Master Google Opal: AI Automation for Everyone

Key Takeaways

So apparently Google is working on Opal which is NA10 like workflow builder where you can prompt and generate AI based workflows. And in this video we're going to look at seven different apps or flows that you can prompt and build from scratch. And we're also going to learn how we can customize these workflows in case you notice any errors. Now the amazing thing about Opal is that it completely integrates within the Google Workspace. So you can save into a drive, you can generate Google docs or slides or you can provide any file from your drive or a YouTube video. So anything that is Google is completely integrated into Opel. So let's get started. So as soon as we open Opel from Google, we'll notice that there are a couple of options here to select from the gallery of apps built by Google team. So these are some of the most common ones. You can see social media post generation or blog post generation. And these are quite nice. If you open any of these, you'll see that it contains some notes or modules here and each of those goes in a sequence. So the first one here is where we provide some input. Then there are some steps happening in the middle and then there is the output. So that's the common format that you'll notice that we have an input an output and a few in bit. So now if we were to start with a new app then we have two options. First is to build something manually or the other option is to type in and let Google do the work for us. And as mentioned before, we can select an input and that goes to some sort of generation or you can think of an agent that is doing the task. So you can pretty much you know supply your input to this agent and then you'll notice right away that there is the user input variable available. So if you were to change the label here, you'd probably see the same updated label and then last is the output. Once you have your flow completed, you can always supply that as an output. And this is where you are going to see how it's going to look like. Now, for each of these components, there are a few settings that you could benefit from. In the first user input, this is where a user is going to provide some info. You can add a prompt here so that your model is going to take the prompt and the user input and then it is going to perform some tasks. Now, for the user input, you can go to the advanced setting. If needed, you can restrict the input to a certain format. So this is quite beneficial. For now, we're going to keep this open. In the generate component, this is your agent that is going to be doing the task. There is the system instructions available here. And then there is the prompt that the agent is going to be using. And then there are options to select different models. There are the simple ones where you can call the Gemini 2.5 flash or you can call 2.5 Pro. U depending on complexity, these could be helpful. But then something that is quite interesting is there are a bunch of additional models here or options that you can use. So you can perform a complex planning and execution task. You can do deep research. So this is something that was made available within Gemini. And then many other providers offers deep research on any given topic. So you can add that as well to your app. Um you can also generate images. You can also use the latest nano banana model if needed. And then of course the notebook LM style podcast generation that's also available and a few additional tool from Google. Now once you have this step this is something where you can have the human in the loop as well. So if you were to select review with user this is where the user is going to approve or it is going to be automatically approved if not selected. And within output there are a few options as well. You can have a manual layout which is the type content. This is what we see in many chat apps. Now there is the web page format as well and then you have the option to save it to one of the Google suite products. In most cases if you're not saving as any of these Googlebased products you're probably going to be using manual which is type content or a web page which is the HTML based content which is also quite nice actually. It's almost like many web app that you're building here. So now that we know how the canvas looks like we are going to go on to the first app that we're going to be building. And quickly before we go to the first prompt, I also want to show that there is an option that you can add assets. So these could be files, YouTube videos, any additional things. And this could also be reference in your app. So for example, if I were to add something like a drawing, I can pretty much use the component make a drawing here of any simple thing, apply it, and then I can use it as an input to the agent. So in addition to the user input, now we have the drawing. So it could be any of these here. And especially something that could be useful is YouTube videos. You can provide any YouTube video or have that option for the user to input that as a video and then you can generate summaries, timestamps or any additional thing that you would like to do from a YouTube video. So let's get started with our app number one which is a slide deck creator. So we're going to start new and we are going to provide a simple prompt where we are going to ask user to upload a PDF document and number of slides required and based of the PDF content we're going to generate the presentation. So this is the exact prompt that I use. The key thing here is that we want the presentation to be in HTML format. So it looks better than just textbased slides. So we're going to give it a try. Now one thing is that we probably are just going to use the PDF content instead of actually the PDF uploading. provide a PDF document content or we could have just said document content since we don't need the PDF document. Cool. So we have our first flow. Now what I noticed is that the PDF document is analyzed here. So it's asking for us to add a document from the drive. Probably that I could have provided a better prompt to use a document content instead of the PDF. So regardless what's going to happen is the content is going to go from here. The number of slides are going to go to the agent. Based on that we are going to be sending the prompt with these details. So now the cool thing I was able to draft the prompt and also some information as what are the input. Um there are also some system instructions those were and with that it can render presentation and then there is also some information about the rendering of the slide. So this is quite nice. Now I'm going to provide a test PDF and this is two pages from a research paper and I'm going to ask for it to generate slides probably four slides is good and let's see how it works. Cool. Seems like the presentation was generated. It's more still like a text and it gives some info about the slides. Um not the best type of slides that I would like but again you can change the initial prompt and then modify the approach here. Um, what you could also do is probably go to the system instructions here and then modify this. There was another run that I had earlier which pretty much did similar tasks but it had better visual representation. So what I'm going to do is I'm just going to take the prompt that was available in this case. Go back to the newly created app and that's where I'm going to add the prompt and also I'm going to modify to look similar to how we had before. So, we're just going to use the simple prompt and paste it here. And we are going to remove the additional details here as well. Let's take that out. And I'm going to connect this one thing here. And then since it's generate presentation, this is the agent. I'm going to remove what we had from before. And this should be fine. So, let's give it a shot. So, the result is a lot better than the first one. So, just HTML with slice. And this is what I wanted. Cool. Second app is going to be a PDF to blog writer. It is going to be a simple prompt here that we're going to be going from a PDF to a well ststructured blog post. Very similar to what we did last time. It's just that instead of the presentation, we want something in writing. This should be quite simple. All right. Seems like there is some issue with the app generation. Even though it should have been easily done cuz it's going from a PDF. I'm just going to go with what it's suggesting right now to see how it work. Okay. Seems like the copy paste doesn't work. So, I'm going to type that in. Or better, I'm just going to use the microphone option here. an app that takes user provided context and generates a well ststructured blog post based on the content. Cool. Let's get started with this. All right, so this should be easy. It's going to render it as a web page. Great. We'll start. All right, so my prompt is the evolution of Steam Engine. Hopefully, it's going to be generated quickly with our blog generator agent. Great. We have the response and this is as expected. It was a simple app to make and the prompt was simple as well. Next, I wanted to generate flashc cards. So, it's very similar. We're just going to go from a given content to flash cards. And this is going to make sure that there's proper formatting for the flash cards and we can flip them. We can see what is behind and we run into same issue. It's weird though that it doesn't want to go with the PDF option even though there is the upload file opt. So I'm just going to probably modify for now to take a user input instead of the PDF and go from there. an app that takes user provided text content and a specified number of flash cards as input and generates the requested number of flash cards based on the provided text. Awesome. So, we have our flow which takes the input flash cards, sends it to the agent, and then generates the final rendering. So, I'm going to give it a try. I'm going to go with the top healthiest natural foods and probably go with something like 10 foods. Each one of those is going to have one fruit or vegetable or whatever it decides to pick. Okay, so it generated the flash cards, but as we saw last time, it is similar to a list. So what I'm going to do is I'm going to remove this. I am just going to go to the flashcard list and make sure that it is web page with auto layout. I am going to remove all of the info here and just ask for it to generate a web page. All right, keeping it simple. Generate flashcards in HTML that are flippable. Let's see if this is something it works with. So I'm going to remove this connection and then just run based on the provided info. So it has info from the generate flashc card and we are going to probably remove the text content since we don't need that. Just simple output from generate flashc card. Now we haven't modified anything with the agent. Hopefully it understands what we want and if not we'll probably modify from there. Cool. We're getting closer. Seems like we have the list of flash cards and then we can study them. This is nice. We could have asked for each flash card on a single page. So something like those slides, but we're just going to give it a go. This looks pretty good for me. And number four, we want to make the app a little more complex. So we want to use the resume of a person or the content of resume, analyze it. And as for our app to do a skill match percentage and also provide suggestions. So this is going to be a little more involved even though it's not that difficult to be honest. If you've used other nodebased tools, then probably you know how to build this. The key here is we want to see abilities of Google Opal especially the agentbased node generation. It's going to take the job URL description go from there to our first agent which is going to pull the web page info and based on that pass it to the next agent which is going to analyze the resume and based on that it has some instructions. So this is nice to be honest. I really like all of the prompt and instructions those are generated for the agent. So this is quite helpful. Final comparison report web page. It is going to be in a webpage format. So, it has certain layout, organization instructions, and some styling instructions. I'm not going to test this one as of now cuz you are probably better judge at looking at the resume and job description and the analysis than a generic resume. So, we're going to pass the testing on this and go on to app number five, which is a landing page optimizer. Now, the goal here is I want to provide one of my product landing page and ask for this app to help me improve the conversion rate. Great. So, the flow and the app is ready and we are going to test this out. For testing purposes, I'm going to use one of the products that I built. This is a PDF conversion platform which takes your PDFs and converts into a nice explainer video. Do check it out. It is used by tons of professors, students, and professionals. So, it is something I would highly recommend. So, I'm going to see what I can improve with this web page. Now, this run is taking quite some time, and I wanted to check what's happening. So the cool thing here is in the console section you can actually see details of what's going on behind the scenes. So I provided an input here for brainy documents. It went to the first model got some info back and this is how the model input and output looks like. So if you wanted to do some sort of troubleshooting or some sort of understanding as how the models are called then this is quite nice. uh once you have a certain output from a component or a node here you're going to see the output of that particular node. So these are three different calls and from all of that we gathered all of this info. This looks great. So this is actually the text from the website and then it went to the second component here and then there's of course the console the output from that component which is the conversion recommendations and then last is the step that we're waiting on the recommendation rendering. Now there is a chance that there probably could have been some issue with the last step or we're just going to give it a few more seconds to see what are the recommendations. In the meantime, I at least have the recommendations available here and I can make a note of what all changes to make to the product. So after a few seconds of running, it seemed like there was an error. So what I did is I just went to the play button for that particular component and I ran it. So it makes it a bit convenient. So I don't have to run the full flow again and it is going to save hopefully a few seconds. Cool. So now I have the output based on all of the recommendations. This is the suggestion. So, it's asking for us to modify some of the messaging on the website to make it look better and with better conversion. Now, this definitely seems like there are a few places I can make changes to the website. This is definitely quite nice. At number six, so we're going to create a simple app that is going to take the occasion from a person along with their name and desired language and generate a nice message to wish them. So, it could be for an anniversary or birthday or any occasion that one is celebrating. So, let's see how this goes. Cool. This seems like a bit more involved. This also has the audio and the music generation. So, I'm going to go on to testing this out. All right. So, I gave the input. Seems like it's going through the different steps here. Going to quickly check the console. Okay. Very nice. So, there is some generation of speech. >> Happy anniversary, Michael. Wishing you a day filled with cherished memories. >> That's pretty good. All right. So, the output is available. >> Wonderful. Here's [music] to many more years of >> good. On to our seventh app and this is going to be a UGC style video from a product image. We're going to provide product image and describe the product. We're going to ask for it to generate a video ad based on this info. All right, so the usual drill. Let's test this out. Now, for this particular product, I'm going to be given my app screenshots for an app that I built. And this is Style Maxi. It takes image of your outfit and gives you a style rating. And also, it's got virtual try on and also wardrobe. So, let's see. For the description, I'm probably just going to grab the info from the description on the app store. Let's see what it is. Seems like it didn't like that I gave multiple images. I'm going to ask for it to modify and use multiple images instead of just one image. It modified it, but there seems to be some error. Not something with what we're doing, but something with Google Server. So, we're going to keep this as is. We already got seven apps built in just a few minutes. So, this is pretty cool. We're going to explore more of these workflow builders and VI coding tools. Drop in the comment section below what you would like for us to build in the next video.