Confused what to use and what to buy

[Delontekoq]

Hello guys, im looking to start a kids channel with 3D illustrated videos. This is for future project i will include later on.
What i need: YT shorts or longer normal videos, promts with a already made Face as main character, promt " 3d illustriated cartoon pixar style....."

And keep having the same "red" line with different stories.

Ive read on tiktok, here and other forums, but there is a jungle to pick.

Question: What would be the best chooise to pick ?
- Leonardo.ai
- Muapi.ai
- Seaart
- Some say Claude + Higgsfield?

I need some input, what do you think is best for my purpose and pump out as many videos as possible, but not spend $10.000 on AI videos

 

[m4rc0 d14z]

My autistic moment has arrived.

i'm a guy who ilustrates and do animations in Moho, but also uses AI a lot.

so, if you say you want to make videos. the best option, as for now (on quality, prompt adherence, and such) is Seedance 2.0 for videos and (arguably) nano banana 2. Although, the audio can be janky sometimes, so i'm telling you right now, you must, and you should spend some time editing the audio on your video editor of your choice.


I'll assume you don't have a somewhat powerfull pc to do some manual generations.

so your best choice would be: Nano banana pro + Seedance.

why nano banana pro?

mostly to get desings, generate the characters, backgrounds and more using nano banana, and with nano banana you can also generate Model sheet for the characters.

once you get the main desing of the characters with the style you want (pixar,dreamworks, arcane, wathever) you make the concept art. and save the model sheet on a folder with the name of the character.

then, when you get that done, assuming you'll be reusing desings, the only thing that you'll need is, with nano banana is to generate the shots of the scenes. Use these generations combining the model sheet of the character and then the background to get the pose/composition you need. And then, you throw that to Seedance and prompt what you want, and... well... that's it.


IF you do have a somewhat powerfull pc, you can save up costs.

the best way, would be, to download stability matrix, then download stable diffussion web UI forge neo (better UI) and download either anima 0.1 or Z image base/turbo (depending of how much Vram you have) with that, you can prompt the characters and iterate a shit ton of times, generate the backgrounds, and use tools like Canny/lineart (if you have some artistic abilities, or just draw some sticks) so the generation goes the way you want.

The workflow, using local generation + nano banana 2 + seedance should be somewhat like this:


Local iterations of scenarios and characters > nano banana to refine/edit characters escenarios > Nano Banana to generate the shots/input frames for seedance > seedance to generate the videos > Kdenlive/premiere pro to edit the cuts and paste it all together > export.

Now, let's say, you don't really know anything about AI, safetensors, api providers, and blah blah blah.

ok, i'm gonna help you with this.

what about costs? what is the best webpage? is seeart better than leonardo?

First of all, i'll kindly tell you, stop going for places like higgsfield, leonardo, seeart or wathever the fuck else pops out.

why? because those webpages are just aggregators of subpar models that charge you to use local models (most of the times) and their prices are a straight up robery.

the best option, if you really don't know, is just to stick with API providers.

but dude, what is a API provider?

An API provider, is a service that allow you to pay FOR USE on the models, you don't pay subscriptions montlhy, you only pay WHAT YOU USE, so if you are inteligent, API can save up money, or straight up bankrupt you.

Best option around, is fal.ai, is more focused to developers, cause the GUI is ugly as hell, don't worry, you don't need fancy GUI, that's for losers.

With Fal.AI you can go, choose nano banana 2 pro, generate an image with fal.ai costo you around 0.15$ per generation, and that's using the PRO version, when iterating, it's important to keep the flash version and only using the pro when you have the desing you need refined and creating the input shots. otherwise you'll spend a shit ton amount of money just "testing" be careful.

there's others API providers around, like replicate, they are good, but i personally prefer fal, you can test the others.

why using an API provider?

because you can use, within the same page, nano banana, flux, recraft, seedance, kling, google audio tts, 3d models generators, you name it.


But, here's comes the tricky part, I told you to use seedance 2.0 is the best nowadays, top of the art video model, good consistency and shit. But is expensive as hell.

let's say you generate batchs of 10sec clips? ok, if you use Seedance 2.0 img to video, with the standard model, is 3.02$ PER CLIP

if you use the fast model, it's 2.5$ Per 10 sec clip.

so take that in consideration when generating, but the best part is that you can go, and just use seedance for the "important" shots, and just use kling for the rest and save up a lot of money.

The Reality Check: Budget vs. Outputs (10-second clips)​

Let’s talk real numbers. If you have a budget of $20 to $30 a month on Fal.ai, here is exactly what that money gets you.

Scenario A: Pure API Workflow (No Local PC)​

Since you don't have a local machine, you have to burn cash on both image creation and video generation.

• The Math:
  
  • Refining images: You'll use Nano Banana 2 (Flash/Standard) for iterations at around $0.03–$0.04 per prompt. Let's assume you do 10-15 test generations to lock in a character or background before pushing to Pro ($0.15). That's roughly $0.50 to $0.70 spent just to get your final input frame ready.
  • Generating the Video: Using Seedance 2.0 Fast ($2.42 per 10s clip) or Standard ($3.02 per 10s clip).
• Total Cost per 10s Shot: Around $3.00 to $3.70 (accounting for image iterations + one video generation).
• What you get per month:
  
  • With $20: You get about 5 to 6 video clips (roughly 1 minute of raw footage).
  • With $30: You get about 8 to 10 video clips (roughly 1.5 minutes of raw footage).

Crucial Note: This assumes zero re-rolls. If the AI gives you a deformed hand or a melting background and you have to generate the video again, you just burned another $2.50+. Realistically, cut those video counts in half if you are picky with quality.

Scenario B: Hybrid Workflow (Local PC + Fal.ai API)​

If you have a decent PC, you do all your heavy lifting, character design sheets, background painting, and composition tests locally using Stable Diffusion (Forge) for $0.00. You only upload the absolute perfect, finalized frame to Fal.ai.

• The Math: Your image generation cost drops to zero. You only pay for the video.
  
  • Seedance 2.0 Fast: $2.42 per clip.
  • Seedance 2.0 Standard: $3.02 per clip.
• What you get per month:
  
  • With $20: You get 6 to 8 video clips (~1.2 minutes of footage).
  • With $30: You get 10 to 12 video clips (~2 minutes of footage).

 

[Delontekoq]

My autistic moment has arrived.

i'm a guy who ilustrates and do animations in Moho, but also uses AI a lot.

so, if you say you want to make videos. the best option, as for now (on quality, prompt adherence, and such) is Seedance 2.0 for videos and (arguably) nano banana 2. Although, the audio can be janky sometimes, so i'm telling you right now, you must, and you should spend some time editing the audio on your video editor of your choice.


I'll assume you don't have a somewhat powerfull pc to do some manual generations.

so your best choice would be: Nano banana pro + Seedance.

why nano banana pro?

mostly to get desings, generate the characters, backgrounds and more using nano banana, and with nano banana you can also generate Model sheet for the characters.

once you get the main desing of the characters with the style you want (pixar,dreamworks, arcane, wathever) you make the concept art. and save the model sheet on a folder with the name of the character.

then, when you get that done, assuming you'll be reusing desings, the only thing that you'll need is, with nano banana is to generate the shots of the scenes. Use these generations combining the model sheet of the character and then the background to get the pose/composition you need. And then, you throw that to Seedance and prompt what you want, and... well... that's it.


IF you do have a somewhat powerfull pc, you can save up costs.

the best way, would be, to download stability matrix, then download stable diffussion web UI forge neo (better UI) and download either anima 0.1 or Z image base/turbo (depending of how much Vram you have) with that, you can prompt the characters and iterate a shit ton of times, generate the backgrounds, and use tools like Canny/lineart (if you have some artistic abilities, or just draw some sticks) so the generation goes the way you want.

The workflow, using local generation + nano banana 2 + seedance should be somewhat like this:


Local iterations of scenarios and characters > nano banana to refine/edit characters escenarios > Nano Banana to generate the shots/input frames for seedance > seedance to generate the videos > Kdenlive/premiere pro to edit the cuts and paste it all together > export.

Now, let's say, you don't really know anything about AI, safetensors, api providers, and blah blah blah.

ok, i'm gonna help you with this.

what about costs? what is the best webpage? is seeart better than leonardo?

First of all, i'll kindly tell you, stop going for places like higgsfield, leonardo, seeart or wathever the fuck else pops out.

why? because those webpages are just aggregators of subpar models that charge you to use local models (most of the times) and their prices are a straight up robery.

the best option, if you really don't know, is just to stick with API providers.

but dude, what is a API provider?

An API provider, is a service that allow you to pay FOR USE on the models, you don't pay subscriptions montlhy, you only pay WHAT YOU USE, so if you are inteligent, API can save up money, or straight up bankrupt you.

Best option around, is fal.ai, is more focused to developers, cause the GUI is ugly as hell, don't worry, you don't need fancy GUI, that's for losers.

With Fal.AI you can go, choose nano banana 2 pro, generate an image with fal.ai costo you around 0.15$ per generation, and that's using the PRO version, when iterating, it's important to keep the flash version and only using the pro when you have the desing you need refined and creating the input shots. otherwise you'll spend a shit ton amount of money just "testing" be careful.

there's others API providers around, like replicate, they are good, but i personally prefer fal, you can test the others.

why using an API provider?

because you can use, within the same page, nano banana, flux, recraft, seedance, kling, google audio tts, 3d models generators, you name it.


But, here's comes the tricky part, I told you to use seedance 2.0 is the best nowadays, top of the art video model, good consistency and shit. But is expensive as hell.

let's say you generate batchs of 10sec clips? ok, if you use Seedance 2.0 img to video, with the standard model, is 3.02$ PER CLIP

if you use the fast model, it's 2.5$ Per 10 sec clip.

so take that in consideration when generating, but the best part is that you can go, and just use seedance for the "important" shots, and just use kling for the rest and save up a lot of money.

The Reality Check: Budget vs. Outputs (10-second clips)​

Let’s talk real numbers. If you have a budget of $20 to $30 a month on Fal.ai, here is exactly what that money gets you.

Scenario A: Pure API Workflow (No Local PC)​

Since you don't have a local machine, you have to burn cash on both image creation and video generation.

• The Math:
  • Refining images: You'll use Nano Banana 2 (Flash/Standard) for iterations at around $0.03–$0.04 per prompt. Let's assume you do 10-15 test generations to lock in a character or background before pushing to Pro ($0.15). That's roughly $0.50 to $0.70 spent just to get your final input frame ready.
  • Generating the Video: Using Seedance 2.0 Fast ($2.42 per 10s clip) or Standard ($3.02 per 10s clip).
• Total Cost per 10s Shot: Around $3.00 to $3.70 (accounting for image iterations + one video generation).
• What you get per month:
  • With $20: You get about 5 to 6 video clips (roughly 1 minute of raw footage).
  • With $30: You get about 8 to 10 video clips (roughly 1.5 minutes of raw footage).

Scenario B: Hybrid Workflow (Local PC + Fal.ai API)​

If you have a decent PC, you do all your heavy lifting, character design sheets, background painting, and composition tests locally using Stable Diffusion (Forge) for $0.00. You only upload the absolute perfect, finalized frame to Fal.ai.

• The Math: Your image generation cost drops to zero. You only pay for the video.
  • Seedance 2.0 Fast: $2.42 per clip.
  • Seedance 2.0 Standard: $3.02 per clip.
• What you get per month:
  • With $20: You get 6 to 8 video clips (~1.2 minutes of footage).
  • With $30: You get 10 to 12 video clips (~2 minutes of footage).

truly appreciate your in deepth explanation! one thing is clear, i do have to look if my pc can handle the api. Its when people like you actually explain what those others are and how they work, sounds super expensive to stick with them!

 

[Dayyhu]

Hello guys, im looking to start a kids channel with 3D illustrated videos. This is for future project i will include later on.
What i need: YT shorts or longer normal videos, promts with a already made Face as main character, promt " 3d illustriated cartoon pixar style....."

And keep having the same "red" line with different stories.

Ive read on tiktok, here and other forums, but there is a jungle to pick.

Question: What would be the best chooise to pick ?
- Leonardo.ai
- Muapi.ai
- Seaart
- Some say Claude + Higgsfield?

I need some input, what do you think is best for my purpose and pump out as many videos as possible, but not spend $10.000 on AI videos

If your main goal is a consistent character across lots of videos, I'd focus less on which tool creates the prettiest image and more on which one maintains character consistency. That's usually the hardest part.

From what I've seen, Leonardo is a solid starting point because it's relatively affordable and good for creating a repeatable character style. The bigger challenge comes when you want that same character appearing consistently across dozens or hundreds of scenes.

Personally, I'd probably use a combination rather than a single tool. Create and refine the character first, build a character reference sheet, then use video tools to animate it. A lot of people spend months generating random videos only to realize their main character looks different in every episode. Solving that problem early will save you a lot of headaches later.

 

[m4rc0 d14z]

truly appreciate your in deepth explanation! one thing is clear, i do have to look if my pc can handle the api. Its when people like you actually explain what those others are and how they work, sounds super expensive to stick with them!

Uh, quick clarification, API calls run entirely on whatever servers the company (google, alibaba, amazon, etc etc etc) has their processing power. API doesn't cost you because you're running on your machine, it cost you because you are running it on THEIR machine, that's why they charge you because you're not running it on your own local machine. that's why people tend to use API calls when they don't have the specs to run local models.

so either you use a latin american goverment laptop with 2 gb of ram, or a ultra high end pc gaming, the cost for running in API calls will be always, the same.

the struggle comes when you want to use some IA models locally on your pc, locally means you use your own hardware for the Tensors to work and output an image/video/text/audio, the cost of suscription is 0, but the electricity cost goes up to the roof if you use a high end model, and if case you're going for locally, you need to check things like, gpu, ram, vram and all that.

so, again.

API calls = More flexibility, no suscriptions, pay what you use.
locally = Uncensored models, freedom, your pc becomes an oven, and electricity goes up.

right now, models like Z image turbo can output very good realistic output, and some stilizations like 3D, pixar, anime, etc etc, you need to get loras out of that, but model base can do it. but to generate model sheet and consistency on character desing between shots. it's gonna be a pain in the ass.

Anima, is a model focused only on ilustrations/anime output, is very lightweight and its capabilities to understand plain language are great, it's a somewhat small model, so any "30xx" nvidia graphic cards can output great stuff.

i Didn't tell you about local video models like WAN, LTX, or whatever is out there, cause to be honest with you? they are a pain in the ass, to configure to make something and to set up in comfy, and it's going to make you quit the project before you even start. so for video, better go for API, and if you want to use any opensource model, API for those model also exist. and it's cheaper and you'll get better results than local and no headache.


also, I know my explanation may seem scary and complicated. And it is.

but i've been dealing with AI and models and shit since 2023, either local or api, or subscription, I have a very focused workflow to generate my stuff, and I use my method to get the best out of each dollar.


So, if i were to suggest you, a good way to save money? just use API calls dude. With it you can generate the pictures you need and then go to video, easy. you just have to be checking your wallet every now and then.

if you want to save up, more money, you can use the free tier of the pages you mentioned, and iterate there, and use the API calls to refine.

for example, open your gemini app, ask gemini to give you a desing, and iterate, once you get your desing, go to the api, refine it, and generate shots. it's gonna save you a bunch of bucks in the long run.

 

[Delontekoq]

Uh, quick clarification, API calls run entirely on whatever servers the company (google, alibaba, amazon, etc etc etc) has their processing power. API doesn't cost you because you're running on your machine, it cost you because you are running it on THEIR machine, that's why they charge you because you're not running it on your own local machine. that's why people tend to use API calls when they don't have the specs to run local models.

so either you use a latin american goverment laptop with 2 gb of ram, or a ultra high end pc gaming, the cost for running in API calls will be always, the same.

the struggle comes when you want to use some IA models locally on your pc, locally means you use your own hardware for the Tensors to work and output an image/video/text/audio, the cost of suscription is 0, but the electricity cost goes up to the roof if you use a high end model, and if case you're going for locally, you need to check things like, gpu, ram, vram and all that.

so, again.

API calls = More flexibility, no suscriptions, pay what you use.
locally = Uncensored models, freedom, your pc becomes an oven, and electricity goes up.

right now, models like Z image turbo can output very good realistic output, and some stilizations like 3D, pixar, anime, etc etc, you need to get loras out of that, but model base can do it. but to generate model sheet and consistency on character desing between shots. it's gonna be a pain in the ass.

Anima, is a model focused only on ilustrations/anime output, is very lightweight and its capabilities to understand plain language are great, it's a somewhat small model, so any "30xx" nvidia graphic cards can output great stuff.

i Didn't tell you about local video models like WAN, LTX, or whatever is out there, cause to be honest with you? they are a pain in the ass, to configure to make something and to set up in comfy, and it's going to make you quit the project before you even start. so for video, better go for API, and if you want to use any opensource model, API for those model also exist. and it's cheaper and you'll get better results than local and no headache.


also, I know my explanation may seem scary and complicated. And it is.

but i've been dealing with AI and models and shit since 2023, either local or api, or subscription, I have a very focused workflow to generate my stuff, and I use my method to get the best out of each dollar.


So, if i were to suggest you, a good way to save money? just use API calls dude. With it you can generate the pictures you need and then go to video, easy. you just have to be checking your wallet every now and then.

if you want to save up, more money, you can use the free tier of the pages you mentioned, and iterate there, and use the API calls to refine.

for example, open your gemini app, ask gemini to give you a desing, and iterate, once you get your desing, go to the api, refine it, and generate shots. it's gonna save you a bunch of bucks in the long run.

Thanks man! yea it seems a bit complicated, but i think ill have to explore to really understand. i think my PC is ok, i have a 3070 card and somewhat OK ram,,,32. I guess ill just be carefull with the cash and see how it goes