animation. When I tried it with text only and a detailed prompt, it was ok, but not totally faithful to the text I entered. AI still has a ways to go.
It's pretty awesome. You're going to have a BLAST with this! Mind is boggling thinking about what someone super skilled with After Effects is going to do as well!
A good first attempt. Still way to inconsistent and weird for my liking to jump in and use it for any project. It's like Midjourney - i wasn't really interested in it until the results yielded acceptable quality. Doesn't mean you shouldn't keep an eye on it or play with it of course. Let's hope more competition will get accessible soon and they push each other to way better quality in future versions.
For sure! I’d call this maybe MJv2 or 3. But, v4 was the BIG jump, so I don’t think we’re far off. Also, now that we’ve seen this, I expect it won’t be long until we get similar leaps from Runway and Pika. And likely few others that we don’t even know about yet!
Yeah, this looks amazing. I've been so busy working on my Apple video that I missed this entirely. Also, you probably already noticed or someone else said something already, but I sure hope the resolution isn't 128 by 720 ;)
Do you ever review AI tools to take existing video and change it's style? (Or is there something even like that out there?). I'd love to see something that could take my 3D animations and make them into a 2D style.
I would very much like to see an extended tutorial with a few student tests. Also, the was one time when you put together a short horror movie. Maybe that would be another cool video, so that we can consider how to create better sequences. I love movies and I am able to remember some fairly long plane sequences from quite a few (I might be able to replay the entire Casablanca in my mind, albeit I can't remember all the dialogue)... But I haven't been able to create any interesting short movie yet. Thanks, Tim !
I feel like this type of AI won't be used as much other than for proof of concept, precisely because of decoherence and just the overall lack of control you have for details. It's probably going to follow the same road 3D printing has taken. It hasn't really disrupted production lines for factories, it just works well for proof of concept and that's about it.
maybe not, it will probably take the route like ToonCrafter, if this would be used i see it being used like interpolation between frames, where you can guide it so it doesnt get lost and stuff like that, just look at the ToonCrafter stuff and you will see what i mean.
Are you kidding me? The photo to video finally allows for some consistency between AI videos. That is huge and will only get better. An artists could make an animated short with a simple sketch
It would be cool if they would incorporate something like Runway’s *‘Multi Motion Brush’* feature. This would be useful for animating multiple different elements from the image, and for posing multiple characters together.
Wow I own a small business. Uploaded a photo of my logo then brought it to life. Amazing! Gonna Use it. Also I sketch for fun. Animated a few of my favorites. I am loving this! This photo to video tech will finally allow for some cohesiveness in AI videos
Pretty incredible. Definitely gonna be using it in my previs. Tho as a film maker, please reconsider demonstrating using copyright material to create new work. Something about using someone’s hard won work without any sort of disclaimer or caveat feels like a dangerous precedent. Love your work and know your heart is in the right place! Thank you for all you do!
Just going through a few image to video generations with a simple prompt gave some tremendously impressive results. Compared to others like Haiper & Pixverse Luma is an order of magnitude better. The videos pack so much depth and motion in to a five second clip. I would call it cinematic. It's not perfect, but it does so much with so little input, it's incredible.
Welll, it’s really cool…but I have 5 clips that started by saying they will be ready in 30 minutes…and after two hours they are still being created. Definitely not ready just yet. I believe the early testers had a fast priority…but for all of the others it is still not good. Also I got to make a few videos and had very mixed results. So, when I see the prices for the subscription they are super high because you will have to repeat many times to get a good shot…loosing a lot of the credits. This seems an amazing tool but still not ready for prime time until they work on the speed and pricing. But…I have great expectations for it. Thanks Tim for a great video as usual. ;)
Yup! If it’s any consolation I got in on Sunday (pre launch) and was able to generate super fast, but as soon as the doors opened up, I’m basically in the molasses with everyone else. When it works, it does well. Rerolling? Sure, but that’s always a part of AIVideo. As far as pricing goes, yeah- I’m not sure many will hop up to the pro tier. That said, if I’ve learned anything over the last two years it’s that when one big leap happens, the others aren’t far behind. I’m looking for big announcements from Runway and Pika soon.
@@TheoreticallyMedia yeah. I’m with you on that. I’m also on the ai video bandwagon since beta testing runway gen-1 and a lot has happen since then (really fun ride). I do believe it’s just a question of time (and it will be fast) till we have close to Hollywood image and movement quality on our hands to make a movie. Im rooting on pika because I believe they have a really fast video creation tool and a great Ui. I don’t know if you noticed but they improved their model in the last few days …it’s closer too Haiper I believe (which I think until now is the best quality…if we forget about dream machine, sora or kling). We are getting there, Tim! ;)
Tried to try Dream Machine and won't let me do it. I get this: Application error: a client-side exception has occurred (see the browser console for more information).
Yeah, I started 4 generations an hour ago, still nothing but the weird thing is, they don't have a timer or anything that suggests the content is still being created. I don't mind waiting, I mind not knowing... but I dare say they just weren't expecting the traffic.
Wow, this is a big step. I like how it animates pictures, not perfect but so much potential already! Still need the lip sync feature for creating actual music video's or movie dialogue.
One of my favorite AI content creators.. I am currently playing with this and will be Paying for it in the future to it as I really see the potential :)
They're getting SLAMMED right now. I'm sure there is a smoking crater where the servers used to be. I'd give it a few hours to let things cool down! The old Hug of Death!
Fascinating. I'm always happy when I see mangled fingers because the rate AI has been progressing is scary and I want to think we still have ways of knowing it's fake.
Having so much fun messing around with uploading my own images to animate. Downside is that there is no way to save the video. Sure I can screen record but wish there was a download option
Ah, excellent to hear! Glad to have you here!! Actually, have another pretty cool one coming up tomorrow! LTX Studios...but, with a music twist! It's actually one of the best videos I think I've made yet!
Wow, thanks for such a quick report and update on this, great content! And I couldn't help but think, the decoherence and morphiness - that's probably all possible to clean up using image generation post-processing, frame-by-frame a.i. automation to get something cinematic, even if the initial results have some wonkiness. *post comment edit - I tried it out and looks great, seems like they're taking the Midjourney approach of allowing enough usage to get people understanding the unique value - but a paid usage model, which is a great way to get people on board.
Great Chanel and great video. I started using dream-machine and yes lots of potential. But certainly dealing with human motions movements need serious work. If you making short videos without any human or any movement the generated videos are stunning.
Wit these ai video things how would you be able ta keep the same character and add them into the next scene? Like where there just one or the other ta maybe having both in the same scene and probably reusing the same character in other areas like instead of being at the beach their at a theme park wit their friends. And ta make that same character talk in that scene or different scene? I have alot of ideas and characters fur movies or shows I'd wuf ta do.
This model is “definitely on a Sora level”? Come on man. Are you joking? This is not even in remotely the same ballpark as Sora. It’s still impressive, all of this stuff is. But let’s not pretend this is anywhere close to Sora’s quality level.
@@injectionAI I have no idea what that’s supposed to mean. This is an ML model that is trained from data using gradient descent-it’s not a product of software developers. Also FWIW, I’m an AI engineer and I work with software devs on a daily basis. So I don’t think it’s because I don’t know how devs work.
100%! That's exactly what I kept thinking when I was using the text to video stuff-- I haven't had this much fun since early Pika! And, when you get the janky outputs, I mean-- Comedy Gold!
@@TheoreticallyMedia Totally superfun! Meanwhile, can we negative prompt in Dream Machine? I am reluctant to try with my free account. Please tell me ObiWan.
Not sure why this is being compared to Sora - the Sora videos I've seen have been sharp and clear, and this is fuzzy and very unsharp, and lacking the clarity and sharpness of Sora videos. True, unlike Sora it is available, and will no doubt keep improving, but it doesn't seem ready for pro work at this point. The image to video option does seem much better in terms of clarity and detail though. Upscaling doesn't really help because it tends to look like bad over-sharpening at this time. Thanks for another great video though, Tim!
Id say in terms of Sora comparisons, it's pretty close on the Text to Video side. And remember, that's mostly all we've seen from Sora. The other big thing with Sora is that what we've actually seen has been heavily cherry picked. We actually haven't seen a lot of the "off" generations that I'm sure it produces. (Actually, confirmed via Paul and The Shy Kids)-- On the Image to Video side, we actually haven't seen anything on that front from Sora. And actually, agreed on your point for pro work for both model...and that's OK. I'm still of the option that AI Video doesn't need to be "industry standard/hollywood death"-- it can be its own thing. It's own weird/wonderful medium of storytelling.
@@TheoreticallyMedia Good points. Especially regarding it being great for its own kind of new art form creations. I have to remember that when I was creating Cybertown (one of the 1st online multiuser 3DVR metaverse-style communities) in the 90's, I didn't care how unrealistic our first 3D world creations looked, I knew they would get better, and I was just super-excited about being able to create impossible things with wild new technologies. These days my clients want realistic videos and so I'm looking forward to when that arrives too.
Hmm, I run my prompt and can see the created window with my prompt below it. I assume I click on the window to view the new video but I get nothing but what my prompt was displayed in the center of the screen. Are there minimum requirements for this?
I tried using it to restore an old photo and it was definitely generative - it turned a boy into a girl, changed the hair styles and made the photo look cleaner but the result was not representative of the original.
I am only 2:42 into this video and blown away. This is the most motion I think we have ever had in a video generator without having major deformation issues. This is sweet.
oh, I'm SURE those servers will be slagged in an hour! If the building hasn't already melted! Haha, might need to wait a few hours for things to stablize.
Young Blonde Princess was Midjourney. Danielle (who is a recurring character on the channel) was Leonardo. I'll see if I can dig up the prompts for you...
you might be interested in checking this video out, where I put together a AI Pirate short film, BUT-- I also go over the costs associated with it: th-cam.com/video/fZ4z4Z2GT2Y/w-d-xo.html
Hi, I great vid. I've been tinkering with Luma with some success, including extension, sometimes... I had a 20 second woman on a rope bridge taking a selfie on a phone and dropping the phone. It was a lot going on for the ai and she ended up over the edge herself😅 I also had a Chuck Norris meme pic and brought him to life to punch cans off a table. I have others I'm working on. For someone with no film creation or editing, this is amazing. Thank Tim Cheers
Pricing is crazy though. 30 generations for free (Could take 30 just to get something usable), then 120 for $29.99, Next level is $99.99 per month then $499.99 .... Sadly as expected totally priced the average user out of it
@@soggybiscuit6098 This is the case with all these "pay per generation" AI tools. Visual generative AI models at the moment function like a slot machines and that kind of randomness doesn't really work in professional environment. And it's not even all about unpredictable generation cost, but rather the time YOU spend praying for the machine gods to bless you with something half decent. It's really easy to spend 1 hour just tweaking midjourney prompt to get "just what you want" and never really get it in the end.
Is it possible with this new Dream Machine app to generate shot by shot a 10 minute video that keeps continuity of how the characters look and move? Thanks
Awesome Video always... just in time when Haiper tightens up the Free plan.... The free plan for LUMA gives you 30 videos a month... WISH I would have known that before run 18 tests. But now you know!
Not exactly. It will not be able to create anything due to the lack of sufficient training materials. Will be limited to what stock video can offer you.
What we really need is an AI generator that only charges for usable results. Luma AI is amazing but I burned through a month's worth of credits making a single 3 minute video. AI is like mining for gold, sifting through dirt hoping for a nugget (and being charged for the dirt).
Make sure you put it a prompt. They were upgrading their GPUs over the last few days (doubling speed) so you might have caught them during the update. That happens to me all the time!
Hi my-name-is-Tim ! Really impressive model in that it's the first one I try that's not in slow motion, really fast movement that came as a surprise.Thanks for the info. I didn't expect to reach the free generation limit that fast though. But that makes sense
Haha, the servers are getting slammed right now. I'd give it a bit to let things cool down. They released about an hour ago...I'm sure there's just a smoking crater where the GPUs used to be.
The easiest thing to create with Luma Dream Machine is too spinning waiting circles, regardless of your prompt. It always get's those right. Everything else it hallucinates into monsters. Want to have nightmares, plug in family photos or of your significant other
Thank you for this awesome video. Your channel looks interesting with ai features videos, so definitely subscribing to get more update on your videos 😉👍
Is it only web based? Can you run any of these models on your own gpu? I think this takes off when we have a stable video type program that runs on your own hardware.
If image to video yields character consistency and it’s possible to add lip sync (eg sync labs) then … just … wow this is truly the end of The Beginning
Hmmh, as always great video, but I end up with an empty thumbnail of the animation every time. Telling me (within the blink of an eye - "invalid date")? Strange 🤨🤔 (but maybe really too much traffic and I have to wait) Looks great anyway, interesting!
Check out Krea's INSANE new AI Video Upscaler here: th-cam.com/video/5mwKcBPlfVM/w-d-xo.html
Free usage only 30 generations p/month.
Very slow now because of everyone using it now my videos don’t generate they just stuck in the queue
"All that is in the past now." That was three days ago. That's hysterical.
like should this be counted as AI history? is someone keeping track of all this?
@@knucklesskinner253 I have a timeline going, but frankly it's down to the second at this point.
@@WifeWantsAWizardThat’s cool
The morphed fingers reminded me of ghost in the shell typing.😅
An epic fantasy battle scene though? That was certainly a dance off. lol
Haha. Lord of the Rings: THE MUSICAL!
Regarding the camera rotation, I've found "cinematic orbit shot" works quite well. That being said, while the camera movements are definitely impressive and so useful for actual projects, the control is still a bit limited, as you say. I gave Luma that as feedback because it's critical when you don't just want a bunch of random shots and a more fluid sequence of clips.
Ah, I was wondering about that! You got some GREAT Orbits going in your film! Did I see some tricks with zooming in on a still image before letting Luma take over as well?
Agreed on controls for movement! Would love to keep it simple still though. Part of me was really relieved by the simplicity of just a prompt box!
@@TheoreticallyMedia I don't really do a lot of pure text-to-video because it's useless for most of the stuff I need to work on where character consistency is key. But I've noticed that you get a lot less of the random takeovers from the model if you use an input image and then use fairly simple prompts. Kinda like old school "partial phrases" like we used to do in MJ.
PS: Another interesting keyword to play with is "dolly shot" or "tracking shot"....of course, assuming that's what you actually want to do.
PPS: Those "tricks" you mentioned are actually 2 different generations of the same input image, but one reversed. If you stick them together... that's the effect you get 😉
I have tested several mainstream AI video generators,
and most of them still have a long way to go,
especially when it comes to character details,
particularly facial features.
Although Dream Machine retains most details at the beginning,
it still loses some when the character starts moving.
However, the overall effect is really impressive.
I'm truly grateful that you could share such a great tool.
I've been waiting for this for so long.
That stuff is insane the way it can bring a still old picture to life like that. That shit is MAD.
Do they allow commercial usage with their standard plan? I cant see any contact us or FAQ section the website :(
That’s going to be mired in fog for quite some time with all these AI tools. Best bet in my opinion is just to make stuff for now.
Definitely the best rendered spinning Icon I've seen so far. 9/10
Great taste in music Tim! First, learned you were a Dream Theater fan and now learn you are an Anthrax fan! Great work as always.
Thanks! As a young comic book/Stephen King fan, obviously Among The Living hit pretty hard for me! Scott was SUPER nice as well. The guy is the real deal!
On the Dream Theater front: I have a video today that's going to make you pretty happy!
I experienced the same issue with trying to have the camera rotate around the subject, zoom in or out works really well as does pan slowly to one side or the other
Just so everyone is aware: anything that’s actually released Is automatically better than Sora. Sora doesn’t exist until it’s actually released as far as I’m Concerned.
Appreciate you keeping helping us keep on top of all this, very helpful!
Reminder that Pika Labs is still a lot of fun. Does cost money for the website version, but they do have some unlimited use paid plans (if these models end up charging per use instead of unlimited when they go pay).
Waiting to see what Midjourney does for video.
Pika is still a TON of fun-- and I'm sure will be releasing an update that is similar to this as well. Once they drop, usually the rest aren't far behind!
As for MJ...yup, can't WAIT to see that as well!
taking your outputs through a controlnet comfyui animatediff workflow would probably clear up a lot of the details.
That’s a great idea! I’m going to run some tests to see what we can do in terms of kitbashing this weekend. There’s a lot of potential here.
@@TheoreticallyMedia right now, I'm working on reconstructions of old, deleted TV programs (Doctor Who) where the originals are lost, but audio and stills exist... Dream Factory might be really useful here - as right now I'm using everything from animatediff to viggle, to deepfakes and lipsync tools)
Wow, that's great! Tim, you should test prompting for mouth and lip movements too. Like for talking, conversation and singing stuff.
I agree with you. Lip-syncing is essential when creating scenes for movies.
yes
I always get so excited when I see you post a new video. You always have the best information! I've made a ton of videos from your Haiper AI recommendation and am off to try Dream Machine now. Thank you so much for your great videos. 😁
Thanks, just tried it and it's probably the best image-to-video platform right now. Amazing job with face and anatomy consistency! However, I'm also looking for prompt consistency. Looking forward to creating the real dream machine.
I consider this a big first step toward getting there. I think we’ll see some massive advancements by the end of the year. November last year had some pretty big announcements, and I expect to see the same cycle this year.
Tried it out immediately. Starting from a picture it works like a charm, even keeping art style consistency (when the style is obviously anime). Now that it is day in Europe, and everybody got the news the queues are clocking up, and I don't get any more pictures...
Enhance Prompt check box is not there anymore :(
Also two versions ?????? nope only get one.
Interesting. They must have pushed something out. I noticed yesterday if I had enhance on, the results were super insane. I think they might be dialing things back a bit,
They seem to be throttling due to high demand. I know they were looking more more GPUs
I've got the best results where I've two photos taken soon after each other. You can supply both of those, and it will create a fill between those two scenes. A lot of the time, you get just a zoom. Sometimes a pan, and with a group of people maybe one or two seem to be alive while the rest stand still.
Ha! That camera control can do some weird stuff! Eventually we’ll have better control over them. But in the meantime, I’ve found some pretty good results by toggling the enhance prompt on and off.
Explain each movement of the folded arms. Like: grab right hip with left hand & grab left elbow with right hand.
Excellent video, thanks. I just tried Luna and I'm not super excited about the results to be honest. It response to parts of my prompts and not others. I tried starting with existing photos and very detailed prompts and all I got were fades and pans, not real animation. When I tried it with text only and a detailed prompt, it was ok, but not totally faithful to the text I entered. AI still has a ways to go.
Fascinating! Finally something proper to play with
It's pretty awesome. You're going to have a BLAST with this! Mind is boggling thinking about what someone super skilled with After Effects is going to do as well!
This AI adventure is crazy!
Every time I think it is starting to die down: WHAM!! Something like this happens!
A good first attempt. Still way to inconsistent and weird for my liking to jump in and use it for any project. It's like Midjourney - i wasn't really interested in it until the results yielded acceptable quality. Doesn't mean you shouldn't keep an eye on it or play with it of course. Let's hope more competition will get accessible soon and they push each other to way better quality in future versions.
For sure! I’d call this maybe MJv2 or 3. But, v4 was the BIG jump, so I don’t think we’re far off.
Also, now that we’ve seen this, I expect it won’t be long until we get similar leaps from Runway and Pika. And likely few others that we don’t even know about yet!
Yeah, this looks amazing. I've been so busy working on my Apple video that I missed this entirely.
Also, you probably already noticed or someone else said something already, but I sure hope the resolution isn't 128 by 720 ;)
Haha, 1280! I was wondering when someone was going to catch that! Ughhhhh, not enough coffee that day!
Do you ever review AI tools to take existing video and change it's style? (Or is there something even like that out there?). I'd love to see something that could take my 3D animations and make them into a 2D style.
I would very much like to see an extended tutorial with a few student tests. Also, the was one time when you put together a short horror movie. Maybe that would be another cool video, so that we can consider how to create better sequences. I love movies and I am able to remember some fairly long plane sequences from quite a few (I might be able to replay the entire Casablanca in my mind, albeit I can't remember all the dialogue)... But I haven't been able to create any interesting short movie yet. Thanks, Tim !
Working on a pretty cool project now that will have a full tutorial for this very thing!
I feel like this type of AI won't be used as much other than for proof of concept, precisely because of decoherence and just the overall lack of control you have for details. It's probably going to follow the same road 3D printing has taken. It hasn't really disrupted production lines for factories, it just works well for proof of concept and that's about it.
maybe not, it will probably take the route like ToonCrafter, if this would be used i see it being used like interpolation between frames, where you can guide it so it doesnt get lost and stuff like that, just look at the ToonCrafter stuff and you will see what i mean.
Are you kidding me? The photo to video finally allows for some consistency between AI videos. That is huge and will only get better. An artists could make an animated short with a simple sketch
It would be cool if they would incorporate something like Runway’s *‘Multi Motion Brush’* feature. This would be useful for animating multiple different elements from the image, and for posing multiple characters together.
Wow I own a small business. Uploaded a photo of my logo then brought it to life. Amazing! Gonna Use it. Also I sketch for fun. Animated a few of my favorites. I am loving this! This photo to video tech will finally allow for some cohesiveness in AI videos
Always with the top quality content! Thanks again! Been enjoying SD3 too today cant wait for fine tuned models
I gotta pop my head into SD3! The last few days have been a bit of a marathon! SO MUCH IS HAPPENING!
thank you for info - Luma is overloaded at the moment:) I wonder if a paid version works as slow as free.
I think it's just an overall bandwidth/hardware issue right now. They just released, so they're getting slammed. Give it a few hours!
@@TheoreticallyMedia I hope so your examples look promising:)
Pretty incredible. Definitely gonna be using it in my previs. Tho as a film maker, please reconsider demonstrating using copyright material to create new work. Something about using someone’s hard won work without any sort of disclaimer or caveat feels like a dangerous precedent. Love your work and know your heart is in the right place! Thank you for all you do!
I have this idea that I can finally watch my dreams as if they're a movie with the right prompt. This is getting me closer and closer to that ideal
Just going through a few image to video generations with a simple prompt gave some tremendously impressive results. Compared to others like Haiper & Pixverse Luma is an order of magnitude better. The videos pack so much depth and motion in to a five second clip. I would call it cinematic. It's not perfect, but it does so much with so little input, it's incredible.
9:10 - "32 extras on a sound stage"... launching into the musical dance battle number
Haha, maybe I should have prompted "Epic Dance Scene" and I would have gotten the epic battle!
@@TheoreticallyMedia 😂🙌
Luma also has some really cool 3D options.
I'm really excited to check out this new video function.
I'm curious to see if at some point we see those two sides of them combine. That could be huge.
Welll, it’s really cool…but I have 5 clips that started by saying they will be ready in 30 minutes…and after two hours they are still being created. Definitely not ready just yet. I believe the early testers had a fast priority…but for all of the others it is still not good. Also I got to make a few videos and had very mixed results. So, when I see the prices for the subscription they are super high because you will have to repeat many times to get a good shot…loosing a lot of the credits. This seems an amazing tool but still not ready for prime time until they work on the speed and pricing. But…I have great expectations for it. Thanks Tim for a great video as usual. ;)
Yup! If it’s any consolation I got in on Sunday (pre launch) and was able to generate super fast, but as soon as the doors opened up, I’m basically in the molasses with everyone else.
When it works, it does well. Rerolling? Sure, but that’s always a part of AIVideo.
As far as pricing goes, yeah- I’m not sure many will hop up to the pro tier. That said, if I’ve learned anything over the last two years it’s that when one big leap happens, the others aren’t far behind. I’m looking for big announcements from Runway and Pika soon.
@@TheoreticallyMedia yeah. I’m with you on that. I’m also on the ai video bandwagon since beta testing runway gen-1 and a lot has happen since then (really fun ride). I do believe it’s just a question of time (and it will be fast) till we have close to Hollywood image and movement quality on our hands to make a movie. Im rooting on pika because I believe they have a really fast video creation tool and a great Ui. I don’t know if you noticed but they improved their model in the last few days …it’s closer too Haiper I believe (which I think until now is the best quality…if we forget about dream machine, sora or kling). We are getting there, Tim! ;)
Tried to try Dream Machine and won't let me do it. I get this: Application error: a client-side exception has occurred (see the browser console for more information).
Not sure, I am waiting for a some test video smore than 1h now and nothing shows up. Maybe their GPUs are busy now :)
oh, they're getting SLAGGED right now for sure. You might need to wait until the horde dies down. Hug of Death for sure right now.
same but this is awesome
Yeah, I started 4 generations an hour ago, still nothing but the weird thing is, they don't have a timer or anything that suggests the content is still being created. I don't mind waiting, I mind not knowing... but I dare say they just weren't expecting the traffic.
@@TheDandonian I mean you're probably not wrong - but how could they NOT know is what I wanna know? :D
Same here more than an hour so far
David Cronenberg sausage 😂 Tim you totally made my day! Nice reference humor
where to get working prompts for Dream Machine? to move the camera and such...? thanks
Wow, this is a big step. I like how it animates pictures, not perfect but so much potential already! Still need the lip sync feature for creating actual music video's or movie dialogue.
One of my favorite AI content creators.. I am currently playing with this and will be Paying for it in the future to it as I really see the potential :)
Thank you! And agreed- this is very v1. By the time they hit v3? We’ll be cooking with gas!
awesome video, but I tried and none of my videos nor prompts work and I have no clue
They're getting SLAMMED right now. I'm sure there is a smoking crater where the servers used to be. I'd give it a few hours to let things cool down! The old Hug of Death!
@@TheoreticallyMediawhy it always be like that bruh don't they think servers will be that
The followers of my ai art have been asking when I will release new content! Now is the time!!! Can’t wait to use!
Fascinating. I'm always happy when I see mangled fingers because the rate AI has been progressing is scary and I want to think we still have ways of knowing it's fake.
Fantastic Job Tim!!! On the cutting edge as always!
Having so much fun messing around with uploading my own images to animate. Downside is that there is no way to save the video. Sure I can screen record but wish there was a download option
Good to discover your channel, great first watch of one of your vids - thanks Tim
Ah, excellent to hear! Glad to have you here!! Actually, have another pretty cool one coming up tomorrow! LTX Studios...but, with a music twist! It's actually one of the best videos I think I've made yet!
I remember when AI text to image was so "primitive" it wasnt so long ago, soon videos will come as far as images have come, I'm so excited
Wow, thanks for such a quick report and update on this, great content! And I couldn't help but think, the decoherence and morphiness - that's probably all possible to clean up using image generation post-processing, frame-by-frame a.i. automation to get something cinematic, even if the initial results have some wonkiness. *post comment edit - I tried it out and looks great, seems like they're taking the Midjourney approach of allowing enough usage to get people understanding the unique value - but a paid usage model, which is a great way to get people on board.
Wow, thank you for letting us know about luma. Amazing generations and still free.
Luma is the one that I subscribed to but it has huge problems. Lots of times it just ignores the initial image and adds random stuff.
Great Chanel and great video. I started using dream-machine and yes lots of potential. But certainly dealing with human motions movements need serious work. If you making short videos without any human or any movement the generated videos are stunning.
Wit these ai video things how would you be able ta keep the same character and add them into the next scene? Like where there just one or the other ta maybe having both in the same scene and probably reusing the same character in other areas like instead of being at the beach their at a theme park wit their friends. And ta make that same character talk in that scene or different scene?
I have alot of ideas and characters fur movies or shows I'd wuf ta do.
This model is “definitely on a Sora level”? Come on man. Are you joking? This is not even in remotely the same ballpark as Sora. It’s still impressive, all of this stuff is. But let’s not pretend this is anywhere close to Sora’s quality level.
No it’s in the ballpark. Not as good yet, but this is early days. It’s exciting too
@@injectionAI It may be exciting, that’s fine. But it’s not in the ballpark of Sora yet, by any reasonable definition of ballpark.
@@therainman7777 maybe you’re not familiar with how software devs work. Don’t worry, just wait
@@injectionAI I have no idea what that’s supposed to mean. This is an ML model that is trained from data using gradient descent-it’s not a product of software developers. Also FWIW, I’m an AI engineer and I work with software devs on a daily basis. So I don’t think it’s because I don’t know how devs work.
@@therainman7777Atm Sora is overhyped because it hasn’t been properly tested by the public.
Neat!! I dig the random ugly non-cohesion aspects, reminds me of when Pika was still weird. Thanks Tim!
100%! That's exactly what I kept thinking when I was using the text to video stuff-- I haven't had this much fun since early Pika! And, when you get the janky outputs, I mean-- Comedy Gold!
@@TheoreticallyMedia Totally superfun! Meanwhile, can we negative prompt in Dream Machine? I am reluctant to try with my free account. Please tell me ObiWan.
Not sure why this is being compared to Sora - the Sora videos I've seen have been sharp and clear, and this is fuzzy and very unsharp, and lacking the clarity and sharpness of Sora videos. True, unlike Sora it is available, and will no doubt keep improving, but it doesn't seem ready for pro work at this point. The image to video option does seem much better in terms of clarity and detail though. Upscaling doesn't really help because it tends to look like bad over-sharpening at this time. Thanks for another great video though, Tim!
Id say in terms of Sora comparisons, it's pretty close on the Text to Video side. And remember, that's mostly all we've seen from Sora. The other big thing with Sora is that what we've actually seen has been heavily cherry picked. We actually haven't seen a lot of the "off" generations that I'm sure it produces. (Actually, confirmed via Paul and The Shy Kids)--
On the Image to Video side, we actually haven't seen anything on that front from Sora. And actually, agreed on your point for pro work for both model...and that's OK. I'm still of the option that AI Video doesn't need to be "industry standard/hollywood death"-- it can be its own thing. It's own weird/wonderful medium of storytelling.
@@TheoreticallyMedia Good points. Especially regarding it being great for its own kind of new art form creations. I have to remember that when I was creating Cybertown (one of the 1st online multiuser 3DVR metaverse-style communities) in the 90's, I didn't care how unrealistic our first 3D world creations looked, I knew they would get better, and I was just super-excited about being able to create impossible things with wild new technologies. These days my clients want realistic videos and so I'm looking forward to when that arrives too.
Hmm, I run my prompt and can see the created window with my prompt below it. I assume I click on the window to view the new video but I get nothing but what my prompt was displayed in the center of the screen. Are there minimum requirements for this?
I tried using it to restore an old photo and it was definitely generative - it turned a boy into a girl, changed the hair styles and made the photo look cleaner but the result was not representative of the original.
I am only 2:42 into this video and blown away. This is the most motion I think we have ever had in a video generator without having major deformation issues.
This is sweet.
Can't wait to see what veo, Kling, luma, sora and all other video models do in the near future!
Awesome !!!! I'm trying right now, the servers must be overloaded as it takes forever to generate a video :D Thank you and awesome short story !
oh, I'm SURE those servers will be slagged in an hour! If the building hasn't already melted! Haha, might need to wait a few hours for things to stablize.
"Young blonde princess" and Danielle as a pirate....how were those still images generated?
Young Blonde Princess was Midjourney. Danielle (who is a recurring character on the channel) was Leonardo. I'll see if I can dig up the prompts for you...
That wizard is 100% Sylvester Stallone in heavy AI makeup.
Ha! Sly has become the Mother from “Stop or my Mom Will Shoot!”
Terry hogan, brothers
Im not getting 2 generations personally unless im looking in the wrong place?
I gotta check. They might be throttling right now due to overload.
thanks for sharing! looks like luma labs is already overloaded, none of the videos are showing up.
absolutly
Just click on the square, it opens. It is not showing that finished rendering.
Wow so damn cool love it ten years ago that would have taken me 2 days now 3 min.
you might be interested in checking this video out, where I put together a AI Pirate short film, BUT-- I also go over the costs associated with it: th-cam.com/video/fZ4z4Z2GT2Y/w-d-xo.html
Hi, I great vid. I've been tinkering with Luma with some success, including extension, sometimes... I had a 20 second woman on a rope bridge taking a selfie on a phone and dropping the phone. It was a lot going on for the ai and she ended up over the edge herself😅
I also had a Chuck Norris meme pic and brought him to life to punch cans off a table.
I have others I'm working on.
For someone with no film creation or editing, this is amazing.
Thank Tim
Cheers
Svd does text to video. Why wouldn't sora be able to do text to video
Pricing is crazy though. 30 generations for free (Could take 30 just to get something usable), then 120 for $29.99, Next level is $99.99 per month then $499.99 .... Sadly as expected totally priced the average user out of it
so it is not free, fake news
Eh at that price, who is that for. At this quality it's only for fun and hobbyists and that's not the price point for it
25-30¢ per roll. Cray cray.
I wouldn’t pay to use it - maybe when it improves I would.
@@soggybiscuit6098 This is the case with all these "pay per generation" AI tools. Visual generative AI models at the moment function like a slot machines and that kind of randomness doesn't really work in professional environment. And it's not even all about unpredictable generation cost, but rather the time YOU spend praying for the machine gods to bless you with something half decent. It's really easy to spend 1 hour just tweaking midjourney prompt to get "just what you want" and never really get it in the end.
Is it possible with this new Dream Machine app to generate shot by shot a 10 minute video that keeps continuity of how the characters look and move? Thanks
This is interesting. It will actually be useful when I can run it off-line locally like you can with SD 1.5
doesnt work for me, browser doesnt show any video :(
Awesome Video always... just in time when Haiper tightens up the Free plan.... The free plan for LUMA gives you 30 videos a month... WISH I would have known that before run 18 tests. But now you know!
Baby steps. These videos will be mind expanding within 2 years. I'm curious to how the VR applications will become
Not exactly. It will not be able to create anything due to the lack of sufficient training materials. Will be limited to what stock video can offer you.
Top quality content as always! Very excited to mess around with THIS!
the Scott Ian clip is nuts!
What we really need is an AI generator that only charges for usable results. Luma AI is amazing but I burned through a month's worth of credits making a single 3 minute video. AI is like mining for gold, sifting through dirt hoping for a nugget (and being charged for the dirt).
You were quick on this one. Nice!
This morning was a REAL rush!! Yeah-- Lots of Coffee!
site not working for me... Application error: a client-side exception has occurred (see the browser console for more information).
The best video AI model, it’s finally free, time to create like a pro, just wait and see! 🎥😄
Did anyone have a problem with generating from their photos? It doesn't do anything. I have tried three different photo pairs...
Make sure you put it a prompt. They were upgrading their GPUs over the last few days (doubling speed) so you might have caught them during the update. That happens to me all the time!
Hi my-name-is-Tim ! Really impressive model in that it's the first one I try that's not in slow motion, really fast movement that came as a surprise.Thanks for the info. I didn't expect to reach the free generation limit that fast though. But that makes sense
What. But... It's just amazing. Finally, it's not just a demo and promises but something we can really try. And the result is... really not bad
Love the American to British banter 😂
Back-to-back impressive videos. I think I'm gonna be busy trying this stuff out. Thanks for all your analysis and the work you do. 😀
The pricing is insane though as expected (it's NOT free, they give you 8 free generations) and this gives one the idea of how much SORA will be.
Can this or an org implement my Metahuman rig as a consistent character?
can't get it to work (Application error: a client-side exception has occurred (see the browser console for more information).)
I tried it out and all I get are blank thumbnails. How long does it take to generate? It's been over 10 minutes now and nothing.
Haha, the servers are getting slammed right now. I'd give it a bit to let things cool down. They released about an hour ago...I'm sure there's just a smoking crater where the GPUs used to be.
The easiest thing to create with Luma Dream Machine is too spinning waiting circles, regardless of your prompt.
It always get's those right.
Everything else it hallucinates into monsters. Want to have nightmares, plug in family photos or of your significant other
How can I extend the video output? Are we stuck at 5 seconds with no option to extend?
I’ve got a trick I mention toward the end of the video!
Clickbait thumbnail 😂😂😂. I am in 😊 and i subscribe. Good job!!😊😊😊😊
Thank you for this awesome video. Your channel looks interesting with ai features videos, so definitely subscribing to get more update on your videos 😉👍
Is it only web based? Can you run any of these models on your own gpu? I think this takes off when we have a stable video type program that runs on your own hardware.
If image to video yields character consistency and it’s possible to add lip sync (eg sync labs) then … just … wow this is truly the end of The Beginning
Hmmh, as always great video, but I end up with an empty thumbnail of the animation every time. Telling me (within the blink of an eye - "invalid date")? Strange 🤨🤔 (but maybe really too much traffic and I have to wait) Looks great anyway, interesting!