Mastering Luma's Dream Machine: From Basics to Advanced!

[@RetrocausalMemories]

Super useful, thanks Tim! Been following you from the beginning, glad to see your channel really growing. 💟👏

[@markrichards5630]

FYI. You can use loop in an extend and it will loop back to the very very beginning (not the start of the extend). This is a great way to make the loop more subtle

[@TheoreticallyMedia]

Brilliant tip!! thank you so much!!

[@AI3EFX]

Appreciate this tutorial man. Any and all tips are welcome 🍻

[@TheoreticallyMedia]

thank you!! Hoping more pop up in the comments!

[@dalecorne-new-mtv]

Thank you for this video. I've been using it to make "The All New MTV". I'm bringing back the original MTV just to do something cool. When I make a logo shot as "first frame/last frame" it makes amazing 3d logo morphing that looks like it was done using a lot of money and special equipment. It's also hosted by droids....lol

[@TheoreticallyMedia]

Haha, this is an awesome use case! I can't wait to see AI Adam Curry and all the other old school VJs! AI Killed the MTV Star!

[@TailesfromtheCryptkeeper]

Thanks Tim and Thanks Luma, I was waiting for someone pulling something like this masterclass up!

[@amzpro5734]

Great video and info about the Luma Ai video 🎥 tool! Agree it is cool at certain things. I'm always interested in people's "keep" ratio versus number of AI video renders they make. I know it can vary from person 🧑 to person 🧑, but to get a usable render that you want to keep ( so a bit like having a "shooting ratio") how many renders do you make average for each clip you keep for a final edit? 3, 5, 10...more?

[@user-on6uf6om7s]

It depends on the subject matter and how complex the action is but I would say 10 conservatively. Sometimes I'm happy in the first 3 but that's balanced out by the shots that take 20-30 generations which might happen every 4 or 5 shots.

[@eddysgaming9868]

Wouldn't a more literal prompt help, like "a man in a blue business suit walking down a busy city sidewalk'? In painting, I was taught to embrace the 'happy accident,' where the results are unexpected and often better than intended. Thanks, Tim.

[@TheoreticallyMedia]

oh, I could 100% clean that prompt up a bit-- it's just that "Man in a Blue Business Suit, walking down a busy city street" is sort of a standard test on the channel. Both for Image Generators and for Video Generation. It's kind of a silly one, but I figure it's such a "stock" prompt, that I'm always curious to see what the models do with it. And yeah: Happy Accidents! LOVE them! Since you brought up painting-- I wonder if I picked that up from Bob Ross? I kinda think I might have!

[@eddysgaming9868]

@@TheoreticallyMedia I think we all did. lol Bob was the best at promoting the concept. RIP.

[@isajoha9962]

Image to Image looks really cool.

[@Lev3lUpLife]

I love Luma! I have also found that using detailed prompts without checking "enhance" usually gives the best results. I've been using Techhalla's Runway prompt generator with my own tweaks mostly, but I think you had a prompt GPT out there. Is it still available? I don't see it in my history but I'm pretty sure I tried it out at one point. 🤨

[@TheoreticallyMedia]

Oh thanks man!! It should be there-- but every once in awhile, OpenAI flags me with an automated policy strike and I have to click for a manual review-- and then it comes back. I presume some user tried to do something that OAI considered a no-no, and the GPT gets flagged. It's a bit annoying. If it isn't there anymore, please let me know and I'll swing over to click the stupid button again!

[@Lev3lUpLife]

@@TheoreticallyMedia Thanks! I just looked again and found it so I stuck it in my sidebar. 😃 I'll probably use it today while using up my Hailuo unlimited free trial.

[@FANTASTICWONDERFUL]

Thank you

[@TheBlessingReport]

another great video as always

[@gumvue.studio]

Hi Tim, please please tell them to fix the website. UX , they have to remove that ugly video background, its just killing my CPU .. so I cant even login to the website.

[@Hazzardhenry]

it is the worst of the bunch

[@gumvue.studio]

@@Hazzardhenry I cant use it, for that reason

[@GloryHouseStudios]

great video I will use the gassian splat

[@MikeDKelley]

And perhaps one more (critical) question: You've mentioned SO many AI products my head is spinning. For someone on a limited budget who wants to explore this stuff but doesn't want to spend more than, say, $25 a month (at the most) is there one in particular you'd recommend? I'm primarily interested in taking still images (just the starting one) and making videos from them. The Hailuo one was a lot of fun to play with and VERY responsive, but also seemed limited in what it could do. Luma's Dream Machine seems very powerful, but I can't really test it because after two hours I still don't even know if I'm in the queue or not (it says I'm on the "waitlist" with no real idea of where on that list I am or any progress). Plus it worries me about how many tries you get - for around $25 I can get 70 or so per month but while that *seems* like a lot it took me 20 or more in Hailuo just to get what I wanted and I could see burning through 70 in a few weeks. So - any advice along these lines?

[@christopheranderson2576]

I've had good results using "camera orbit" left or right on image to video.

[@EllenVaman]

I dig Luma …it performs best by uploading first and end images and good prompting .. and voila!!! Magic 🙃 And Grazie for the laughs again 😁

[@TheoreticallyMedia]

1000%!! Laughs are the best part of AI! And totally agree: the i2v method really shines!

[@amkire65]

I was pleased to notice that you can delete videos (maybe it was always there and I just missed it) because some of my old videos from way back about 4 months ago, turned out to be real horror stories! I keep going a "you are on the waitlist" message, so either it's really busy, or they are features I don't have yet (on the free method).

[@TheoreticallyMedia]

Yup! I noticed the delete recently as well-- which is.a WELCOME addition. Would love to have some more organization tools (like Project Folders or better date sorting), but I'll give them time to develop those!

[@MikeDKelley]

Can Dream Machine work from an image, like you showed with the other product recently, Halioia? I've been having SO much fun with it (but now my trial has run out).

[@TheoreticallyMedia]

It can! And even more so, it can work with TWO images. A beginning image and an end image!

[@MikeDKelley]

@@TheoreticallyMedia Okay, great - then I think I will sign up for it. I had SO much fun with that other AI thing so perhaps this will be just as much fun.

[@MikeDKelley]

@@TheoreticallyMedia Also, thanks for the reply and all you do for this. While I'm not sure I'll ever do anything with AI (the way I do with 3D animation) it's still a ton of fun to play around with. What I *really* want is to take my 3D animation and have the AI process it so it looks like 2D. When we get there I'll be thrilled (and I do think we will get there someday).

[@TheoreticallyMedia]

@@MikeDKelley You can try playing around with some of the Video To Video processes. You can see a bit of it in principal in this video: kzbin.info/www/bejne/npvOZZRvnN18gsU And while I was using more of a live action to 3d to animation workflow, I think you can obviously extrapolate some ideas to be used with straight 3d. Or, maybe even come up with some wilder ideas! That's a lot of what I love about this early era of AI-- no rules, so everyone can bring their own background and mojo to it!

[@MikeDKelley]

@@TheoreticallyMedia Thanks again for all the info. I assume if I pay for Luma I get ahead in the queue - right now I'm waiting and there doesn't appear to be any particular information about how long or where I really am in the process (unlike that other program you talked about, which gave me a time at least to gauge).

[@zeeyannosse]

Interesting ! Lucky me : I decided to watch the full video :)) at first was thinking 🤔 hummm the blue dude show is really bad indeed ! More than the kindness you put in your description trying to save the boat sinking :))) and like you said, it’s kind of a promo video … it felt like not being honest about the obvious situation ;))) 😅 anyhow ! Glad that the rest of the video is really informative and helpful ! Keep it up ! Keep it real ;))) cheees !

[@TheoreticallyMedia]

Yup! While I think Business Suit blue guy isn't horrible, I'll say that it isn't mindblowing either. But that's something I always like to showcase on the channel: I don't like to cherry pick and set up unrealistic expectations of any of these AI Models. That said, once we get to Image to Video, I think things really take off! And-- I still say there are uses for T2V-- it does well with environments/locations etc-- just maybe not always great with cinematic people!

[@gogidolim]

Question: is there a way to create an AI video by upscaling rudimentary previz-style cg animation to make it look more photorealistic?

[@vrl1109]

Do you mind telling me what prompt you used to get that post apocalyptic look in the movie at the end of this video? Thanks.

[@MagnumOpusx]

Luma can be creative but the artifacts , blur , pixelation and distortion keep me using Runway for more of pro level output

[@Herson_ricardo_alarcon]

Thanka forcthe video

[@LouisGedo]

I'm only seeing a notice now for this video drop 😲 Typically I get notices much earlier

[@TheoreticallyMedia]

Oh, you know what-- I'm in Europe right now (headed back to the states on Monday)-- Since I uploaded and posted from here, I was wondering if YT was going to do anything weird. And there you go: It did!

[@Ibo87Mmm]

Hello my friend and thank you for your efforts in clarifying and explaining I hope you can give me advice regarding a reliable site to convert Arabic text to audio without spelling errors, especially in videos where the shot is a dialogue between two people in the same shot. Is there a way to produce a shot that is longer than 10 seconds? Your follower from Iraq

[@wadeva69]

Why dont the links to Luma work? Their website doesn't seem to open

[@pongtrometer]

Do you think we will see the rise of ‘ The directors AI cut ‘ re edits or current post production, as well as alternative interactive story telling, that the viewer can semi remix the narrative.

[@Hazzardhenry]

we will just do it on our own, ive beem downloading all older movie workprints, and laserdisk copies fto train our model

[@Onion224]

8:19 you only have those camera options? for me there are a lot more..

[@TheoreticallyMedia]

I think you can still prompt for different ones and you may or may not get them (typical of AI Video)-- but things like "tilt" or "handheld" often work. I think they just put those into the dropdown, since they're the most common.

[@knoxieman]

Much better to use Ideogram to create an image then use it with image to video, its pretty bad at creating videos from scratch compared to minimax, both work much better with image to video

[@TheoreticallyMedia]

Yeah, I2V is always the preferred way of getting (mostly) what you want, for sure!

[@K-Digital-Dreamworks]

How do you get them not to pan in any direction.. I just want a fireplace flicker no other movement..Please and Thank you.

[@dirtydevotee]

If you know "down the street" is a problem, why not change the prompt at 1:40 to read "down the sidewalk"? That's not really Luma's fault. Also, 3:29, deer in real life are humongous.

[@eddysgaming9868]

Agree, it's taking his prompt literally. So, he's walking "down the street." Deer can be large and very skittish.

[@Maylin-ze6qx]

Very beautiful ❤❤❤❤

[@PeterStrmberg007]

The man in a blue suit was a bit unfortunate. The guy himself looked warped in the face on all samples. Luma is actually good from an image prompt.

[@TheoreticallyMedia]

Yeah, I don't like to cherry pick too much-- since I think that sets up unrealistic expectations. I think the I2V examples later in the video showcase the real power here.

[@ECHSBACHS]

Cool

[@Mooooty]

Dream Machine still takes foreverrrrrrrrrrrrrr....to build out your result.

[@capitannegro100]

My video has been in queue for a month now 🤣

[@scoobydoobydoooo]

Same, but only for the last few days

[@Hazzardhenry]

same, luma and kling are just not ready for mainstream traffic yet

[@zakaris7259]

I nearly forgot the existence of sora so luma is og

[@TheoreticallyMedia]

Haha, you know what's crazy? Sora "dropped" (or at least made the big announcement) in February. Yeah...this year. So like, 8 months ago. And yet, it feels like it's been years.

[@lorenzozen]

what's the best AI voiceover tool now that I could use in my videos?

[@aitechwitht]

elevenlabs is still the best

[@armondtanz]

Luma is slipping... With kling and minimax and runway... its defo 4th. 😢

[@TheoreticallyMedia]

Never count a model down! I’ve learned that every one is one update away from taking the top spot!

[@armondtanz]

@@TheoreticallyMedia Ha. Yea. I forgot, sora, meta, seaweed... Ooof. Busy busy 2025.

[@bloxyman22]

I think it many ways it shares third place with runway unless you can justify getting the unlimited plan. Runway is generally more coherent and higher quality but is also incredible lazy when it comes to movement in the scenes and more censored. Kling I have not used other than the lower quality free version, which is not too great. But minimax is for sure a big step up with exception of lacking features like extensions and first and last frames.

[@hqcart1]

@@armondtanz and adobe :)

[@scoobydoobydoooo]

That woman in red on the beach, her footsteps were confusing. It looked like her legs were corkscrewing around each other

[@TheoreticallyMedia]

Yeah, I'm always a bigger fan of Image to Video, as that really helps the model in terms of posing/camera angle etc-- and I think you tend to get more cohesive outputs that way. Text to video is still kind of fun though-- although, I tend to think it works better for things like "big city shots" or animals...basically, the non-human subject kind of things.

[@EnhanceYourJourney]

I am looking for an AI video generator that allows me to upload an audio file. If anyone here knows of an easy way to generate it a simple kind of music video I would appreciate knowing about that and if lumen can do that that would be great too.

[@TheoreticallyMedia]

I've seen a few in development-- I'll let you know as soon as I see anything drop on that front! I think Suno just dropped a feature where you can generate a song from an image in your camera roll. Not quite what you're looking for, but getting close!

[@pookie67]

Luma took a step backwards with their last update in my opinion. And trust me I used it A LOT since it first launched, including when image to video first launched. There is no better authority than me on this stuff I've used it on some very complex projects. While their own models and generator has improved from where it was at the beginning, the speed and complexity of movements has taken a step back to slow-motion land. I was getting a lot of really amazing results with image to vid a few months back, and I have loads of footage to prove it. It can no longer do what it was capable of

[@TheoreticallyMedia]

Yeah, they're in a bit of a transition for sure. One of the things I've noticed with a lot of the AI Video models is that they all tend to have their moments of ups and downs, depending on model updates. Kind of how like Midjourney's (in my opinion) 6.1 isn't as good as 6. And to be totally fair, I still think that v4 is the coolest one. The thing on the video side, is that we don't really see the big version number changes. It mostly seems to happen in the background without a lot of announcements-- at least w/ Luma. The other thing to keep in mind is that even when companies DO make a big deal about version numbers (Like Gen-3), do take note that it is version 3. And if you take a look at Midjourney version 3...well, that also had a long way to go!

[@pookie67]

@TheoreticallyMedia Runway did the same thing back around December of 2023 when they upgraded/changed Gen-2 for what seemed like the 2nd or 3rd time and added new models. The first release of Gen-2 actually produced some pretty versatile motion for it's time, albeit extremely low quality details. It went to higher quality, especially when image to vid was offered but MUCH slower motion up until Gen-3 (which still could use a lot of work in my opinion but I'm sure they're working on it) Fast, complex movements are the most difficult thing for AI to do while retaining details and quality

[@NEURAMos]

I think it's time for the LUMA platform to consider an update and make video generation higher quality because there are AI platforms that are much better...I don't see any point in spending money on a LUMA subscription.

[@SamoDaDam]

For 6 days Kling AI has not been working properly. Massive amount of users can not create anything, no matter they are free or paid. You should also say something about this. Idea for video: HOW GOOD IS TECH SUPPORT on each generative AI platform ?

[@Hazzardhenry]

same, I was coming here to see if he posted about it. I found Kling thru this channel, now Im out 300$ and they wont reply

[@choochypoo]

they are also deleting comments

[@SamoDaDam]

@@Hazzardhenry exactly the same. Kling was introduced to me here and the same, they took my money and silence... PLATFORM NOT WORKING THEY DON'T CARE

[@TheoreticallyMedia]

Can't speak the deleting of comments, but they did recently address the failed generations. Apparently it was caused by API abuse and that was causing the system to false flag genuine users. They said to change your password and put up a post in their discord w/ your user ID to reset. Agreed though: Communication lag on their part. They need to get better about that.

[@TheoreticallyMedia]

a bunch of people tagged me on X and DM'ed me about it, but funny enough, I'm hopping around Europe right now, so with timezones, I was getting the messages at like 3am! haha, took me a bit to dig into it. But yeah, they're apparently (finally) addressing the issue. Agreed: they need to get better/faster at communicating issues. I mean, I'm just one guy with a YT channel, so I have an excuse!

[@Comic_Culture]

Geesus Cries..luma is no match against hailouai...and luma is charging for this craps wdf...😂 Lmao

[@Dazzastrous]

Seems like your taking a step back with this video, this is no competition for the likes of Kling and Minimax or others around the corner, quality is really bad compared to kling and minimax and I kinda see why you did this they have been overtaken and are getting less attention because of the better models and are trying to remain relevant by reaching out to yourself.......its nothing new. I enjoy watching for the next exciting development not the same old same old

[@TheoreticallyMedia]

So, I've learned by covering these tools for awhile now that you never count a model down for the count! They're all one update away from being "king of the hill" again-- or, releasing a whole new an interesting workflow. Like, Kaiber (which I just did a video about)-- everyone was like: I guess they're done! But-- nope, they were quietly working on a total revamp (which, oddly includes Dream Machine)-- And...man, how many times has Midjourney been declared "dead"-- So, yeah-- when a company reaches out, I always like to do these tutorial type videos. Where we learn the ins and outs of the platform. The output quality is going to shift back and forth for awhile, as the tech advances. So, for a video like this, I think it's fun to look at basics and techniques.

[@shauncaruna387]

Luma is not good now. Its like Pika's failed little brother. Runway is starting to become 💩 too. So many generations to get one decent one. Some of these ai video generators are becoming scammers.

[@WarpedTrekker]

Yep same with the music AI generators. I use Suno, and it takes sometimes dozens of tries or more to get decent output per each part of a song.

[@pongtrometer]

Why do all these ai video generators love slo motion and give people 5 secs in this hyper fast world 🤪 I know it’s early days, but life isn’t about slow panning.

[@Lifebeats_A.I]

I love A.I apps and use them every day,It wouldn't do anything so I've Launched it in to the trash,A.I needs to save me time not waste it

[@TheoreticallyMedia]

Yeah, I was talking about that a little while back. I get your point from a time standpoint-- but, that's not really where we are yet. But, if you also think about it as getting a shot of some woman walking on a beach, it DOES save you time, as you don't have to travel to a beach, find an actress, wait for magic hour, and do all the other physical production aspects. So...when you look at it like that?

[@MrSzwierzak]

Luma works way better with pictures from MJ.

[@hqcart1]

thanks for not picking the good vids only!!!

[@TheoreticallyMedia]

I always think it's important not to cherry pick. Sets up WAY too many unrealistic expectations of what the tech can and can't do!

[@meikmeiker2615]

If you want to burn money you should create AI videos.

[@TheoreticallyMedia]

I gotta say, it depends on how you look at it. I went over a whole thing in the back of this video (kzbin.info/www/bejne/nIuXq2eQZ6yHaLs) basically, comparing the cost of an AI short film vs the cost of other production methods. So, for sure, you're going to spend money, the question is: How much money are you saving?

[@alphacentauri424]

unfortunately Tim your title is kind misleading... this is not taking anyone from basics to advanced...you just went through the settings mostly.. Luma is sooooooo annoying and bad with their controls consistency...there isn't actually real control, it seems more like spend money and regenerate until you get something decent. I was hoping this video was an actual deep dive into using Luma...but it wasn't...so PLEASE, stay away from click bate and be more honest....this is how you show respect to your community. I like you and hope this was an error on your side, that won't happen anymore

[@DannyFerguson-b3q]

How could you NOT notice her feet.. That's a fail.. IS she a troll?

[@chrisjlee2013]

lol Luma Labs still can’t do any of these? Kling and Runway is waaaay way way ahead

[@Zerobytexai]

Luma Labs is garbage now LOL why are they even still trying? They should become free they're so behind. Make sense why you were forced to make this video because you partnered with them for money 🤑😅 That move does reflect badly on this channel though FYI. If possible I would only partner with top tier from now on but I understand this is your business. You DO take a hit on your channel every time you make a video with low tier models just for the sake of business though. You should pick quality over business next time, otherwise people here on KZbin can see right through it and this channel becomes less appealing and loses subscribers or won't gain as quickly as it can. Just trying to help you.

[@siheedsalaam4712]

😂😂😂😂Relax😂Dream machine really isn’t that bad if you use image to video and follow the prompting advice. I’m a runway ml subscriber as well(it has a hard time following directions and warps a lot which is disappointing for the price), kling (overrated but has potential), pika( just mid). Dream machine is powerful, it follows your prompt as you direct it better.

[@Skylineblue1997]

Vidu is better than those all,in term of movements....

[@siheedsalaam4712]

@@Skylineblue1997 it’s just like the others as far as quality, sometimes it’s great! Sometimes it goes left 😂it’s still the early phases for all of these platforms though, all have their pros and cons

[@Skylineblue1997]

@@siheedsalaam4712 yes it create fail videos but when it create something good its really good, they give more chances to play witu prompt as it take only 5 -10 sec to create video,aslo provide free daily credits to experiment with prompt, I'm now able to create NSFW things too😂😂

[@Zerobytexai]

@@Skylineblue1997 Nope Vidu barely even makes the top 5. If you want actual cinematic movie quality where people can't tell the difference between AI and reality, then only Minimax is on that level. Vidu is not even in the same stratosphere.

[@WarpedTrekker]

I've tried them all. First, they are all overpriced. Second, they need to do away with the "credit-based" system. Or at least make it easier to see how many generations you have left. Make it "1" credit per video generation. Not some number like 10 where you have to divide credits to figure out how much you get. Also right now I tried DM and it said I was on a wait list. Can't even give it an image to generate. Also what you pay for in price, most of these don't give enough generations per month to do anything practical. You can maybe make a 5-min short film if you are lucky with the subscriptions. Not enough render time is given for what you pay. I know servers with the Nvidia GPU's are super expensive, but come on.