YouTube2Text

Stop Wasting Credits on AI Video Generators Like Seedance 2.5 — Transcript

by Dan Kieft · 3,087 words · 424 segments · language en · Watch on YouTube

Full transcript

  1. 0:12No, no, no, no.
  2. 0:15I need credits. I need to find credits.
  3. 0:30So, there are three methods of how you
  4. 0:31can save yourself money on AI video
  5. 0:33credits that we'll be covering in this
  6. 0:35video. We'll be using Higgs with for
  7. 0:37this as in my opinion, this is one of
  8. 0:39the best AI video generator tools out
  9. 0:41there and they have created a new method
  10. 0:43that will save you a lot of credits
  11. 0:45which I'll be covering later. I will
  12. 0:47leave a link to them in the description
  13. 0:49down below. The first method is choosing
  14. 0:51the right model and this often gets
  15. 0:53overlooked by a lot of people. Everyone
  16. 0:56nowadays thinks they have to use
  17. 0:57Seedance 2.5, but to be honest, it's
  18. 1:00expensive as [ __ ] A different model can
  19. 1:02easily achieve 80 or 90% of the same
  20. 1:05result and it's [music] a lot cheaper.
  21. 1:07So, let me show you an example of what I
  22. 1:09mean by that. So, I have one video that
  23. 1:11is generated using Seedance 2.5 and the
  24. 1:14other video is generated using an older
  25. 1:16model which is Kling 3.0 which is
  26. 1:19severely outdated by now. I won't tell
  27. 1:21you which one is which yet, but you tell
  28. 1:23me which one you like the most and which
  29. 1:25one you think is Seedance.
  30. 1:40Now, the first video was Kling 3.0 and
  31. 1:42the second video was Seedance 2.5. To
  32. 1:45me, these results are really similar,
  33. 1:48but this generation in Kling cost me 12
  34. 1:50credits while the same generation in
  35. 1:52Seedance 2.5 cost me 72 credits. That's
  36. 1:56the difference of 60 credits. So, if we
  37. 1:58compare the two, I can generate six
  38. 2:00videos using Kling 3.0 for the same
  39. 2:03price as one video on Seidance 2.5. And
  40. 2:06that's why you got to identify when you
  41. 2:08can use a cheaper model. Because some of
  42. 2:10the models might not be suitable for
  43. 2:12your generation. For example, let's take
  44. 2:14the topic of anime. I've done a lot of
  45. 2:17research on this. I've asked around and
  46. 2:19I've tried it myself. And in my
  47. 2:21experience, Minimax is the better
  48. 2:24solution to use here compared to
  49. 2:25something like Seidance. So, here I have
  50. 2:28another example.
  51. 2:42>> I am the AI lord.
  52. 2:59I am the AI lord.
  53. 3:00>> Here, the first generation was Seidance
  54. 3:022.5 and the second one was Minimax. I
  55. 3:05liked the result of Minimax a lot more
  56. 3:07and it's also a lot cheaper. The
  57. 3:10difference here in price is 180 credits
  58. 3:12for Seidance 2.5, while it has only 30
  59. 3:15credits for Minimax H3. And if you use
  60. 3:18the max version, it's not even that much
  61. 3:20more expensive. It's 37.5. And the other
  62. 3:23thing is, Minimax is pretty
  63. 3:25unrestricted, while Seidance 2.5 has the
  64. 3:28tendency to not allow a lot of different
  65. 3:30generations. Like, for example, if you
  66. 3:32want to make copyrighted stuff, you
  67. 3:34can't really do that with Seidance. But,
  68. 3:36with Minimax on the other hand, I made
  69. 3:38this.
  70. 3:40>> Rick, I I told you we can't use
  71. 3:42copyrighted visuals.
  72. 3:43>> Yes, we can, Morty.
  73. 3:44>> No, we can't.
  74. 3:45>> Yes, we can, you stupid human.
  75. 3:47>> We definitely can, guys.
  76. 3:49>> No, we can't.
  77. 3:54>> He's too young. Now, if you're just
  78. 3:56getting started with AI, it might be
  79. 3:57super overwhelming trying to know
  80. 3:59everything about which model you need to
  81. 4:01use. And that's why I made this website
  82. 4:04for you, where you can upload your own
  83. 4:07prompt to it, and it will tell you which
  84. 4:09model you should use. So, call this the
  85. 4:12AI video model selector. How it works is
  86. 4:15you simply add in your prompt. This is
  87. 4:17powered by Jeff, so it does the hard
  88. 4:19work for us. It's run by me, it's super
  89. 4:21cheap. I will just give this away for
  90. 4:22free. You put in your prompt, then you
  91. 4:25add in the reference that you have, and
  92. 4:27then you tell it the duration that the
  93. 4:28video will be. As you can see, if you're
  94. 4:30trying to make longer videos than, for
  95. 4:33example, 15 seconds, you only have a
  96. 4:35choice of these three models. If you go
  97. 4:37longer than that, you only have the
  98. 4:38choice to do something like SeaDance or
  99. 4:41One 3.0. So, let's do 10 seconds as this
  100. 4:43first simple test. What it will do is it
  101. 4:46will evaluate your prompt, and it will
  102. 4:48look through the database of the things
  103. 4:51that I've uploaded there, and then it
  104. 4:52will tell you which model will get you
  105. 4:54the best result. And then it's up to you
  106. 4:56to make the judgment call if it's worth
  107. 4:58it. For example, here it's only giving
  108. 5:0033% to do it with SeaDance 2.5. And it
  109. 5:04will also give you the expected credit
  110. 5:06cost. But for 35 credits cheaper, which
  111. 5:09is 45 credits in total, you can also use
  112. 5:11SeaDance 2.0. And [music] the budget
  113. 5:13pick would be One 3.0. So, it will give
  114. 5:15you those percentages, and if you are
  115. 5:17And depending on what budget you have,
  116. 5:19you can make your choice of that. And
  117. 5:21once you like what you see, you click on
  118. 5:22generate, and then it will bring you
  119. 5:24over to Higgsfield, where you can
  120. 5:25generate your video.
  121. 5:33Okay, now let's try a more complex scene
  122. 5:35to show you how this really works. So,
  123. 5:37now I added in a longer prompt with a
  124. 5:40lot more words, and it's kind of like a
  125. 5:42sitcom type of prompt. I also have a
  126. 5:45couple different references. And then
  127. 5:47I'm going to select the length of how
  128. 5:49long I want this video to be, which is
  129. 5:5115 seconds. For the resolution, I want
  130. 5:53to do 1080p. And now we're going to
  131. 5:55evaluate, and then it tells us that this
  132. 5:59might be best to do with Cidence 2.5.
  133. 6:01Again, it's the most expensive model,
  134. 6:03but if you want the absolute best
  135. 6:04result, you want to be using this. If
  136. 6:06you want to do 4K, it will automatically
  137. 6:09drop that model because 4K is
  138. 6:11unavailable, but you can get 1080p. So,
  139. 6:14based on this recommendation, let's
  140. 6:15actually generate this scene using
  141. 6:17Cidence 2.5. And now it has added in the
  142. 6:20references, it has added in the prompt,
  143. 6:23and then we can hit generate.
  144. 6:26>> I finally found a friend.
  145. 6:27>> Please, everyone knows you don't have
  146. 6:29friends.
  147. 6:30>> Oh, yeah? You sure?
  148. 6:32>> Yeah, I'm sure.
  149. 6:33>> So, look at this.
  150. 6:36Hey, I'm Bruiser, Dan's new friend.
  151. 6:40>> So, yeah. Looking at this result, having
  152. 6:42multiple different camera angles, three
  153. 6:44different characters talking. It's not
  154. 6:46too complicated in terms of camera
  155. 6:47movement, but I would say Cidence 2.5
  156. 6:50really nailed this. And that's how I'm
  157. 6:52hoping to save credits for you. But
  158. 6:54let's say we do something completely
  159. 6:56different. So, let me remove all of
  160. 6:57these references, and then we add in a
  161. 6:59different prompt. This time I'm going to
  162. 7:01do a motion prompt. This is going to be
  163. 7:03for a 2D kinetic typography and motion
  164. 7:07graphics, 50 seconds. So, I've added in
  165. 7:09my prompt. I have added in the
  166. 7:11resolution, all of that. We're going to
  167. 7:13evaluate it, and now it's telling us
  168. 7:15that Minimax H3 might be the best option
  169. 7:18to use here because it is super good at
  170. 7:20making these motion graphics. And if we
  171. 7:22generate this, we get a result like
  172. 7:24this.
  173. 7:28>> [music]
  174. 7:35[music]
  175. 7:40>> So, all we did, and by we I mean me and
  176. 7:42team, is fed the data that we had when
  177. 7:45using these models, and based on that,
  178. 7:47we built this tool for you, so you can
  179. 7:49hopefully save yourself some credits.
  180. 7:51Again, this tool is free to use. I will
  181. 7:53leave the link in the description down
  182. 7:54below. The second method to saving
  183. 7:56yourself credits is using draft mode.
  184. 7:59This is only available for students 2.5,
  185. 8:01and ever since this launched, I've saved
  186. 8:03a lot of credits since then. Because, as
  187. 8:05you can see, the 1080p version is quite
  188. 8:08expensive. But as soon as we switch over
  189. 8:10to the draft mode, we get the video for
  190. 8:1245 credits. The downside, it's only
  191. 8:15480p. You might wonder, like, why would
  192. 8:18you do that? So, for example, I used to
  193. 8:21generate a lot of 1080p videos, and I
  194. 8:24regenerated them until I got it right.
  195. 8:27Each generation cost me like over $10
  196. 8:30with students 2.5 1080p. Now, with draft
  197. 8:33mode, this cost has significantly gone
  198. 8:35down, and then once I found the perfect
  199. 8:38generation, all I had to do is just
  200. 8:40upscale it. Because draft mode is 45
  201. 8:43credits, and then it will cost you an
  202. 8:44extra 180 credits to upscale it. So, let
  203. 8:47me give you an example of what I mean by
  204. 8:49this. So, my goal is to kind of like
  205. 8:50create a Star Wars scene. I have me
  206. 8:53myself as a character sheet in there,
  207. 8:55and I have Bruiser Robot as a character
  208. 8:58sheet in there, and then I've added in
  209. 8:59my prompt. I'm not sure like if this
  210. 9:02will work, because it's quite a complex
  211. 9:04scene. What I want to see is I want to
  212. 9:05have a battle in space. I want to have
  213. 9:07some interaction in terms of talking
  214. 9:09with me and the other character, and I
  215. 9:11want everything to like flow nicely. So,
  216. 9:14instead of me just going straight into
  217. 9:16the 1080p mode, which will cost me a lot
  218. 9:18of credits, I first generated this
  219. 9:20draft. Now, that got me this first
  220. 9:23generation.
  221. 9:24>> [groaning]
  222. 9:27>> Wow, that was really close.
  223. 9:38You really need to learn how to speak.
  224. 9:41>> I've always known how to speak. It was
  225. 9:42just to piss you off.
  226. 9:43>> So, to me this doesn't really make sense
  227. 9:45because first we have the robot in the
  228. 9:47starfighter like this
  229. 9:49and then it's flying like outside of it.
  230. 9:52So, I won't be able to use this
  231. 9:54generation in my video. So, I have to
  232. 9:56regenerate it. But what I do now is I
  233. 9:58can just go back and I can just tweak
  234. 10:00this prompt. So, I look through my
  235. 10:02prompt. I find the parts that I want to
  236. 10:04tweak. I rewrite it and then I hit
  237. 10:06generate again again in draft mode. And
  238. 10:10that got me this generation.
  239. 10:15>> Wow.
  240. 10:16That was really close.
  241. 10:27You really need to learn how to speak.
  242. 10:29>> I've always known how to speak. It was
  243. 10:30just to piss you off.
  244. 10:32>> So, that is already a lot better. But
  245. 10:34now that I'm watching this, I am
  246. 10:35thinking like why don't I have the robot
  247. 10:38like sit outside of the spaceship
  248. 10:41instead of like in the cockpit with me.
  249. 10:43I go back into my prompt. I add in these
  250. 10:46changes and then I generate it again
  251. 10:49with the draft mode. And that generated
  252. 10:51me this video.
  253. 10:55>> Wow. That was really close.
  254. 11:07>> You really need to learn how to speak.
  255. 11:09>> I've always known how to speak. It was
  256. 11:11just to piss you off.
  257. 11:12>> And that's exactly how I want the video
  258. 11:14to be. Like this to me is great. Like I
  259. 11:16can use this. And then now that I have
  260. 11:19the final version, I had to pay one time
  261. 11:21a 240 credit fee to get it in the 1080p
  262. 11:25version. So, in total that would be 420
  263. 11:27credits. So, if you put them side by
  264. 11:29side, you would have saved yourself 300
  265. 11:32credits using the draft mode. But, if we
  266. 11:35compare this to how I used to do this
  267. 11:37the old way, is I used to have to
  268. 11:40generate this video three different
  269. 11:42times for 240 credits each, which will
  270. 11:45bring out the total price to 720 credits
  271. 11:48to now me generating this three times
  272. 11:50using the draft mode, which is 60
  273. 11:53credits per generation. The third method
  274. 11:55to saving yourself money is using the
  275. 11:58API. And in particularly, I'm talking
  276. 12:00about the Hixfeld API as they have
  277. 12:02launched this. Because if you're unsure
  278. 12:03if AI is something for you, or you find
  279. 12:05it expensive, and you don't have to lock
  280. 12:07yourself into a annual or a monthly
  281. 12:10membership, you can just pay for what
  282. 12:12you use. If you're unsure of what to
  283. 12:14use, then let me explain it to you.
  284. 12:16First, if you are using AI casually, and
  285. 12:20you're not spending all of the credits
  286. 12:21that you have in your monthly
  287. 12:23subscription, then get the API. You only
  288. 12:26pay for what you use. If every time you
  289. 12:28use AI and your or subscription is
  290. 12:31finished, you might be better off
  291. 12:33upgrading your subscription or seeing
  292. 12:35like what you can get with the API. But,
  293. 12:38if you are a heavy user, then I would
  294. 12:39still recommend using a subscription.
  295. 12:42I've actually made this graph for you,
  296. 12:43which exactly explains for which
  297. 12:45scenario it is best to either use the
  298. 12:47API or get a subscription and what
  299. 12:50subscription you should get. If you are
  300. 12:52making just one generation, API is
  301. 12:54obviously cheaper because you're wasting
  302. 12:57$49 on the plus plan and $310 if you're
  303. 13:00getting like a maxed out ultra plan if
  304. 13:02you don't use these credits. Now, if you
  305. 13:04make five generations, then the plus
  306. 13:06plan on an annual plan is cheaper. But,
  307. 13:09if you want a monthly plan, then it's
  308. 13:11about the same as the API. If you want
  309. 13:13to make more than five generations, so
  310. 13:15let's say 10 for example, you cannot
  311. 13:17even get that in the plus plan and we're
  312. 13:19talking about students 2.5 1080p
  313. 13:2215-second generation videos. So, these
  314. 13:24are long videos. Keep that in mind. You
  315. 13:26can also compare this to the Cidens 2.0
  316. 13:29and Kling 3.0. Over here, as you can
  317. 13:31see, most of the times the API is
  318. 13:33cheaper to use if you're using these
  319. 13:35models. But when you're using Cidens
  320. 13:362.5, which is the model that most people
  321. 13:38want to be using, then you have to
  322. 13:40decide for yourself which one makes more
  323. 13:41sense. For example, if you're a heavy
  324. 13:43user, you might benefit more from having
  325. 13:46an ultra plan at $310, as it's like
  326. 13:48twice as cheap as using the API here.
  327. 13:51Keep in mind, this is as of making this
  328. 13:53video, these prices might change. So,
  329. 13:56what is actually the benefit of the API
  330. 13:57and how do you start using this? Well,
  331. 14:00let me explain. What you need to do is
  332. 14:01you need to go over to the Higgsfield
  333. 14:03dashboard, then you click on API. This
  334. 14:05will open up a new tab, and here you can
  335. 14:07explore the API. You can then go to the
  336. 14:10quick start page right here, and you can
  337. 14:12copy this prompt and send this to your
  338. 14:14agent. If you copy this, you put it
  339. 14:16inside of your ChatGPT or your Claude or
  340. 14:19whatever you're using, you paste it in
  341. 14:21there, and then you hit send, and now it
  342. 14:23will start building your own AI video
  343. 14:26model tool for you using the Higgsfield
  344. 14:29API. Now, for me, this took like 10
  345. 14:31minutes, and it made basically our own
  346. 14:33variation of Higgsfield, but it is using
  347. 14:36the API. Like, it's exactly the same as
  348. 14:38Higgsfield. You have all of the models
  349. 14:39that you want to use, you have all of
  350. 14:41the image models that you can try out
  351. 14:42here, and you can add in your start
  352. 14:44frame, end frame, tweak these settings,
  353. 14:46everything you want. All you have to do
  354. 14:48here is you have to connect your API
  355. 14:50key, and then you can get this to work.
  356. 14:52Keep in mind that if you go to your
  357. 14:54Claude, Codex, or even Gemini chat where
  358. 14:57you start building this, that you can
  359. 14:59customize everything out of this. For
  360. 15:02example, I made a better type of
  361. 15:04Higgsfield, I would say, where I use
  362. 15:06this API access, and I customized it
  363. 15:09using all of my different workflows that
  364. 15:12me and my team are using internally. And
  365. 15:14I basically put this into the LLM, and
  366. 15:16then I've made variations of that, and
  367. 15:18that is this result. So, now there are a
  368. 15:21cool things which I can do. For example,
  369. 15:23if I added my prompt right here, I have
  370. 15:25this prompt optimizer and this prompt
  371. 15:27optimizer is made based on my own MD
  372. 15:31files. So I have had that in there. So
  373. 15:33it's using my own optimizer skill. It's
  374. 15:35not using anyone's else, completely
  375. 15:37tailored to my liking. So let's generate
  376. 15:39an example here. I'm just going to do
  377. 15:40the low quality mode. So then we hit
  378. 15:43generate and now it's using the API to
  379. 15:45make this. So this is the video that it
  380. 15:47made.
  381. 15:58And as you can see, it has used my MD
  382. 16:01file. I don't have to do this in my LM
  383. 16:03anymore because it's has done that for
  384. 16:06me through this new tool that I built.
  385. 16:08And you can access this right now. So
  386. 16:09the all you have to do is once you
  387. 16:11access this site, you just need to add
  388. 16:13in your own API key. You need to add an
  389. 16:15open AI API key. You need this so it can
  390. 16:18do the prompting for you and then you
  391. 16:19get access to this. Another cool feature
  392. 16:21that I built into this is the style
  393. 16:23feature and here you can copy the style
  394. 16:26of any type of reference image. So for
  395. 16:28example, I really like the look of this
  396. 16:31image and if I now drop this into the
  397. 16:33style right here, it will apply that
  398. 16:34style to my videos. So now I'm going to
  399. 16:36describe my scene. So I'm going to add
  400. 16:38in a simple prompt of a cyberpunk style
  401. 16:41samurai walking into a flower field and
  402. 16:43now we're going to hit generate and now
  403. 16:45it's using the style of this image and
  404. 16:47it's optimizing my prompt to generate
  405. 16:49something like this.
  406. 17:00So that is pretty cool. I've also got
  407. 17:02one more feature which I'm not going to
  408. 17:04share with you, but like this is a tool
  409. 17:06we use internally and that's why I like
  410. 17:09API so much because I can tweak my own
  411. 17:11MD files, my own skill files and I can
  412. 17:14just share it internally with my team,
  413. 17:15build my own tools, host it on my own
  414. 17:17servers, and pay only for what I use.
  415. 17:20So, if you apply all of these different
  416. 17:21methods, you're able to save yourself
  417. 17:24some credits. And hopefully, this was
  418. 17:26useful to you. If it was, please let me
  419. 17:28know in the comments down below what you
  420. 17:29want to see next. And if you want to
  421. 17:31learn how to make super realistic AI
  422. 17:33videos, then click the video that's on
  423. 17:35the screen right now, and hopefully, you
  424. 17:37will learn a lot more.

About this transcript

This page contains the full transcript of Stop Wasting Credits on AI Video Generators Like Seedance 2.5 by Dan Kieft, generated from the public captions YouTube serves with the video. The transcript has 3,087 words across 424 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.