YouTube transcripts

Is This the Best Way to Use MiniMax? Save Latents Tutorial: video thumbnail

Is This the Best Way to Use MiniMax? Save Latents Tutorial transcript

Ok Todd AI · @OkToddAI

Published September 19, 202617:56747 views

Watch this video on YouTube

Transcript analysisComputed from the caption text

Words

3,873

Runtime

17:56

Speaking pace

216wpm

Reading time

16min

216 words per minute, above the 201 75th percentile of 349 measured videos. That distribution comes from the 349-video hook study.

Opening (first 30 seconds)

Okay, today I got a brand new workflow that I found. Actually, two workflows that I found. And I'm super excited about these Miniax H3 workflows because they fix an issue that we were dealing with. And I'll show you. I made a video that I'm about to show you, but I want to tell you where I got these workflows from first. So, um, this this guy right here, Sedo Drive, no, Sato Dive, my bad. Uh, awesome, awesome. Totally recommend following him. Uh, this is the video that he made and I'll I'll link it. And then this is the latent continuation that he created as

108 words, the words spoken in the first 30 seconds at 216 words per minute.

Sentence shape

MeasureThis transcript
Sentences353
Average words per sentence11.0
Longest sentence61 words
Questions asked15
Sentences containing a number30

Most used terms

  • gt44
  • video39
  • uh30
  • gt gt22
  • reference12
  • okay11
  • frames10
  • seed10
  • um10
  • workflow10
  • hit9
  • latent9

Filler phrases

87 in total: uh 30 · like 21 · you know 11 · um 10 · actually 7 · basically 4 · kind of 3 · I mean 1.

A literal whole-word count of the same phrase list the Prepublish browser extension uses, so a phrase inside another word is not counted and a phrase used in its ordinary sense still is. It is a count and not a judgement.

What this transcript is

Every word below is the caption track YouTube publishes for this video, pulled from the video itself and reproduced unchanged. It is not Prepublish's writing, not a summary, and not a re-transcription: it is the video's own published captions. English captions, generated automatically by YouTube, in the video’s original language. Source: the video on YouTube. A channel that would rather this page did not exist can ask for its removal through the contact page, and it is removed.

Transcript

Okay, today I got a brand new workflow that I found. Actually, two workflows that I found. And I'm super excited about these Miniax H3 workflows because they fix an issue that we were dealing with. And I'll show you. I made a video that I'm about to show you, but I want to tell you where I got these workflows from first. So, um, this this guy right here, Sedo Drive, no, Sato Dive, my bad. Uh, awesome, awesome. Totally recommend following him.

Uh, this is the video that he made and I'll I'll link it. And then this is the latent continuation that he created as well. And this is what's so powerful about it. So, um, he made these two workflows and he's got great videos that explain them. I I changed a few little things in there. Um, you know, I love Singularity. I'm I'm sold on it now. And so, I was able to set up the the dual clock node here, uh, so that Singularity works well with it.

So, I changed that and a few other things. But let me go ahead and show you the video that I made and then we'll explain how we made it. And a couple people have asked questions about how you build the clips in the end. So I'll try to include that as well. Hopefully the video won't be too long, but we'll do what we can. All right, so let's watch the video first. >> You ate my cheese. >> It was community cheddar. >> You're cute when you're mad. >> You're buying more cheese.

We already spent all our cheese money. >> Then what's left of us? >> I have an idea. I know how to save the day. >> Really? You do? What do you have planned this time? >> We'll have cheese, milk, and butter for life. >> You couldn't buy cheese, but you think we can afford a cow? >> Buy it? We're just borrowing it indefinitely. >> Okay, so that was it. I think it did really good.

Uh there's a lot of things I like. So, these are the characters at the very beginning. Now, there's still there's still frames like that, but you know, it's motion, so it does pretty good. Um, so that's what he started out looking like right there. And then that's what he ended up looking like right there. So, the issue that I was having with the other workflow that I just love that I absolutely love is it kept building on the I'll call them raster images.

It kept building on the images instead of the latent. And when you do that, um, let me see if I can find it. Okay, I think I found it. All right. So, when you do that, when you just build on the images of the past video, this is this is what happens. So, I'll just show you this right here. All right. So, this is on a different workflow right here. And you don't you don't need the the sound for this, but you just need to see the people.

Okay. So, you see her and him. All right. But then a few if you if you keep using these images and you keep using them. So, this is four down. And look how much more detail has been added to her face versus here. That versus that. Now, you could, you know, you could say, "Oh, it's just zoomed in." But it's not really. Uh, it's it's because it continues to use the image and it continues to to grab more things out of it.

So, the beauty of this other workflow that we're using, let me switch back over, is that this guy figured out how to save the latent. I'm looking to see where it is. Here it is. He figured out how to save the latent. And so when you save the latent, then we're able to bring it into his second workflow here, load up that latent. And the cool thing is that you save it. So if you name it something, then you can come back to it later.

Whereas with these other workflows, I haven't been able to do this. Uh so this is super powerful. Just absolutely awesome. Uh so I think this is going to be my my new my new thing. Now I say that the workflow that I've been using and that I've been loving, he just released an update to it. So this is the one that I've been loving from Fox for Essence Films. and this is his point two or 2.0, but he's still doing the same thing with uploading the video and then enhancing the video and going off of that.

So, I think I'm going to love certain things about this new video, but I think this one with the latence, that's really the answer. So, I think this is going to be it. All right. So, a lot of people said that I can just kind of skim over how to create it. So, I said, you know what? This time, we're actually going to just build a new video so that you can see what we're doing step by step. So, for this, I added all this.

I haven't I haven't put this in yet, but I got it typed up already. So, we're just going to make a 5-second video, and we're going to do it at 640p. Uh, my my system is 16 GB graphics card, and anything over that, it it can do it. 5 seconds at 768, it can do that just fine. But, I feel like for me, 640 is kind of my sweet spot. So, that that's where I'm going to stay. Uh, and so we got three images here. So, we're going to leave all of these set to one right here.

Let's see what else. Resolution is there. Seconds is there. And uh just one thing to keep in mind, whatever your first video is, and that's what I wrote right here. Whatever your first video is, if it's 640, every video after that has to be the exact same. Otherwise, it'll it'll throw up an error. So, let's go over here. And this is this is the the prompt that I created. And then I said, you know what? I only was using these two images.

So, I want to reference the outside of the house from the last video. So, I said, "Hey, should I put this in?" And it said, "Yeah, let's add it." And so, this is it. So, what we're going to do is we're going to copy this right here. And then we're just going to grab that. Actually, I've been leaving a thing up here, style definitions, just just to keep it extra. And notice as soon as I pasted that in, you see how it added these images?

Now, if these were in the wrong order, see how it says one, two, and then you can't see it, but it says three right there. One, two, three. It's got to be in the right order, otherwise it won't have the it won't pick it up in the right way. All right, so we got that. So now, let me go back to chat. And then let me grab this extra little part for picture three. And then I'm going to just put it under subject definitions.

Use image three as this. We'll see if that works. We'll see what it does with that. All right. So then that's good to go. I don't have to change anything over here. I'm going to leave this as is. And this right here, it's just good to to name name your latence. So, I might just say cartoon, if I can spell cartoon couple new, because this is this is actually going to be a new a totally new scene. All right, so we've got everything good there.

Let's just run this and see what happens. One thing I'm I'm going to jump in here to say I also added this right here which is the uh preview just so because one thing I really liked from the other workflow was that I would run it at you know.3 really low and then I would get a couple options and I could see what was going on. So at least with this preview this helps me to to make sure that it's kind of lined up with what I want.

So better than nothing. All right, so that's done. So now let's see what that did right here. >> Why are there so many cheeses? Different cows, different dreams. >> Okay, so just some dumb ridiculous thing, but uh it worked really good. Now, the voices are probably a little bit different than before because there was nothing to reference and that's where the second one comes in. But it created the video. Uh it kept the background very similar and then it also saved the laten.

So that's what you use this first one for. Awesome. I just absolutely love it. Perfect. Okay, so uh one other thing I want to mention here. If you paste in all your your text here and these things don't highlight like this, then what it says to do, and I've saved it right here just to remind us, he told us to just do a cut and paste. And then that'll just tell the system to put it back in place. So that's that's the easy way to fix that.

All right. So then once we've got that, now we come over to our second one. And I put this really big because if you forget, then it won't work. So let's go ahead and hit R for refresh. And we're just like we do with models, we're refreshing it so that it finds the laten. Okay. So see now this is the one we just made. Cartoon couple new. That's not the one I want to use for this. I want to use the one before it. And that's what's so powerful about this is that you can go back if you save these.

Obviously you can tell I stopped naming them. But if you're diligent and you actually save them correctly, then you can go back and you can build from any scene that you've made in the past. I just love this. It's absolutely awesome. All right. So, this is the one we want to do here. So, at the end of the video, remember the guy had the cow. Let me go back just so you can see where we're at. So, we're not on this one, we're on this one.

So, we're right there. Um, and another thing that I messed up on, you notice how he's cartoon, but this turned into like three 3D Pixar. Mini Max wants to do basically like that 3D Pixar look. So, what I should have done is, and I noticed it too here, this is almost realistic in the bathroom. uh any of those things. You just got to prompt ahead of time in a different, you know, with Korea or if you want to use chat DPT, whatever you want to use, but you got to get all your slides in the format you want.

So, I should have got her and the bathroom and the cow, everything in that cartoon format beforehand and used it as references. When you don't do that, you're just leaving Miniax open to its interpretation. So, we're going to go from this last frame right here is what we're going to try to do. And let's see. Let's go back to chat. I had a this right here written. So what I told it was this right here. I said continue outside the house.

I I fed it the last clip, which is what I just showed you. And I said 22 frames of this reference will be used. So then I want the policeman to come up. So this is what it said. It said, you know, the 22 frames will occupy this much. So you get nine new seconds of action. So we're going to go ahead and copy that right there. And then we're going to do the exact same thing. So I'm going to leave the style definitions in there.

And then I'm going to delete that. I'm going to paste that in right there. So, I've got my two references here. I'm not using this bathroom reference anymore. And then I've loaded in this last video that I just showed you right here. Now, let me just take a second to show you a few things. I mentioned 22 seconds earlier, and that's right here. The context frames, you can change it all the way up to 56, but 22 is a good middle ground is what he said.

So, it's taking 22 frames from this video here, which is fine. That's that's exactly it's almost a second, but that's perfect because this is this is where we were at the very end. So, we're going to keep those 22 frames. And then also video output for this is going into seed video and seed reference video. So, one thing I noticed was, let me go back to my video. This part right here, I went from this scene to this scene outside.

So, I switched completely. And yeah, yeah, I know far away. His face is all jacked. So when I did that, I didn't want this video of the bathroom to mess up the new video. And so that's where I put my note here, right here. If you want to reference the video, but you want a totally different scene, which I wanted it outside, then I said to leave it in the seed reference, but to delete it from seed video. And if you watch his video, he also explains why you would use this second one up here.

This is if like your first video is just real closeup. Let's say this is perfect. It's real closeup on the truck or real close up on their faces in the bathroom and the next scene is going to show the whole bathroom. Then you'll need to use this one. You'll plug it into the the seed reference. I know it's backwards, but you would plug this into the seed reference because you want it to reference the whole room instead of just the small part here.

So seed video is basically I know I botched that up but seed video is basically the continuation of the video whereas seed reference is just referencing the images as far as getting a picture of the whole scene. So hopefully that makes sense. If not, put it in the comments. Blast me for it and I'll I'll try to I'll try to do better next time. Okay, so we've got this right here. We've got our latent loaded that we want.

Uh now I will say that the last one, let me go back. What was this one? The last one took 3 minutes to do. When you do a continuation and you're referencing the video, it's going to take longer. And so that's why for my 16 GB system, 640 works pretty good. Um 10 seconds is fine for what what I want for this scene. And I don't think we need to change anything else on here. So, let's just run it. And then there's one other thing at the end that I'm going to show you.

So, let's run it and see see how it does. So, this is why I try to get all my videos done before I'm filming. Uh, I know it says 9 seconds, but let's go into the history. So, it took 17 minutes to run that. And part of the problem was I had Da Vinci open, and I had my OBS recording. Uh, so that slowed it down as well. But when I was watching it, everything was maxed out. My RAM was maxed out, my GPU was maxed out, my VRAM was maxed out.

And so, it's, you know, 16 GB, it's doing the best it can at 640. So, let's let's watch what it did. >> Definitely. >> Sir, where did you get this cow? I'm just borrowing it. >> All right. So, just a little short extra thing. And the cool thing is that it it'll stitch it on to the reference video that we did. And it does this automatically. So, so watch this one. >> We'll have cheese, milk, and butter for life. >> You couldn't buy cheese, but you think we can afford a cow? >> Buy it.

We're just borrowing it indefinitely. >> Sir, where did you get this cow? >> I'm just borrowing it. >> Okay. All right. So, that's pretty cool because now we're going to take it into Da Vinci and I'm going to show you how to to add these things together. Uh, one other thing that I added to this workflow is what I noticed after I run it, uh, the GPU drops back down, the VRAM drops back down, but the RAM was staying peaked.

It just takes longer to clear out the RAM. And so what I did was I just added in this little thing here and it's just any image is fine. And then it just cleans up the RAM. And so all you got to do is just do like that right there and just hit pre play preview. And then it just makes sure everything's dropped back down. So I I think I'm going to start putting this on every single one of my uh workflows because uh 16 gigs you just every single bit matters.

So it's so important. All right. So that those are the two workflows right there. So that that went over everything. Um, so if that's what you needed, great. If you want to see the Da Vinci extra part, then then keep watching. But, uh, like I said, I just think this this dude here just unbelievable work. Sato dive, just unbelievable work. So, totally subscribe to him. Just amazing. Uh, this is going to change everything.

I really think loading latent and being able to save your latent, that's just awesome. And the size of the laten, I checked it was like five megabytes per one. So, it doesn't even take up much space. All right. So, I'm gonna pull up Da Vinci now and we'll see how to add in the video. All right. So now I've got Da Vinci loaded up and I've got the video loaded up. So um these here, let me just zoom out to the whole thing.

Zoom out a little bit more. These I'm just going to grab and I'm just going to get them out of the way because that's that's the completed thing there. All right. So then this was the last clip. Okay. Yeah, I just need to make this smaller. I was getting confused. So these are my subtitles up here. This is the last clip right here. Last whole clip. Yep. All right. So what we're going to do is we're going to drag in the one that we just made.

You could drag in the stitched one that I just made or you could drag in this one here. Doesn't really matter. Uh, whatever whatever works best for you. So, I'm going to just throw this into here. It's 20 right there. And then I'm going to put it just like that right there. So, now we can we can make this a little bit bigger so that you can see everything. Subtitles can get really small. All right. So, then this is where we are.

So, I like to come up here to this thing and say dual viewer. And then I put it right. You could even move this back if you wanted to. I put it right there and I hit F. And what F does is it it highlights this frame. And so now all I got to do is I got to come back in here and it should be about 22 frames is where it is. So if this is 242, then it should be at 220 is where it paused. So you see it's close, but it's not.

Okay. I think that's it right there. I'm just looking at the two. So, so I mean that's that's basically um I did like Seed Hunter better for that because I could actually specify exactly 60 frames. Uh whereas this one, this is 204 is where it's at. And if you if you're not seeing this, you might be seeing the time like that. You can just right click and then say frame. I I like to see the frames more. So then once we got that, then I'm going to hit B and I'm going to cut that.

And then I'm going to hit D. Actually, I'm going to hit A, select it, and then hit D like that. So then I'm going to drag this over to there. And then I'm just going to test it from there. and see. So, I'm gonna just hit space bar. All right. So, that's it. And that's how that's how I did the whole thing. And so, you see, uh, when I load in a new clip, I just add it to a second, you know, I just go back and forth. I don't have to keep going higher and higher.

But then I just cut it and I just I don't delete it just in case I messed up. So, that way, you know, if you hit D, you turn it back on. D, you turn it off. Uh, but then you can just keep building as you go. And the cool thing about this workflow with the laten is that it's also retaining the voice. And so that's another thing on the other workflow that I would use. If I cropped 60 frames, but I didn't get the voice, then it might not carry it over.

Whereas the laten is so powerful because it carries all that information over. Uh, so hopefully this helps. I know once again ended up being a long video, but uh, there were a lot of parts to it that I thought were that were helpful to me. And like I said, watch his video as well because he goes into these things and I'm sure he he goes into things that I missed as well. So hopefully that helps. And probably the next one, what I was thinking I might do is a comparison between this video and then the updated 2.0 of the one that I used to like.

So that might be one of the the next videos that I create. So thanks for watching. See you next time.

The words are the caption track's own and nothing is reworded or re-transcribed. Paragraph breaks are placed between sentences so the text reads as prose.

Use this transcript

Three free tools that work on the material around a video like this one. No signup, no login.