YouTube2Text

Gemini Live Avatars — Transcript

by Sam Witteveen · 2,012 words · 289 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Welcome. I'm Jennifer. I'm ready to dive
  2. 0:02into the latest F1 action with you.
  3. 0:05>> Hey Jennifer, who won the latest Grand
  4. 0:06Prix?
  5. 0:07>> George Russell won the latest Grand Prix
  6. 0:09in Azerbaijan. He had a fantastic race,
  7. 0:12holding off Max Verstappen to take the
  8. 0:14win.
  9. 0:14>> So, where did Kimi Antonelli come?
  10. 0:17>> Kimi Antonelli finished in fifth place
  11. 0:19after a really impressive recovery drive
  12. 0:21from 16th on the grid.
  13. 0:23>> Okay, so that was Google's new live
  14. 0:26avatars with Gemini 3.8 live. So, I've
  15. 0:29talked about the live models in the
  16. 0:30past. These are the bidirectional models
  17. 0:32that basically have multimedia going in,
  18. 0:35being audio and video. And up until
  19. 0:37recently, it could just produce text or
  20. 0:39audio back. Well, Google's now updated
  21. 0:41that so it can actually have a live
  22. 0:43rendering face that kind of lip-syncs
  23. 0:46what the actual audio is. So, up until
  24. 0:48recently, if you wanted to do any sort
  25. 0:50of AI agent with a face that could
  26. 0:52actually hold a conversation, you would
  27. 0:54generally sort of stitching together
  28. 0:56four different things. You had some kind
  29. 0:57of speech to text, you had an LLM, you
  30. 1:00had a TTS model creating the voice back,
  31. 1:03and then there was some sort of avatar
  32. 1:04service on the end doing lip-sync. And
  33. 1:06the challenge has always been with that
  34. 1:08kind of thing is that every one of those
  35. 1:10points adds a lot of latency. And by the
  36. 1:12time the mouth starts moving, the
  37. 1:14conversation has often already started
  38. 1:16to fall apart. Now, I won't say that
  39. 1:18Google's got this perfect, but what
  40. 1:19they've done is pretty impressive.
  41. 1:22They've made their Gemini 3.8 live
  42. 1:24avatar system generally available. And
  43. 1:26there are actually a few caveats to
  44. 1:27that, which I'll mention later on in the
  45. 1:29video. But this whole thing basically
  46. 1:31collapses down those four models that
  47. 1:33you had to take care of yourself into a
  48. 1:35single stream. It can listen, it can see
  49. 1:38what's on your screen, it can call tools
  50. 1:40in the background. As you can see in the
  51. 1:42example I just showed, it was using
  52. 1:44Gemini search to basically get the
  53. 1:46latest results from Saturday's Grand
  54. 1:48Prix. And the cool thing here is it can
  55. 1:50do all of this in 97 languages. So, I've
  56. 1:53been playing with this for a while, and
  57. 1:55I built some demos that I'm going to go
  58. 1:56through in a second and show you. But
  59. 1:58let's put this into a little bit of
  60. 1:59context. So, while we're still yet to
  61. 2:01see a Gemini Pro model come out now for
  62. 2:04I think it's about 7 months, this has
  63. 2:06still been a pretty busy week for Google
  64. 2:08on the audio side. So, in my last video
  65. 2:10I talked about the whole new TTS system
  66. 2:12that they launched with the voice
  67. 2:14cloning, with those sorts of things. And
  68. 2:16actually before that, they updated
  69. 2:17Gemini 3.8 live and live extended
  70. 2:20thinking. And now they've released live
  71. 2:22avatar. And like I said in the last
  72. 2:24video, whatever you think about Google
  73. 2:26is in the LLM race at the moment, the
  74. 2:28multimodal and audio stuff has been
  75. 2:31really strong. They've had Omni, live
  76. 2:33translate, transcribe, the TTS models,
  77. 2:36and now this live avatar. So, I guess
  78. 2:38quickly what does generally available
  79. 2:40actually mean here? Basically, it's
  80. 2:42ready for production on Google Cloud.
  81. 2:45But the sad thing here is that not
  82. 2:46everything is generally available. You
  83. 2:48get a library of pre-built avatars. If
  84. 2:50you want a custom avatar, one made from
  85. 2:53your own reference photo, your own sort
  86. 2:55of voice, that kind of thing, that's
  87. 2:56behind an allow list and a verification
  88. 2:58process. And even things like the live
  89. 3:01extended thinking version of live is
  90. 3:02still in private preview. So, in this
  91. 3:04video I'm just going to show what you
  92. 3:05could go and do and build right now and
  93. 3:08show that in my demos, etc. If you want
  94. 3:10to see an example of the custom agents,
  95. 3:12they've got a video in there with a demo
  96. 3:14of making a custom avatar for an avatar
  97. 3:17with a really bad Australian accent that
  98. 3:19I think only a bank owned by Crocodile
  99. 3:21Dundee would actually want to use. So,
  100. 3:23with what they've actually made
  101. 3:24generally available in here, you've got
  102. 3:27a bunch of different visual avatars that
  103. 3:29you can pick from. So, here I've made an
  104. 3:30avatar studio, which is basically a
  105. 3:33front end app that's running this and
  106. 3:36streaming the live avatars through the
  107. 3:37back end in here. So, here we can
  108. 3:39basically just pick any of the avatars
  109. 3:42that we want to. I can then basically
  110. 3:44choose a voice that I've got and they've
  111. 3:46got a set of voices that we can actually
  112. 3:48pick. We can pick the sort of persona
  113. 3:51preset that we want. Let's go for a time
  114. 3:53traveling tour from 2126, and then we've
  115. 3:56got the system instructions. Now,
  116. 3:57obviously, the system instructions, you
  117. 3:59can fully customize yourself. You can
  118. 4:01get all of that set up. And then you can
  119. 4:03add in tools if you want. So, here I've
  120. 4:05basically got Google Ground Searching
  121. 4:07for this. Now, since we're talking to
  122. 4:09someone from the future, that actually
  123. 4:10might not be good. I will leave that off
  124. 4:12for now. And then basically, what I can
  125. 4:14do is I can just start it. And then I've
  126. 4:17got the choices. Once I start
  127. 4:19>> Wow, it is so incredible to finally see
  128. 4:22the quaint 21st century in person.
  129. 4:25>> Okay, and you can see there that now I'm
  130. 4:28talking to Vera. Vera, how are you doing
  131. 4:30today?
  132. 4:31>> I am doing absolutely wonderful. Thank
  133. 4:33you. Exploring this era is just pure
  134. 4:36magic. These are actually a perfect
  135. 4:39vintage replica of 2026 fashion. So, you
  136. 4:42could say I dressed the part to blend in
  137. 4:44perfectly.
  138. 4:45>> Okay, so Vera, what I want to show is
  139. 4:46that we can actually type to you as
  140. 4:48well. So, I'm going to ask you a
  141. 4:50question in a second, and I only want
  142. 4:52you to answer the question when you see
  143. 4:53the typed question.
  144. 4:55>> Understood. I am ready and waiting for
  145. 4:57your typed question. I'm doing
  146. 4:59absolutely fantastic. Just so clear the
  147. 5:02incredible energy of 2026.
  148. 5:06>> You can see there that basically, she
  149. 5:07was able to take a type or voice, and
  150. 5:10she's still listening to me, aren't you,
  151. 5:11Vera?
  152. 5:12>> Exactly. I'm processing everything
  153. 5:14perfectly, no matter how it reaches me.
  154. 5:16>> Okay, I'm going to stop there. So, you
  155. 5:19can get a sense from that this is
  156. 5:20actually pretty responsive, right? Now,
  157. 5:23there is a delay, and obviously, we're
  158. 5:25going to get the transcript back ever so
  159. 5:27slightly before the actual voice comes
  160. 5:30back, which kind of makes sense. But you
  161. 5:32can see for something where now we can
  162. 5:33go through and pick any of the different
  163. 5:35avatars. You can pick photorealistic
  164. 5:38avatars, or you can pick sort of
  165. 5:40cartoonish avatars in here. And so, the
  166. 5:43cool thing here is you can actually make
  167. 5:44a bunch of fun demos with this. So, not
  168. 5:47only can we do all of this in English,
  169. 5:49we can actually come in here and we can
  170. 5:51pick different languages. So, if we want
  171. 5:54to have a cafe and we want to put it in
  172. 5:56a different place, we can actually sort
  173. 5:58of make this a conversational app for
  174. 6:01teaching us to speak a different
  175. 6:03conversation, etc. So, you can see here
  176. 6:05we've basically got something where it's
  177. 6:08going to teach me Let's try Japanese.
  178. 6:11Let's see how that goes. And what I'm
  179. 6:13going to do is try and have a
  180. 6:14conversation with her. Hi there. Can you
  181. 6:16tell me how to order a coffee in
  182. 6:18Japanese?
  183. 6:25Kohi is coffee?
  184. 6:31Okay, so how do I say I would like one
  185. 6:33coffee?
  186. 6:38How do I pronounce it? Can you give me
  187. 6:40the corrections for how I should
  188. 6:42pronounce it?
  189. 6:50Kohi o kudasi.
  190. 6:56Okay, so enough of me embarrassing
  191. 6:58myself Nothing going to speak Japanese,
  192. 7:00but you can see here that we can
  193. 7:02basically set this up to teach us
  194. 7:05vocabulary, to do a whole bunch of
  195. 7:06different things. And we could change it
  196. 7:08so that we start with something that's
  197. 7:10really beginner and something that then
  198. 7:12gets right through to something much
  199. 7:14more advanced in here. Another thing
  200. 7:16that you can actually do for fun is that
  201. 7:18you can actually have multiple
  202. 7:20characters and have them communicate to
  203. 7:22each other. So, here we're going to
  204. 7:24start a debate over whether pineapple
  205. 7:27belongs on pizza or not. And you can see
  206. 7:30Moderator, esteemed guests, culinary
  207. 7:33purists often mistake rigidity for
  208. 7:35tradition, but love and gastronomy
  209. 7:39thrives on daring unions. To deny the
  210. 7:42pineapple its rightful throne upon the
  211. 7:44pizza is to deny the very spirit of
  212. 7:46romance, which seeks the unexpected
  213. 7:49harmony of sweet and savory. Imagine the
  214. 7:52sun-drenched tropics marrying the rustic
  215. 7:54warmth of baked dough. It is not just a
  216. 7:57topping, Carmen. It is poetry on a
  217. 7:59crust.
  218. 8:00>> Quickly, it is then. Leo, while I
  219. 8:03appreciate your poetic vision, culinary
  220. 8:06harmony requires more than just sweet
  221. 8:09and savory colliding. From a practical
  222. 8:11standpoint, the excessive moisture in
  223. 8:14pineapple compromises the structural
  224. 8:16integrity of the crust, turning a crisp
  225. 8:18base into a soggy mess.
  226. 8:21>> You are right, moderator. There is
  227. 8:23wisdom in her words that I must respect.
  228. 8:26Yet, Carmen, true passion often
  229. 8:28withstands a few
  230. 8:30>> Okay, I'm going to stop it there, but
  231. 8:31you get a sense there of that these two
  232. 8:33can actually sort of communicate to each
  233. 8:35other. So, you can have some sort of fun
  234. 8:37with that. Probably I'd go back and
  235. 8:39actually change the system prompt to
  236. 8:40make them a little bit more concise for
  237. 8:42my demo. And it doesn't stop there. You
  238. 8:44can also do things like where because
  239. 8:46you can also activate the camera, you
  240. 8:49can then hold up things and have a
  241. 8:50conversation about them with the avatar.
  242. 8:53Now, if you want to see a video of me
  243. 8:54actually explaining the code and going
  244. 8:56through it a little bit and stuff like
  245. 8:58that, I showed you how to get started,
  246. 8:59etc. I'm going to put the link for that
  247. 9:01in the description. That's going to be
  248. 9:02on my second channel. You can go through
  249. 9:04and get this all started yourself. Okay,
  250. 9:07so just to finish off, I would say that
  251. 9:08while this doesn't have the flexibility
  252. 9:10that perhaps we would like at the
  253. 9:12moment, again, this is Google having to
  254. 9:15be conservative with how they do things.
  255. 9:17They don't want 500 people uploading
  256. 9:20pictures of Trump or making a Trump
  257. 9:21avatar or something like that. My guess
  258. 9:23is there's going to be quite a number of
  259. 9:24safety features around that kind of
  260. 9:26avatar system. You can imagine also that
  261. 9:28Google's going to be pretty strict about
  262. 9:29the different use cases for this. If you
  263. 9:31want to try it out, this is in Google
  264. 9:33Cloud now. You can actually just come in
  265. 9:35and look at the Gemini live avatar. You
  266. 9:38can pick from the avatars that they've
  267. 9:39actually got set up in here. So, you can
  268. 9:41certainly get a demo of this up for a
  269. 9:43business or to show people how to use
  270. 9:45it, etc. Now, the pricing for this is
  271. 9:47going to be pretty pricey coming out. I
  272. 9:49think the simple fact that this just
  273. 9:51uses so much compute means that we're
  274. 9:54probably not going to see this get super
  275. 9:55cheap in the really near term. But, it's
  276. 9:57definitely going to be interesting to
  277. 9:58see where this goes, and I'd love to
  278. 10:00hear from people where you see the best
  279. 10:02use cases of this are actually going to
  280. 10:03be. Let me know in the comments what you
  281. 10:05think about those. And hopefully it
  282. 10:07won't be too long before we see an open
  283. 10:08version of this, perhaps from Nvidia or
  284. 10:11from one of the other companies that are
  285. 10:12actually releasing open models out there
  286. 10:14as well. Anyway, as always, if you found
  287. 10:16the video useful, please click like and
  288. 10:18subscribe, and I will talk to you in the
  289. 10:19next video. Bye for now.

About this transcript

This page contains the full transcript of Gemini Live Avatars by Sam Witteveen, generated from the public captions YouTube serves with the video. The transcript has 2,012 words across 289 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.