How are people creating +20 minutes videos using only AI? Veo 3.1 hacked accounts?

[drakos]

So I keep seeing YouTube niches where most of the videos are created using 100% models like Veo 3.1 or Kling 2.6
I'm doing research for a new niche and when I calculated how much it will cost me to generate a 15-minute video using 8 seconds clips from Veo 3, the amount was absurd (I need audio on the clips).

I know there is Meta AI that animates your images, but it's without sound. There is Grok AI that animates them pretty good, but there are limitations to it.
So, I know for a fact that these niches are using this type of models, but with simple calculations (using 8s clips to create the videos), I can see the videos are SUPER expensive to make. Even most of them are not even worth it.

There must be a catch here. I heard about Google accounts with unlimited plans or AI platforms that provide this AI models with a discount, but I don't really know much about them.
Does anyone know how this is done? Any recommendations of sites where its much cheaper?
Thanks!

 

[IM Dude]

If they are using veo3 they are using gemini ultra account. There is some way to get untra account cheap or free(student account??). That's the only way to to generate more than 4 8sec videos in a day.
With a ultra account you can generate 160 mins per month. ((25000/20 )*8)/60 = 166.67
Using veo 3 fast.
Which is insane, if you are a professional videographer/movie maker this is a golden opportunity. That's a chance to create a whole movie.

 

[drakos]

If they are using veo3 they are using gemini ultra account. There is some way to get untra account cheap or free(student account??). That's the only way to to generate more than 4 8sec videos in a day.
With a ultra account you can generate 160 mins per month. ((25000/20 )*8)/60 = 166.67
Using veo 3 fast.
Which is insane, if you are a professional videographer/movie maker this is a golden opportunity. That's a chance to create a whole movie.

I already tested the student account and It's amazing, but not nearly close to produce 15-min videos consistently. It's enough maybe for one video (taking into account possible failures and re-generations), yet many of these channels are posting each 2 days. Seems crazy to me, how are they doing it without blowing so much money...

 

[Amaz0n]

I already tested the student account and It's amazing, but not nearly close to produce 15-min videos consistently. It's enough maybe for one video (taking into account possible failures and re-generations), yet many of these channels are posting each 2 days. Seems crazy to me, how are they doing it without blowing so much money...

More student accounts?

 

[Proxy Cake]

No real catch, most just use a paid plan and reuse lots of clips then add the sound later in editing instead of generating audio every 8 seconds.

 

[drakos]

More student accounts?

how easy is to get students emails?

 

[alienacademy]

If they are using veo3 they are using gemini ultra account. There is some way to get untra account cheap or free(student account??). That's the only way to to generate more than 4 8sec videos in a day.
With a ultra account you can generate 160 mins per month. ((25000/20 )*8)/60 = 166.67
Using veo 3 fast.
Which is insane, if you are a professional videographer/movie maker this is a golden opportunity. That's a chance to create a whole movie.

You get only 1000 credits per month on student account but that should also work if you can register multiple accounts.

I already tested the student account and It's amazing, but not nearly close to produce 15-min videos consistently. It's enough maybe for one video (taking into account possible failures and re-generations), yet many of these channels are posting each 2 days. Seems crazy to me, how are they doing it without blowing so much money...

With multiple students/pro accounts or cheap ultra accounts.
I myself run AI YouTube channels and create both short and long videos.

 

[IM Dude]

I already tested the student account and It's amazing, but not nearly close to produce 15-min videos consistently. It's enough maybe for one video (taking into account possible failures and re-generations), yet many of these channels are posting each 2 days. Seems crazy to me, how are they doing it without blowing so much money...

Student , pro, ultra all give you access to the same veo 3.1. It has 2 versions , fast and quality. Fast takes 20 credits and quality takes 100 credits. Like I said if you have some kind of real video production experience it goes long way to properly prompt veo3 to make consistently similar video. You gotta learn all the lingo of the video production game. Like say a dialog in particular way etc. You can also create all the images of your characters in nanobanana for consistency throughout the video.

 

[VelocityServ]

additionally, they're more likely using google cloud trial credits, which is far easier than being bottlenecked by daily restrictions on any account type.

 

[splishsplash]

So I keep seeing YouTube niches where most of the videos are created using 100% models like Veo 3.1 or Kling 2.6
I'm doing research for a new niche and when I calculated how much it will cost me to generate a 15-minute video using 8 seconds clips from Veo 3, the amount was absurd (I need audio on the clips).

I know there is Meta AI that animates your images, but it's without sound. There is Grok AI that animates them pretty good, but there are limitations to it.
So, I know for a fact that these niches are using this type of models, but with simple calculations (using 8s clips to create the videos), I can see the videos are SUPER expensive to make. Even most of them are not even worth it.

There must be a catch here. I heard about Google accounts with unlimited plans or AI platforms that provide this AI models with a discount, but I don't really know much about them.
Does anyone know how this is done? Any recommendations of sites where its much cheaper?
Thanks!

Ultra plans. You can get multiple 3 month discounts. The UK ones are only £80/mo for 3 months. US is iirc $140. I have an ultra, but I don't use it much. I have 3 claude code max and a gpt pro which is more useful for what I do.

But with the ultra you get 333 minutes of veo fast, or 33 minutes of quality.

So they're likely using the fast generation. 15-20 15-20 minute videos per month.

And a lot of people probably re-use the same clips and merge them in different ways.

 

[Segedino]

Student , pro, ultra all give you access to the same veo 3.1. It has 2 versions , fast and quality. Fast takes 20 credits and quality takes 100 credits. Like I said if you have some kind of real video production experience it goes long way to properly prompt veo3 to make consistently similar video. You gotta learn all the lingo of the video production game. Like say a dialog in particular way etc. You can also create all the images of your characters in nanobanana for consistency throughout the video.

I learn something more today, I usually use the color grading terms and it works, I dont know it responds to lingo too

 

[SproutGigsTeam]

Please note that AI isn't limited to online models; there are local models you can combine to create videos with audio, achieving acceptable quality if you have good hardware. This can be done using WAN 2.1 or 2.2 and Stable Audio for sound creation. There are also models like InfinityTalk that allow you to create podcast-style videos with multiple speakers. (These models are just examples; more local models exist.)

 

[Peter Ngo]

There’s no secret unlimited plan or magic account. The math you did is correct. Generating a full 15 minute video with 8 second AI clips plus audio using models like Veo or Kling gets expensive fast and most of the time it’s not worth it.


From what I’ve seen and tested, the “catch” is usually one of these:


Most channels are not doing 100 percent AI video, even if it looks like it. They mix short AI clips with stock footage, static images, zooms, cuts, overlays and fast pacing. The AI part is just filler, not the entire video.


Many of them don’t generate audio in the clips. They use narration, background music and sound effects on top. Viewers rarely notice that the original clip has no sound.


Another common thing is using longer clips and reusing them. Instead of 8 seconds everywhere, they generate fewer clips, stretch them with cuts, slow motion, zooms or loop small sections.


Some creators also use cheaper tools like Runway or Pika for parts of the video and reserve expensive models only for key scenes. Quality difference is acceptable if editing is decent.


And a lot of these videos simply aren’t profitable. People assume they are killing it, but when you run the numbers, many channels are burning money or barely breaking even.


Bottom line, there’s no hidden discount that changes the economics. If the niche can’t support the cost, the model doesn’t scale. Most successful channels reduce AI usage instead of going all in.


Hope that helps.

From memory, it is also super hard to monetise with pure AI content, right?

 

[drakos]

You get only 1000 credits per month on student account but that should also work if you can register multiple accounts.



With multiple students/pro accounts or cheap ultra accounts.
I myself run AI YouTube channels and create both short and long videos.

Would love to hear more about how are you managing the creation of your videos as I think we are in similar niches (if you are open to talk about it), can I dm you?

Student , pro, ultra all give you access to the same veo 3.1. It has 2 versions , fast and quality. Fast takes 20 credits and quality takes 100 credits. Like I said if you have some kind of real video production experience it goes long way to properly prompt veo3 to make consistently similar video. You gotta learn all the lingo of the video production game. Like say a dialog in particular way etc. You can also create all the images of your characters in nanobanana for consistency throughout the video.

Yes! That was important to me, optimizing the actual settings of Veo to reduce credits consumption. But, even in the lowest possible (1 result,720, Veo 3.1 Fast), it is still 20 credits for 8 second. Enough for shorts but not nearly for long format.

Please note that AI isn't limited to online models; there are local models you can combine to create videos with audio, achieving acceptable quality if you have good hardware. This can be done using WAN 2.1 or 2.2 and Stable Audio for sound creation. There are also models like InfinityTalk that allow you to create podcast-style videos with multiple speakers. (These models are just examples; more local models exist.)

That is what I was expecting, thanks for the answer! I tested some of these models on hugging face spaces, but they seem pretty bad. What are the ones that do you recommend? And what is the expected hardware for some of them? Also, it's even worth it to rent cloud space to host it? I experience something like that when self-hosting a text-to-speech model and it was pretty expensive and not that well performance.

There’s no secret unlimited plan or magic account. The math you did is correct. Generating a full 15 minute video with 8 second AI clips plus audio using models like Veo or Kling gets expensive fast and most of the time it’s not worth it.


From what I’ve seen and tested, the “catch” is usually one of these:


Most channels are not doing 100 percent AI video, even if it looks like it. They mix short AI clips with stock footage, static images, zooms, cuts, overlays and fast pacing. The AI part is just filler, not the entire video.


Many of them don’t generate audio in the clips. They use narration, background music and sound effects on top. Viewers rarely notice that the original clip has no sound.


Another common thing is using longer clips and reusing them. Instead of 8 seconds everywhere, they generate fewer clips, stretch them with cuts, slow motion, zooms or loop small sections.


Some creators also use cheaper tools like Runway or Pika for parts of the video and reserve expensive models only for key scenes. Quality difference is acceptable if editing is decent.


And a lot of these videos simply aren’t profitable. People assume they are killing it, but when you run the numbers, many channels are burning money or barely breaking even.


Bottom line, there’s no hidden discount that changes the economics. If the niche can’t support the cost, the model doesn’t scale. Most successful channels reduce AI usage instead of going all in.


Hope that helps.

Appreciate so much your answer! You are 100% right, even tho the channel I'm talking about is doing everything with sound and each clip is different, you are right about the "no sound" clips and the re-use of already generated clips to stretch the video.
I might need to analyze some more videos to see all the tricks they are doing...
The main issue is that my niche is completely untouched. There is no channel doing what I'm going to do, so there is no real example of how it must look, but Im sure this tricks can also be applied.

I think I can shed some light on this. It's interesting that you're seeing many channels posting high-quality videos consistently, despite the costs. I've heard of people using Google Cloud trial credits, as mentioned earlier, to get around the daily restrictions on AI model usage. However, I'm curious to know if anyone has explored other options like AWS or Azure, which might offer similar discounts or promotions. Have you considered reaching out to the creators of these channels and asking about their production process, or do you think they'd be tight-lipped about their methods? Additionally, I'd love to know more about your research on YouTube niches and how you're planning to utilize these AI models in your own content creation.

So, there is the Google Student account where you get a bunch of credits. That is one way.
Then, I recently discover the Google Cloud Credits free tier, they give you $300 in credits that can be used on Vertex to create content (this one is amazing)
I know for experience, that using GitHub student, can get you free credits on Azure. And Azure has AI models also.
Lastly, there is the AWS Free Tier, I Think it is like $150 or $200, and they DO have many AI models also there.
Maybe combining ALL this can get you somewhere. Hope it helped, man!
In relation to my niche. Its completely untouched, I could not find long format videos or channels doing exactly what I'm going to create.

From memory, it is also super hard to monetise with pure AI content, right?

I'm pretty sure It's hard, yes. Not because of the actual content as many of the content we consume on YouTube its mainly AI or similar, but more because of the competition. Everyone is creating content with AI nowadays, so monetizing from zero a channel with it its kind of hard...

 

[komedya]

I'm also very curious about the answer to this. My research is still ongoing. I will write if I find an answer.

 

[Davidwolf]

ForgotToHitSave already dropped a really solid reality check and I’m with him on that. The math is brutal and there’s no secret magic account that suddenly makes 15 min fully AI videos cheap and scalable without cutting corners somewhere.

Also just to clear one thing up, YouTube is not “penalizing” AI content by default, what they care about is whether the content is actually useful or engaging. You can monetize AI content just fine, but it has to be very relevant to the viewer and not feel like generic filler. Pure AI doesn’t kill channels, boring AI does.

Since your niche is untouched and you don’t really have references, I’d actually study channels that are clearly AI heavy, even if they’re in different niches, and reverse engineer the structure, pacing, reuse of scenes, how long they stay on one visual, how they hide loops, how much is narration vs visuals. The tricks are usually in the edit, not in the model.

On the cost side, like others said, most people are using fast modes, reusing clips, adding sound later, mixing in stock, images, zooms, motion effects, and only using expensive models for hero shots. Full Veo or Kling for every 8 seconds with audio is basically the Ferrari way of doing Uber Eats.

Also don’t sleep on Chinese models. Stuff like Wan and Qwen based tools are improving fast and a lot cheaper for certain styles, especially if you’re okay with slightly less cinematic output but more volume. For some niches, consistency and frequency beats ultra cinematic anyway.

Another thing people underestimate is prompting like a video person, not like an AI person. If you describe shots, camera movement, lighting, framing, continuity, you get way more reusable footage and fewer failed generations. That alone can cut your costs a lot.

And yeah, some channels are probably not even profitable, they just look good from the outside. Happens a lot in AI YouTube, everyone assumes the guy posting every two days is printing money, meanwhile he’s feeding credits to the machine and hoping AdSense catches up.

If your niche is truly new, you actually have an advantage because you don’t need movie level visuals to validate the idea. I’d test with cheaper workflows first, see if people even care about the concept, then upgrade production once the numbers make sense. Better to be early and rough than cinematic and broke

You’re thinking about the right problems though, which is already more than most people do before they start burning credits.

 

[drakos]

ForgotToHitSave already dropped a really solid reality check and I’m with him on that. The math is brutal and there’s no secret magic account that suddenly makes 15 min fully AI videos cheap and scalable without cutting corners somewhere.

Also just to clear one thing up, YouTube is not “penalizing” AI content by default, what they care about is whether the content is actually useful or engaging. You can monetize AI content just fine, but it has to be very relevant to the viewer and not feel like generic filler. Pure AI doesn’t kill channels, boring AI does.

Since your niche is untouched and you don’t really have references, I’d actually study channels that are clearly AI heavy, even if they’re in different niches, and reverse engineer the structure, pacing, reuse of scenes, how long they stay on one visual, how they hide loops, how much is narration vs visuals. The tricks are usually in the edit, not in the model.

On the cost side, like others said, most people are using fast modes, reusing clips, adding sound later, mixing in stock, images, zooms, motion effects, and only using expensive models for hero shots. Full Veo or Kling for every 8 seconds with audio is basically the Ferrari way of doing Uber Eats.

Also don’t sleep on Chinese models. Stuff like Wan and Qwen based tools are improving fast and a lot cheaper for certain styles, especially if you’re okay with slightly less cinematic output but more volume. For some niches, consistency and frequency beats ultra cinematic anyway.

Another thing people underestimate is prompting like a video person, not like an AI person. If you describe shots, camera movement, lighting, framing, continuity, you get way more reusable footage and fewer failed generations. That alone can cut your costs a lot.

And yeah, some channels are probably not even profitable, they just look good from the outside. Happens a lot in AI YouTube, everyone assumes the guy posting every two days is printing money, meanwhile he’s feeding credits to the machine and hoping AdSense catches up.

If your niche is truly new, you actually have an advantage because you don’t need movie level visuals to validate the idea. I’d test with cheaper workflows first, see if people even care about the concept, then upgrade production once the numbers make sense. Better to be early and rough than cinematic and broke

You’re thinking about the right problems though, which is already more than most people do before they start burning credits.

This is it, thanks for the answer! I'm trying to cut some edges as some of you suggested when generating the content. Doing some test right now to see how I can produce the videos. Would it be interesting if a share the things I learn?

 

[Davidwolf]

This is it, thanks for the answer! I'm trying to cut some edges as some of you suggested when generating the content. Doing some test right now to see how I can produce the videos. Would it be interesting if a share the things I learn?

Absolutely

 

[MsLin]

ForgotToHitSave already dropped a really solid reality check and I’m with him on that. The math is brutal and there’s no secret magic account that suddenly makes 15 min fully AI videos cheap and scalable without cutting corners somewhere.

Also just to clear one thing up, YouTube is not “penalizing” AI content by default, what they care about is whether the content is actually useful or engaging. You can monetize AI content just fine, but it has to be very relevant to the viewer and not feel like generic filler. Pure AI doesn’t kill channels, boring AI does.

Since your niche is untouched and you don’t really have references, I’d actually study channels that are clearly AI heavy, even if they’re in different niches, and reverse engineer the structure, pacing, reuse of scenes, how long they stay on one visual, how they hide loops, how much is narration vs visuals. The tricks are usually in the edit, not in the model.

On the cost side, like others said, most people are using fast modes, reusing clips, adding sound later, mixing in stock, images, zooms, motion effects, and only using expensive models for hero shots. Full Veo or Kling for every 8 seconds with audio is basically the Ferrari way of doing Uber Eats.

Also don’t sleep on Chinese models. Stuff like Wan and Qwen based tools are improving fast and a lot cheaper for certain styles, especially if you’re okay with slightly less cinematic output but more volume. For some niches, consistency and frequency beats ultra cinematic anyway.

Another thing people underestimate is prompting like a video person, not like an AI person. If you describe shots, camera movement, lighting, framing, continuity, you get way more reusable footage and fewer failed generations. That alone can cut your costs a lot.

And yeah, some channels are probably not even profitable, they just look good from the outside. Happens a lot in AI YouTube, everyone assumes the guy posting every two days is printing money, meanwhile he’s feeding credits to the machine and hoping AdSense catches up.

If your niche is truly new, you actually have an advantage because you don’t need movie level visuals to validate the idea. I’d test with cheaper workflows first, see if people even care about the concept, then upgrade production once the numbers make sense. Better to be early and rough than cinematic and broke

You’re thinking about the right problems though, which is already more than most people do before they start burning credits.

Thank you for sharing. I’ve burned through so many failed renders despite having several accounts. I’m gonna shift my focus to Wan/Qwen for volume and lean more into editing techniques instead of praying for a flawless AI generation every time.

 

[ltesocksio]

How do i setup a student account fro a cheaper rate?