WEBVTT 00:00:00.000 --> 00:00:02.600 Imagine the promise. You just click one single 00:00:02.600 --> 00:00:06.500 button and a completely formatted professional 00:00:06.500 --> 00:00:10.099 slide deck just materializes in two minutes. 00:00:10.220 --> 00:00:12.380 Oh, man. It sounds like the absolute ultimate 00:00:12.380 --> 00:00:14.820 productivity dream. It really does. But then, 00:00:14.839 --> 00:00:16.760 well, then you look closely at the result. Yeah. 00:00:16.859 --> 00:00:20.199 And it's just this plain, generic, totally uneditable 00:00:20.199 --> 00:00:22.760 mess. Oh, totally. It's the classic trap of early 00:00:22.760 --> 00:00:25.359 automation. We've basically been conditioned 00:00:25.359 --> 00:00:28.320 by the tech industry to expect instant magic. 00:00:28.539 --> 00:00:32.369 Right. and we completely bypass the actual architectural 00:00:32.369 --> 00:00:36.090 work required to make that magic, you know. actually 00:00:36.090 --> 00:00:38.770 useful. Welcome to this deep dive. We're really 00:00:38.770 --> 00:00:41.490 glad you're joining us today. I've been thinking 00:00:41.490 --> 00:00:44.210 a lot lately about the friction between our creative 00:00:44.210 --> 00:00:46.530 intentions and the actual output we get from 00:00:46.530 --> 00:00:48.829 AI tools. It's a huge pain point for so many 00:00:48.829 --> 00:00:51.770 people right now. It is. So our mission today 00:00:51.770 --> 00:00:53.850 is to kind of step outside of that one -click 00:00:53.850 --> 00:00:57.109 illusion. We are going to break down a professional 00:00:57.109 --> 00:01:00.530 three -step presentation workflow inside Notebook 00:01:00.530 --> 00:01:02.789 LM. Which is such a game changer once you understand 00:01:02.789 --> 00:01:06.040 it. Exactly. So first, we'll look at how to build 00:01:06.040 --> 00:01:09.140 a walled garden of trusted sources to completely 00:01:09.140 --> 00:01:12.500 eliminate data hallucinations. Essential. Then 00:01:12.500 --> 00:01:16.299 we'll explore how to craft slide by slide prompts 00:01:16.299 --> 00:01:19.719 for really rigorous brand control. And finally, 00:01:20.219 --> 00:01:22.840 we're going to solve one of the most maddening 00:01:22.840 --> 00:01:24.879 bottlenecks in this entire process. Oh, I know 00:01:24.879 --> 00:01:27.140 what you're going to say. Yeah. The flat image 00:01:27.140 --> 00:01:31.159 export limitation. Yes. It is a critical shift 00:01:31.159 --> 00:01:33.280 in mindset. You really have to move from being 00:01:33.280 --> 00:01:36.159 just a passive consumer of a software feature 00:01:36.159 --> 00:01:39.260 to acting as an active art director. I like that, 00:01:39.319 --> 00:01:41.659 an active art director. Exactly. Because the 00:01:41.659 --> 00:01:44.719 tool is wildly powerful, but only if you constrain 00:01:44.719 --> 00:01:48.060 it first. It's all about setup. generation, and 00:01:48.060 --> 00:01:50.159 targeted revision. You know, whenever I sit down 00:01:50.159 --> 00:01:52.140 to build a presentation, the immediate temptation 00:01:52.140 --> 00:01:54.140 is to just jump straight to the design phase. 00:01:54.299 --> 00:01:56.159 Oh, we all do it. Right. We want to pick the 00:01:56.159 --> 00:01:57.959 fonts, we want to arrange the colors, and just 00:01:57.959 --> 00:02:00.159 make it look pretty. Yeah. But the foundation 00:02:00.159 --> 00:02:02.680 of any meaningful deck isn't the aesthetic layer. 00:02:02.799 --> 00:02:05.480 No, not at all. It's the data layer. If we skip 00:02:05.480 --> 00:02:07.299 straight to the generate button, we are basically 00:02:07.299 --> 00:02:10.099 just guaranteeing mediocrity. Because the prettiest 00:02:10.099 --> 00:02:13.539 typography in the world cannot save factually 00:02:13.539 --> 00:02:16.750 bankrupt content. Right. NoBookLM is fundamentally 00:02:16.750 --> 00:02:19.629 a source -based tool, and that makes its underlying 00:02:19.629 --> 00:02:22.030 architecture very different from a generalized 00:02:22.030 --> 00:02:25.669 model, like chat GPT or Gemini. Let's define 00:02:25.669 --> 00:02:28.210 that clearly for a second. A source -based AI 00:02:28.210 --> 00:02:30.990 strictly analyzes the specific files you upload. 00:02:31.289 --> 00:02:34.229 It doesn't scour the open internet unless explicitly 00:02:34.229 --> 00:02:36.870 told to. It's kind of like giving a student an 00:02:36.870 --> 00:02:39.310 open book test, but you only allow them to bring 00:02:39.310 --> 00:02:42.330 three very specific textbooks into the exam room. 00:02:42.509 --> 00:02:45.270 That is a perfect analogy. That constraint is 00:02:45.270 --> 00:02:48.069 exactly what gives it power. But, you know, human 00:02:48.069 --> 00:02:50.150 nature kicks in, right? Always. We're busy. We're 00:02:50.150 --> 00:02:52.030 on a deadline. We think, oh, I'll just use the 00:02:52.030 --> 00:02:54.490 auto web search feature and let the AI pull the 00:02:54.490 --> 00:02:56.500 background context for me. I mean, I have to 00:02:56.500 --> 00:02:58.919 be honest here. I still wrestle with prompt drisht 00:02:58.919 --> 00:03:01.219 myself when I don't give the AI rigid boundaries. 00:03:01.400 --> 00:03:03.680 Really? Oh, yeah. I get lazy. I'll just drop 00:03:03.680 --> 00:03:06.500 in one vague link, skip the deep source curation 00:03:06.500 --> 00:03:09.099 entirely, and just hope the model magically intuits 00:03:09.099 --> 00:03:12.099 my entire business strategy. We all do it. It's 00:03:12.099 --> 00:03:15.099 so tempting. But relying on live web searches 00:03:15.099 --> 00:03:17.879 for your core data is essentially just inviting 00:03:17.879 --> 00:03:20.460 chaos into your workflow. It's a mess. It's like 00:03:20.460 --> 00:03:22.849 a dog catching a random scent on a walk. the 00:03:22.849 --> 00:03:25.810 AI just wanders off. You're pulling in SEO spam, 00:03:25.949 --> 00:03:28.009 you're grabbing irrelevant sidebars, completely 00:03:28.009 --> 00:03:30.490 out of context quotes. It completely dilutes 00:03:30.490 --> 00:03:32.750 the core message of your presentation because 00:03:32.750 --> 00:03:35.169 the AI is trying to synthesize too many conflicting 00:03:35.169 --> 00:03:38.090 realities. So the solution is heavily curating 00:03:38.090 --> 00:03:40.430 that left -hand source panel. We can upload up 00:03:40.430 --> 00:03:43.849 to five trusted formats, local files, live links. 00:03:44.539 --> 00:03:47.379 PDFs, Google Drive documents, or even YouTube 00:03:47.379 --> 00:03:49.979 videos. Let's ground this in a real scenario. 00:03:50.620 --> 00:03:53.240 Say you are tasked with building a really complex 00:03:53.240 --> 00:03:55.919 deck on climate change mitigation. Oh, wow. OK, 00:03:56.000 --> 00:03:58.379 a massive topic. If you just hit generate. with 00:03:58.379 --> 00:04:00.639 an open web search on climate change, the AI 00:04:00.639 --> 00:04:02.939 is going to drown in contradictory noise and, 00:04:02.939 --> 00:04:04.939 you know, political opinion. Exactly. But if 00:04:04.939 --> 00:04:07.560 you purposefully combine a verified NASA PDF 00:04:07.560 --> 00:04:11.060 on atmospheric data, a really well curated Wikipedia 00:04:11.060 --> 00:04:14.060 summary and a specific YouTube explainer video 00:04:14.060 --> 00:04:16.339 breaking down carbon capture, you are essentially 00:04:16.339 --> 00:04:19.899 curating the entire universe the AI is allowed 00:04:19.899 --> 00:04:22.529 to perceive. Waldgarden. Yes. you're building 00:04:22.529 --> 00:04:25.149 a verifiable reality. And the fascinating part 00:04:25.149 --> 00:04:27.990 is just watching how it synthesizes those specific 00:04:27.990 --> 00:04:32.370 inputs. It cross -references the dense academic 00:04:32.370 --> 00:04:35.689 data tables in that NASA PDF with the highly 00:04:35.689 --> 00:04:38.550 accessible conversational transcript from the 00:04:38.550 --> 00:04:41.730 YouTube video. It creates this factual foundation 00:04:41.730 --> 00:04:44.610 that is completely bulletproof, but still really 00:04:44.610 --> 00:04:47.069 easy for a general audience to digest. And it 00:04:47.069 --> 00:04:48.769 gives you a quick summary in the center panel, 00:04:48.949 --> 00:04:50.949 which sort of acts as a confirmation that it 00:04:50.949 --> 00:04:53.430 has actually ingested the reality you've built. 00:04:53.889 --> 00:04:56.709 Exactly. So stepping back for a second, why exactly 00:04:56.709 --> 00:04:59.250 does pulling from live web searches dilute the 00:04:59.250 --> 00:05:01.730 quality of the slides so much? Basically because 00:05:01.730 --> 00:05:04.029 it introduces completely uncontrolled variables 00:05:04.029 --> 00:05:06.569 and just totally irrelevant context into the 00:05:06.569 --> 00:05:10.019 AI's synthesis engine. OK, so bad sources equal 00:05:10.019 --> 00:05:13.439 mixed results. Verified uploads equal laser -focused 00:05:13.439 --> 00:05:16.399 accuracy. That is the anchor of the whole workflow. 00:05:16.620 --> 00:05:19.360 You have to wall off your reality first. So once 00:05:19.360 --> 00:05:22.079 we've successfully walled off the AI's universe 00:05:22.079 --> 00:05:25.949 with our trusted PDFs and videos, The next logical 00:05:25.949 --> 00:05:28.889 hurdle is visual. Yep. How do we stop it from 00:05:28.889 --> 00:05:31.930 turning that great data into a boring gray wall 00:05:31.930 --> 00:05:34.709 of text? We have to move our attention from the 00:05:34.709 --> 00:05:37.670 source panel over to the prompt box. And this 00:05:37.670 --> 00:05:40.310 transition is where a lot of users just totally 00:05:40.310 --> 00:05:43.389 drop the ball. They do. They assume the AI inherently 00:05:43.389 --> 00:05:46.329 knows how to pace a narrative arc. But before 00:05:46.329 --> 00:05:49.250 you even hover over that generate button, you 00:05:49.250 --> 00:05:50.889 have to interact with the visual description 00:05:50.889 --> 00:05:54.029 box. You absolutely must define your structural 00:05:54.029 --> 00:05:56.829 goal. Let's unpack the formats Notebook LM offers 00:05:56.829 --> 00:05:59.670 first. I notice in the advanced options, there's 00:05:59.670 --> 00:06:02.389 a distinct split between presenter slides and 00:06:02.389 --> 00:06:05.129 a detailed deck. Yes, and it's a vital distinction. 00:06:05.670 --> 00:06:07.449 Presenter slides are lightweight visual layouts 00:06:07.449 --> 00:06:10.370 meant strictly for live public speaking. Right, 00:06:10.370 --> 00:06:12.529 so if I'm speaking on a stage, I want to force 00:06:12.529 --> 00:06:14.810 the AI into presenter mode. Exactly. I don't 00:06:14.810 --> 00:06:17.089 want it generating a massive paragraph that basically 00:06:17.089 --> 00:06:18.829 acts like a teleprompter on the screen behind 00:06:18.829 --> 00:06:21.970 me and a detailed deck. Those are heavy text 00:06:21.970 --> 00:06:24.490 sides designed for people to read alone. So think 00:06:24.490 --> 00:06:27.310 of an internal company memo or a really dense 00:06:27.310 --> 00:06:29.709 research report. That's just formatted as a presentation 00:06:29.709 --> 00:06:32.750 Okay, so you set those guardrails in the advanced 00:06:32.750 --> 00:06:35.509 options menu by clicking the small arrow next 00:06:35.509 --> 00:06:38.389 to the slide deck button You choose the format 00:06:38.389 --> 00:06:41.759 the language and a rough slide length Under 10 00:06:41.759 --> 00:06:44.620 slides is short, 10 to 15 is the default, and 00:06:44.620 --> 00:06:47.740 anything over 15 is long. But the real leverage, 00:06:47.939 --> 00:06:50.980 the actual design control, comes from the description 00:06:50.980 --> 00:06:53.920 box itself. And so many people leave it entirely 00:06:53.920 --> 00:06:56.680 blank. I know. Or they write something incredibly 00:06:56.680 --> 00:06:59.519 weak. A classic weak prompt is just typing something 00:06:59.519 --> 00:07:02.060 like, create a professional presentation about 00:07:02.060 --> 00:07:04.019 digital marketing. Which sounds like a recipe 00:07:04.019 --> 00:07:07.779 for a completely generic, soul -crushing corporate 00:07:07.779 --> 00:07:10.319 snooze fest. So it guarantees it. The AI will 00:07:10.319 --> 00:07:13.160 default to the safest, most statistically average 00:07:13.160 --> 00:07:15.579 interpretation of professional. You get slate 00:07:15.579 --> 00:07:18.560 gray backgrounds, standard bullet points, and 00:07:18.560 --> 00:07:21.040 incredibly dry pacing. So how do we force it 00:07:21.040 --> 00:07:24.009 to act like a high -end design agency? We have 00:07:24.009 --> 00:07:27.850 to be brutally specific. Contrast that weak prompt 00:07:27.850 --> 00:07:31.350 with a strong one. You don't just ask for a deck. 00:07:31.850 --> 00:07:35.230 You specify a 10 -slide deck on 2025 marketing 00:07:35.230 --> 00:07:39.110 trends. You demand a dark navy background, white 00:07:39.110 --> 00:07:42.649 text, vibrant orange accents, and bold sans serif 00:07:42.649 --> 00:07:45.410 fonts. You're stepping into the role of an art 00:07:45.410 --> 00:07:48.300 director. you are assigning a highly specific 00:07:48.300 --> 00:07:51.279 creative brief. And you should dictate the structural 00:07:51.279 --> 00:07:53.899 rhythm, too. How so? Well, instructed to use 00:07:53.899 --> 00:07:56.860 one main headline, exactly two to three short 00:07:56.860 --> 00:07:59.639 bullets, and one simple visual per slide. And 00:07:59.639 --> 00:08:02.199 you define the tone. You tell it, make this practical, 00:08:02.620 --> 00:08:04.660 confident, and written for non -technical business 00:08:04.660 --> 00:08:07.100 owners. Exactly. But if we connect this to the 00:08:07.100 --> 00:08:09.860 absolute ceiling of what the tool can do, the 00:08:09.860 --> 00:08:11.959 ultimate control method isn't just writing a 00:08:11.959 --> 00:08:14.480 global style prompt. Right. It's dictating the 00:08:14.480 --> 00:08:16.740 entire presentation. slide by slide. This is 00:08:16.740 --> 00:08:18.480 where it gets incredibly powerful. It obviously 00:08:18.480 --> 00:08:20.660 requires a lot more upfront labor, but the resulting 00:08:20.660 --> 00:08:23.860 polish is just staggering. It really is. You 00:08:23.860 --> 00:08:26.300 essentially write a script for the AI, pacing 00:08:26.300 --> 00:08:29.160 out the entire narrative, and paste that detailed 00:08:29.160 --> 00:08:31.319 brief directly into the visual description box. 00:08:31.480 --> 00:08:33.080 Let's walk through what that looks like in practice. 00:08:33.100 --> 00:08:35.559 Okay. Imagine we're building a deck on digital 00:08:35.559 --> 00:08:38.399 tools for remote teams. Yeah. You start by establishing 00:08:38.399 --> 00:08:41.529 that global style we talked about. clean, minimal, 00:08:41.870 --> 00:08:44.110 light gray background, flat icon illustrations, 00:08:44.590 --> 00:08:46.649 friendly professional tone. Then you literally 00:08:46.649 --> 00:08:49.830 script the sequence. You type, slide one, title, 00:08:50.470 --> 00:08:54.669 the remote team toolkit, visual, a simple minimalist 00:08:54.669 --> 00:08:56.850 desk illustration. Perfect. Then you move to 00:08:56.850 --> 00:08:59.649 the problem statement, slide two, title, why 00:08:59.649 --> 00:09:03.090 remote teams struggle, visual, an illustration 00:09:03.090 --> 00:09:05.490 of confused expressions on a video call grid. 00:09:05.610 --> 00:09:07.549 And then you get highly specific on the solution, 00:09:07.710 --> 00:09:11.590 slide three, loom. You explicitly tell the AI 00:09:11.590 --> 00:09:13.970 the key takeaway you want on screen, which is 00:09:13.970 --> 00:09:16.610 record a three minute video, and you demand a 00:09:16.610 --> 00:09:19.070 specific icon like a red record button. Whoa. 00:09:20.009 --> 00:09:21.629 Just imagine the scale of this for a second. 00:09:22.009 --> 00:09:24.669 Imagine generating 10 completely distinct, perfectly 00:09:24.669 --> 00:09:26.330 tailored presentation decks for 10 different 00:09:26.330 --> 00:09:28.850 audiences. Ranging from highly technical engineers 00:09:28.850 --> 00:09:31.690 to absolute beginners. Exactly. All pulling from 00:09:31.690 --> 00:09:34.490 the exact same source material just by swapping 00:09:34.490 --> 00:09:36.450 out that master prompt in a matter of minutes. 00:09:36.549 --> 00:09:38.730 Just tweaking the prompt, hitting Generate, and 00:09:38.730 --> 00:09:41.149 letting it run. That completely redefines how 00:09:41.149 --> 00:09:43.389 we think about presentation prep. Oh, entirely. 00:09:43.629 --> 00:09:45.649 Now, the generation process does take two to 00:09:45.649 --> 00:09:47.909 four minutes, and you have to keep the browser 00:09:47.909 --> 00:09:50.509 tab open while it works. Yeah, that's true. But 00:09:50.509 --> 00:09:52.909 because you've mapped the sequence, the story 00:09:52.909 --> 00:09:55.269 stays tightly connected. The visual language 00:09:55.269 --> 00:09:57.549 doesn't just wander off halfway through. Right. 00:09:57.710 --> 00:10:01.610 But wait. Doesn't writing a massive, highly detailed, 00:10:01.830 --> 00:10:05.129 slide -by -slide brief sort of defeat the purpose 00:10:05.129 --> 00:10:08.110 of using a fast AI tool to begin with? Not at 00:10:08.110 --> 00:10:11.629 all. Because it forces the AI into your specific 00:10:11.629 --> 00:10:14.649 story flow, saving hours of back -end revision. 00:10:14.929 --> 00:10:17.250 Right. Putting in the time up front saves hours 00:10:17.250 --> 00:10:20.370 of frustrating formatting later. Exactly. Front 00:10:20.370 --> 00:10:22.269 -load the creative thinking so you can actually 00:10:22.269 --> 00:10:24.809 enjoy the final product. Okay, so you followed 00:10:24.809 --> 00:10:26.960 the blueprint? You've walled off your sources, 00:10:27.519 --> 00:10:29.379 you've written a meticulous slide -by -slide 00:10:29.379 --> 00:10:32.940 prompt, and the AI has generated a gorgeous deck. 00:10:33.419 --> 00:10:36.360 You download the PPTX file, open it up because 00:10:36.360 --> 00:10:40.919 you spotted a tiny typo on slide 4. You double 00:10:40.919 --> 00:10:43.330 -click the text. And nothing happens. Nothing 00:10:43.330 --> 00:10:45.529 happens. The sheer panic of that moment. Oh, 00:10:45.629 --> 00:10:47.870 yeah. You've got a stakeholder meeting in 20 00:10:47.870 --> 00:10:50.110 minutes. You're furiously clicking the title 00:10:50.110 --> 00:10:52.289 to change your revenue number. And you realize 00:10:52.289 --> 00:10:55.029 you're just dragging the entire slide around 00:10:55.029 --> 00:10:57.669 the screen like a photograph. It's the most critical 00:10:57.669 --> 00:10:59.850 limitation in Notebook LM right now. And you 00:10:59.850 --> 00:11:01.669 absolutely must understand it before you build 00:11:01.669 --> 00:11:05.750 a workflow around it. Notebook LM exports. slides 00:11:05.750 --> 00:11:09.809 as flat images. The system relies on a backend 00:11:09.809 --> 00:11:12.889 rendering process called NanoBanana. Let's define 00:11:12.889 --> 00:11:15.429 that quickly for everyone. It's the backend software. 00:11:15.679 --> 00:11:18.759 rendering your slides into flat, single -layer 00:11:18.759 --> 00:11:21.960 images. Which means it makes text natively uneditable. 00:11:22.159 --> 00:11:24.700 It does not create editable text layers or vector 00:11:24.700 --> 00:11:27.220 shapes or individual elements the way a standard 00:11:27.220 --> 00:11:29.960 PowerPoint or Apple keynote file does natively. 00:11:30.279 --> 00:11:32.379 It's kind of like someone handing you a beautiful, 00:11:32.659 --> 00:11:34.919 perfectly framed painting of a spreadsheet instead 00:11:34.919 --> 00:11:39.080 of the actual Excel file. Wow. That is a painful 00:11:39.080 --> 00:11:42.539 but incredibly accurate analogy. You can admire 00:11:42.539 --> 00:11:46.000 the painting. of the Q3 revenue numbers on the 00:11:46.000 --> 00:11:48.440 wall. But you cannot click into the cells and 00:11:48.440 --> 00:11:50.580 update them. So we are stuck with a painting. 00:11:51.100 --> 00:11:54.320 How do we fix a typo on a painting? We have three 00:11:54.320 --> 00:11:56.620 distinct paths for a revision. Let's start with 00:11:56.620 --> 00:11:59.220 option A, which is revising entirely inside Notebook 00:11:59.220 --> 00:12:02.279 LM. This is your best strategic move if you need 00:12:02.279 --> 00:12:05.080 to make large content updates that affect multiple 00:12:05.080 --> 00:12:07.259 slides. You just hit the revise button inside 00:12:07.259 --> 00:12:09.539 the tool to open the slide editing view. But 00:12:09.539 --> 00:12:11.860 again, the AI is not a mind reader. If you use 00:12:11.860 --> 00:12:14.860 a vague prompt like, make this slide better, 00:12:15.100 --> 00:12:17.080 you're going to get a mess. Because better is 00:12:17.080 --> 00:12:19.580 a highly subjective human concept. Right. The 00:12:19.580 --> 00:12:21.960 AI has absolutely no idea what better means for 00:12:21.960 --> 00:12:23.899 your specific audience. You have to give it surgical 00:12:23.899 --> 00:12:26.240 structural instructions. Exactly. You define 00:12:26.240 --> 00:12:28.799 an update goal, make the slide feel more modern. 00:12:29.039 --> 00:12:31.220 than a specific title update. Replace the current 00:12:31.220 --> 00:12:34.299 title with, how AI automation saves teams 10 00:12:34.299 --> 00:12:37.059 hours every week. And you must constrain the 00:12:37.059 --> 00:12:40.379 content changes. Reduce the long paragraphs into 00:12:40.379 --> 00:12:43.279 three short bullet points. Crucially, you have 00:12:43.279 --> 00:12:45.480 to remind it of your original design instructions 00:12:45.480 --> 00:12:48.179 so it doesn't lose the plot and generate something 00:12:48.179 --> 00:12:50.799 completely mismatched. Right, like keep the dark 00:12:50.799 --> 00:12:54.159 background and blue accent colors. Yep. It will 00:12:54.159 --> 00:12:57.080 re -render the slide as a brand new image. Now, 00:12:57.220 --> 00:12:59.659 the backgrounds might shift slightly because 00:12:59.659 --> 00:13:02.179 it's generating a fresh picture, but it creates 00:13:02.179 --> 00:13:04.879 a new version without overwriting your original 00:13:04.879 --> 00:13:06.960 deck, which is a really nice safety net. That's 00:13:06.960 --> 00:13:09.539 good. Yeah. But what if my content is actually 00:13:09.539 --> 00:13:11.500 perfect and I just need to change the structural 00:13:11.500 --> 00:13:13.460 order of the presentation? That's where we move 00:13:13.460 --> 00:13:16.700 to option B. Exporting the PPTX file. OK, so 00:13:16.700 --> 00:13:19.120 this is ideal for basic layout adjustments. You 00:13:19.120 --> 00:13:21.620 download the file and you open it in Google Slides 00:13:21.620 --> 00:13:24.700 or standard PowerPoint by going to File, Import 00:13:24.700 --> 00:13:27.120 Slides, and Upload. Right. The slides themselves 00:13:27.120 --> 00:13:29.960 are still completely flat images, but now they 00:13:29.960 --> 00:13:32.919 exist in traditional presentation timeline. You 00:13:32.919 --> 00:13:35.519 can drag slide four to the front. You can delete 00:13:35.519 --> 00:13:38.740 slide seven entirely. And if you're truly desperate 00:13:38.740 --> 00:13:41.799 and in a rush, you can always just draw a new 00:13:41.799 --> 00:13:44.850 editable text box. right over the top of the 00:13:44.850 --> 00:13:46.990 flat image to cover something up. Oh yeah, it's 00:13:46.990 --> 00:13:48.809 a bit of a hack, but it definitely works for 00:13:48.809 --> 00:13:51.190 quick fixes. But what if I don't want to hack 00:13:51.190 --> 00:13:53.750 it with layered text boxes? What if I just need 00:13:53.750 --> 00:13:57.129 to fix one single word on one slide and I need 00:13:57.129 --> 00:13:59.490 it to look perfectly native? Then you need option 00:13:59.490 --> 00:14:03.009 C. Gemini Visualize. Walk me through how this 00:14:03.009 --> 00:14:06.149 actually works. It is absolutely brilliant for 00:14:06.149 --> 00:14:08.009 lightning fast single slide edits. You don't 00:14:08.009 --> 00:14:10.029 even need to leave Google Slides. You just use 00:14:10.029 --> 00:14:12.240 the Visualize side panel. right there in the 00:14:12.240 --> 00:14:14.879 interface. So I click on the flat image slide 00:14:14.879 --> 00:14:17.580 I want to change. I open that visualize panel 00:14:17.580 --> 00:14:20.940 and I just describe the correction. You literally 00:14:20.940 --> 00:14:23.899 just type change the title of this slide to three 00:14:23.899 --> 00:14:27.080 habits of highly productive workers. Gemini analyzes 00:14:27.080 --> 00:14:29.700 the original image, matches the font style, matches 00:14:29.700 --> 00:14:32.120 the background aesthetics, and generates a new 00:14:32.120 --> 00:14:34.279 corrected image version of the slide right there. 00:14:34.519 --> 00:14:38.120 Wow. It seamlessly rewrites the title and swaps 00:14:38.120 --> 00:14:40.220 the image, saving me the massive headache of 00:14:40.220 --> 00:14:42.980 going all the way back into Notebook LM, writing 00:14:42.980 --> 00:14:45.539 a new prompt, waiting for the render, and re 00:14:45.539 --> 00:14:47.779 -exporting. Exactly. It's a surgical strike. 00:14:48.240 --> 00:14:50.620 It fixes the pain point without abandoning the 00:14:50.620 --> 00:14:53.080 visual quality. I keep coming back to this design 00:14:53.080 --> 00:14:55.820 choice though. Why would the developers engineer 00:14:55.820 --> 00:14:59.120 it to export as uneditable flat images in the 00:14:59.120 --> 00:15:01.820 first place? It feels so counterintuitive to 00:15:01.820 --> 00:15:04.080 how we've worked for the last 30 years. Basically, 00:15:04.340 --> 00:15:07.299 rendering them as images ensures a much cleaner, 00:15:07.600 --> 00:15:10.320 highly polished visual output. without formatting 00:15:10.320 --> 00:15:12.639 breaking across different devices. So it trades 00:15:12.639 --> 00:15:16.000 easy text editing for a much more polished, consistent 00:15:16.000 --> 00:15:18.399 visual design. It's a highly calculated trade 00:15:18.399 --> 00:15:20.700 -off. Think about it. If it exported a bunch 00:15:20.700 --> 00:15:23.480 of portally aligned, messy text boxes that broke 00:15:23.480 --> 00:15:25.059 when you opened them on a different operating 00:15:25.059 --> 00:15:27.899 system, users would immediately complain that 00:15:27.899 --> 00:15:30.600 the AI was terrible at design. Fair point. Okay, 00:15:30.740 --> 00:15:32.320 let's step back and look at the whole picture 00:15:32.320 --> 00:15:34.480 here. What does this all mean for our workflow? 00:15:34.759 --> 00:15:37.139 Let's recap the big idea from today's deep dive. 00:15:37.340 --> 00:15:40.460 The overarching philosophy is that you must treat 00:15:40.460 --> 00:15:44.279 Notebook LM as a rapid visual framing tool. You 00:15:44.279 --> 00:15:46.899 cannot treat it like a traditional granular slide 00:15:46.899 --> 00:15:49.179 editor. Right. The heavy lifting, the actual 00:15:49.179 --> 00:15:51.820 secret to a perfect deck, does not happen in 00:15:51.820 --> 00:15:54.490 the editing phase. It is heavily weighted in 00:15:54.490 --> 00:15:57.549 the upfront prompting and relentlessly curating 00:15:57.549 --> 00:16:00.230 exact sources. If you put the intellectual work 00:16:00.230 --> 00:16:03.070 in before you click generate, the output feels 00:16:03.070 --> 00:16:06.230 incredibly precise. So remember the best practices. 00:16:06.669 --> 00:16:09.730 Wall off your reality. Always use verified sources 00:16:09.730 --> 00:16:12.190 when accuracy matters. Bring your own material 00:16:12.190 --> 00:16:14.669 and do not rely on random web searches. Make 00:16:14.669 --> 00:16:16.330 sure you're matching your format to the room, 00:16:16.570 --> 00:16:19.029 too. Use detailed decks if the audience is reading 00:16:19.029 --> 00:16:21.299 it alone. and present your slides if you're speaking 00:16:21.299 --> 00:16:23.860 on a stage. Keep your slide instructions in plain, 00:16:23.960 --> 00:16:26.720 direct language. No flowery descriptions needed. 00:16:27.059 --> 00:16:29.980 Just clarity. Break complex revisions into single 00:16:29.980 --> 00:16:32.179 steps so the AI doesn't get overwhelmed and drop 00:16:32.179 --> 00:16:34.539 details. And don't be afraid to create multiple 00:16:34.539 --> 00:16:36.580 presentation versions within a single notebook. 00:16:37.019 --> 00:16:39.080 You can test completely different styles without 00:16:39.080 --> 00:16:41.460 starting from scratch every single time. Thank 00:16:41.460 --> 00:16:43.929 you for joining us on this deep dive. You now 00:16:43.929 --> 00:16:46.330 have the exact blueprint to take scattered notes, 00:16:46.809 --> 00:16:48.710 build a walled garden of truth, and turn them 00:16:48.710 --> 00:16:51.070 into professional polished presentations without 00:16:51.070 --> 00:16:53.330 falling into the one -click trap. It really is 00:16:53.330 --> 00:16:55.509 a paradigm shift once you adapt to the boundaries 00:16:55.509 --> 00:16:58.789 of the tool, sources, structure, and targeted 00:16:58.789 --> 00:17:01.070 revision. Before we go, I want to leave you with 00:17:01.070 --> 00:17:03.370 a final thought to mull over, something that 00:17:03.370 --> 00:17:05.630 builds on everything we just discussed. I love 00:17:05.630 --> 00:17:07.589 these. Let's hear it. We've spent this whole 00:17:07.589 --> 00:17:10.549 time figuring out how to precisely manipulate 00:17:10.549 --> 00:17:14.400 an AI to output static flat images that mimic 00:17:14.400 --> 00:17:18.559 legacy software like PowerPoint. But if AI can 00:17:18.559 --> 00:17:21.779 already perfectly synthesize, brand, and visualize 00:17:21.779 --> 00:17:25.500 our dense data in mere minutes, how long until 00:17:25.500 --> 00:17:28.160 the traditional slide deck becomes entirely obsolete, 00:17:28.660 --> 00:17:31.619 replaced by dynamic AI -generated visual dashboards 00:17:31.619 --> 00:17:33.220 that we interact with in real time?