Getting the transcript
Reading the captions from YouTube. A video nobody has opened here before takes 10 to 30 seconds; this page fills in on its own.
Getting the transcript
Reading the captions from YouTube. A video nobody has opened here before takes 10 to 30 seconds; this page fills in on its own.

Stefan 3D AI · @stefan_3d_ai
This video has no Most replayed graph yet: YouTube shows one only once a video has enough views. These are the moments viewers replayed most in Stefan 3D AI's most watched videos.
Most replayed moment at 4:08
7.0x that video's typical replay level
and it will also have the access to full ai apis. Now let's do some serious tests. The first test will be to do an animation of a character that will be split in parts. The first test will be to create animation. Animation of a character that will be an
Said at 4:01
Most replayed moment at 9:39
5.7x that video's typical replay level
Overall I think this guy's made really amazing job, bring it it all together and I really had a lot of
Said at 9:41
Most replayed moment at 20:36
4.1x that video's typical replay level
Just also because I was playing that game when I was kid. Okay, let me see if I remember how to play that. Let's start with the road, right? Let's try to build a, oh, there are cars. There are cars. It generated nice cars and they're like, wow.
Said at 20:29
The graph counts replays. It does not show where viewers stopped watching.
Words
3,142
Runtime
16:43
Speaking pace
188wpm
Reading time
13min
188 words per minute, between the 181 median and the 201 75th percentile of 349 measured videos. That distribution comes from the 349-video hook study.
Opening (first 30 seconds)
Claude is back. I was really not expecting that Opus 5 .5 will be such a big one because it's not a major release like a GPT -6, but it's really good, especially in 3D. Just let me show you. I asked Claude and GPT to create me a scene from a famous movie with the DeLorean and the clock tower. It's quite interesting and dynamic scene potentially. So let's see who going to do that better. Of course, I provided with full package, Blender access, all the APIs and boom, Astro decided to
94 words, the words spoken in the first 30 seconds at 188 words per minute.
Free, no signup. See how the first 30 seconds hold attention, with rewrites.
Sentence shape
| Measure | This transcript |
|---|---|
| Sentences | 425 |
| Average words per sentence | 7.4 |
| Longest sentence | 38 words |
| Questions asked | 21 |
| Sentences containing a number | 34 |
Most used terms
Filler phrases
176 in total: like 104 · actually 40 · I mean 10 · you know 7 · right? 6 · sort of 6 · basically 3.
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.
Free, no account. See where attention is likely to drop, with a rewrite for each weak line. The free check shows the scores and the one issue costing the most. Or run it on the words above first.
Free · No login · See a sample audit first if you prefer.
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, published by the channel, 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.
Claude is back. I was really not expecting that Opus 5 .5 will be such a big one because it's not a major release like a GPT -6, but it's really good, especially in 3D. Just let me show you. I asked Claude and GPT to create me a scene from a famous movie with the DeLorean and the clock tower. It's quite interesting and dynamic scene potentially. So let's see who going to do that better. Of course, I provided with full package, Blender access, all the APIs and boom, Astro decided to not use any side API and generate anything.
So it actually did everything by its own, like not even a single image. So you can see the whole time less what it did. And this is the result. OK, fine, I guess. Buildings or surroundings, not bad. But DeLorean is not really that accurate. Here's what Opus 5 .5 did. Here's the time lapse. Opus 5 .5 also did most of the things procedurally. He only generated some images with Nano Banana Pro. And look at this results.
Crazy thing. And this final action with fire traces. Insane. Like, wow. I mean, I was really not expecting to see any of this. At most something like Astra. And you can see the stats. Both of them were working around one and a half hour, spending around 25 bucks. And you know, guys, this is actually interesting because Opus series is not Frontier series such as Fable. Well, Astra is Frontier series. Also, within announcement, we know that newest Opus will actually cost much less.
Well, also, they compare themselves by performance with GPT -6 Astra. And many numbers are actually higher. Well, again, they're sort of not the same tier. And by the way, OpenAI same day also dropped Soul and Luna version for GPT -6. But here's exactly a second tier models for everyday needs, I guess, because this model basically can tuning Soul and Luna series with reduced cost, which is amazing. You can call of them become cheaper.
But even OpenAI says that they are less in performance than GPT -6 Astra. So in this video, we're going to compare Opus 5 .5 with GPT -6 Astra only. Anyway, there are more crazy tests to see like this one. How about we ask it to procedurally create Notre Dame? You might actually get this famous cartoon where I actually got this reference from. OK, so let's see what they cooked. Let's start with GPT -6. You can see the time lapse.
Exactly how it builds in, you know, like it's very interesting to see because they do use different approaches, like some build from primitive, some with splines, etc. And by the way, to do all this test, I actually using my own workspace, which I recently shared, and it actually packed with plenty of MCP, such as Blender MCP. I really think that's the smartest way to actually use this AI, not. computer. Use And one of my favorite plugin and MCP is fixed, which I going to use most of the time.
Actually, Blender plugin itself is. Absolutely free. It's just expose some MCP tools for your agent. And you can also connect your Hicksfield account and get access to 3D. Generation 2D generation, video generation, all in one place, through Blender or through your agents if you use Hicksfield CLI. But even that is not all what Hicksfield offer right now, because they just launched Hicksfield API where you can get even more.
I. Really recommend you to give it a try. Links will be in the description. And thanks Hicksfield to sponsor this video. So this one is versioned by. GPT -6 and it's sort of fine. I don't know why, but it feels like it lacks some details. Although. Like GPT -6 usually adds a lot of details. Let's take a look on Opus 5 .5. Here we can see the time lapse, how. It builds things. I actually like the scene itself that like around the river.
I think that's more accurate to real Notre. Dame And here we go with Blender file. All right. I don't know. It just feels like here around there are more details. And I don't know, maybe like the surrounding is better like for me. Oh, look like how much birds there are. Like, wow, like there's a lot of birds. OK not sure, not sure in this case. Look like. they. Both did not bad job. I mean, let me know in the comments which colors you like more, because I think like this one is more.
Natural Like it tried to make a sunlight, etc. And it feels like there are more details. I don't know, or they are more accurate. Can't. Really tell who win this one. But what I can tell is that GPT -6 spent twice less time on this task. Forty two. Minutes against ninety six. Interesting. I can all test that around around on Max effort. Let's do another one. How about we ask. Them to create a ride? Well, I chose this Cadillac Eldorado.
I think that's amazing, right? Let's show you what this boy is going to. Create Here's the GPT -6 time lapse. And here we go with the results. Oh, well, I mean, it has taste. Wow. I. Really like this real lights. No way. It looks so sick. I really want to see what Opus created. Here's the time lapse and here's the results. All. Right I mean, this is like neat and clean. But I immediately feel there are like Astra added more.
Details Right here. You can see both of them on the screen same time. And I think, yeah, maybe actually to keep that in mind, we're going. To trace how much details they put and for expenses they both spend pretty much same time and approximately cost the same. Well, actually,. GPT costs a little. bit more. But having a car as a rider is pretty basic, right? How about having a pirate galleon as a ride? Let's see how this guy is going to handle this one.
Also, must be completely procedural. Starting with Astra. Here we can see the time lapse. Actually, this time lapse is one to watch. And here's the result. Well, this one is quite cool, I think, because there's a lot of symmetrical details that you can mathematically describe. And the result is quite decent. Like I really like this one. Let's compare it with the Opus. Here's the time lapse from Opus. OK, also not bad.
We can see all the iteration it did. And here is the results. All right. Well, I think in this case, GPT -6 is more impressive. Well, I really like it more in terms of expenses by time. It's a tie. Like basically they spent one hour exactly. But in terms of costs, it seems like GPT -6 is cheaper twice. Very interesting. Next one, you know, I'm actually building my game and I really, really want to try how good it will handle the effects.
Because I really need the effects for my game right now. And by the way, guys, if you still not follow me on X, it's really a best time to join because I share a lot, including this test there on X and how I progress on my game. I'm really taking that serious. Now, the Test is not really easy. We have magic circle. We have wind with leaves. We have some ice shards attack and some shaders. Let's take a look. And guys, you're going to be surprised.
Here's what Astro made. And this already looks super fun. Look, wow, I'm actually currently creating the effects with Astro. So, yeah, I mean, that's super cool. Right. It basically created a complete scene, a sequence of this VFX is all together. And I think that's amazing. And the materials like they are sick. Look at this fireball. I mean, layers or particles all fit so nice. The magic circle like, wow, wow. I really like that.
I mean, like, I'm going to use this with purpose, but I already saw what Opus created. Wow. This camera animation, like when I first see like, like what? And this leaf animations. Wow. That's super sick. And guys, you know, this is actually a rendering in real time in 3DS, like in my browser. I'm just opening the web page. Dang, that's crazy. I think like for artistic stuff, Opus five definitely a winner. Maybe like, I mean, some effects like these balls.
Maybe GPT 6 did a little bit better job, but the intro, this magic circle is 100 percent super cool and much better in Opus point five. Wow. If you were ever struggling with VFX and stuff, guys, it is now sold, I think like, I mean, you just need to imagine some stuff. It will. Be able to help. And it was actually quite easy task for both these AIs. They both spent around 30 minutes of time and you. Can see not much token spent.
I also tried to create complete games, but before I show you that, there is one another font that how good. It actually understands 3D, because you guys fairly admit that Claude doesn't have vision like that good as. GPT So how about I give it like few 3D models and ask it to unwrap? That is quite challenging. Trust me. And all. 3D models, by the way, I created right here in Hicksville plugin. You can just drop the image here and choose one of the available model.
And generate right here and it will be added to the Blender or even access any newest model right here through this. Plugin like GPT -6 Astra and Opus five point five. Now, let's take a look. Cape haha. OK. Well,. Doesn't look any good, right? First of all, you can see like I provided like separate objects and it's like separated. Additionally It's quite interesting. Like it separated this thing and this thing. OK, that's cool, right?
But it's sort of useless. Because like UV still mess. Like if you see something like this, this is it's not acceptable. Let's take a look on the boots. All. Right So not much better, really. And I don't really understand why it doesn't fit it like in whole square. Why? The. Bag is absolute nightmare. Like this is absolutely terrible. No, not good at all. Maybe close now. No no, no. Well, Astra failed this task. Opus five point five.
OK, so objects are not split. I appreciate that. OK, here we can see the Cape. We can see the results. Well, it's not that crazy, but still again,. These small islands. That's not really acceptable. OK, let's take a look on boots. OK. Well, I also don't get it. Like why. They do not use all the space. Why? No, it's like a common issue. Look at this. I think again, like Opus five did like a. Little bit better job, but still it's a disaster.
Look at many small islands. Only one piece that looks more or less. Decent Actually, probably this is like a decent if it could pack it nicely to take all space, then I would say like that was a win. But yeah, that's like the only close. The boss actually failed it. Astra with vision still didn't manage that. And that actually took. 20 minutes and 10 bucks of each. But how about animations and creatures animations? I actually tried that with Astra and you know already that.
That actually works and that's crazy good, but I still was spending time to iterate and do some fixes. Let's see if Opus. Will handle that better or at least the same. OK, so here's my bad boy. Also, I generated right here. Quite simple,. and you can create both image and 3D models right here. OK, this guy, by the way, is like sort of like Opus five point five creature. You know, like I guess I'm going to put it on a thumbnail.
First of all, you can see that there is a decent trick, which is amazing, but we already saw that. Let's take a look on the animations. OK, then I really like the walk animation. And you can see like it also does a little bit until it's slow, but it works. Jump. Oh, my days. Yeah, this is not a jump. It's very bad. It's failed. Attack. OK. Yeah, I mean, like with that reference again, like it sort of gets not really impressed.
OK, walk animation. Yes. A rig not bad, but up to animations not really work for me. Opus five point five. OK, for the rig, it's pretty much the same again. Like you can see controls here, here, like the pretty much same thing. OK, let's review the animation. I can't wait. Walk animation. All right. Let me turn off like the control. OK, first of all, it's in place, which is not bad. I think like the movement here is a little bit too harsh, too sudden.
Although, again, like if this will be normalized, it's quite good walking animation and in place jump. OK, well, but jump animation is actually much better than GPT 6. I'm not sure they usually jump like with two. Legs simultaneously. Need to review that and the attack animation. Well, this one is much better. That's what I'm talking, guys. Like we. Do have a solid competition here. Like this one is super cool. I'm happy just because I'm going to use it for my.
Game And this is smart enough so I can guide it to do plenty of animations, VFX, etc. Opus was working for 46 minutes,. GPT -6 for 63 minutes, and you can see all the expenses. Well, creatures are fun, but how about humanoids? How. About Lego me? I think I never tested something like that. Here's the prompt and I provide it with the 3D model of me as Lego. Here we. Go Here is how GPT imagined me walking. All right. And cracking down.
All right. Let's bring back. OK. Well, I think. Lego animation can definitely be much more fun. And Opus five. OK, let's do the walking. Crack down and. Bring back. All right. Well, actually, maybe Opus five is a little bit better. But I actually thought like they will do. Some more funny thing because like I know like Lego is limited by structure. But, you know, like when you watch Lego movie and stuff like there. Are much more funny.
Anyway, to the next one. Now let's escalating a bit. How about we create a castle extremely famous castle? But. There will be my channel's name under the task is on the screen. Again, all procedurally. And. We're going to ask to do the render. And I also asked both of them to create a time lapse of how they actually built it. Let's start. With GPT six. Here is the time lapse. By the way, GPT six did wide detail time lapse.
And this is really fun to watch. And here. Is the result that not bad, right? It's really not bad. And actually, I posted in the next and some of you said like they. Really like this version better than Opus five. Because look at this. Here is the time lapse. Not that detailed time lapse, to be honest. But the results is sort of much better. Look at this. I like the fireworks and I like how my name. Appears Maybe GPT six actually made more detailed castle.
But this just looks better in dynamic with all. These effects. I don't know. Just drop a comment. What you think about it, which one is the best? But I definitely going to use both of them for my work. And one more. Test that I really wanted to create for a while. This look like this is about to create a game playable as one 3G This HTML file. But the. History is that someone actually made this already three months on eggs.
And there were a lot of people saying that actually it's not really like that. And I also tried to made it that time. And Fable five was not able. to. Do such a smooth experience, especially from one prompt to believe whether that was true or not. It's for you to decide. But. I gave exactly the same prompt to my guys. And here's what they built. GPT six drop in. Let's go. OK All right. This is actually GPT six Astra.
And that test was made with Opus five, like three. Months ago. I don't know. I guess like I do bad prompting. Maybe maybe that's possible. But yeah, here's what I got actually, like from one. Prompt It's very important. OK, like I think that this one looks quite boring. What I can do and why is stuttering like. Crazy Oh, I can jump actually. All right. OK. All right. This is actually quite boring. Like,. Oh well, OK.
So will I be able to jump like I'm almost stopped now? Let's take a look what. Opus created. All right. Well, first of all, I really like the intro and there is no play button. It just starts. OK. And there is like a. Sounds that's quite noisy and unmuted. OK. Oh, wow. But this one is actually more fun and. Way more dynamic. OK. And I can see the trampoline. Let's approach it. Let's go. Oh, nice. This one is much more fun.
You also can see like that it builded from primitives, right? Which is expected, but it. Plays really nice. I don't know, like it really gets some stuff. OK, I can. do like that. Oh, wow. But this is much more fun than GPT 6. Not going to lie. Well, I actually like this one and I can actually jump in place. I can do like that. Great. Nice. Crazy. Boom. For this test, this one is my favorite 100 percent. And I think I know why Opus actually did it better.
It spends almost three times more time and also cost almost three times more. But it did the job. And again, both of them was running on max effort with exactly the same prompt. In next video, I definitely going to test it in game engine, such as Unity and maybe Unreal Engine. Well, let me know in the comments, what should I test? Like what crazy stuff? But for now, that's it. Like and subscribe. And I see in the next video.
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.
Free tools for your own script: paste a draft and see where it stands before you record it.
Paste your draft and see where viewers are likely to drop off, with a rewrite for each weak line.
Paste the first 30 seconds of your own draft for a hook score and rewrites.
Check your draft against YouTube's advertiser-friendly guidelines before you record it.
Read this channel's public videos and transcripts, and download a writing brief for it.