I Make 30s AI Videos for FREE Every Day (Google Flow)

0:00<i>I don't exist.</i> <i>Google made me this morning.</i> <i>For free.</i> <i>And you probably didn't</i> <i>notice until I told you.</i> He's right. His name is Mr. Smith, and I made him in Google Flow with less than one day's free credit. With Google Flow, you can repeat making a 30 second video similar to this one for free, every day.

0:21And in this video, I'll show you exactly how to make a video just like this one. I make 30 second shorts using my own avatar, and this time I wanted to see if a Flow character could hold up just as well. First, we make our character, Mr. Smith. Then we create three 10 second clips in a sequence. Flow charges credits for every video,

0:38and Google gives you 50 credits a day, even on a free account. A 10 second clip costs 15 credits. So the three clips that we're gonna make going to cost 45 out of that 50. To start, go to flow.google and sign in with your Google account. Create a new project, and you'll have a choice to make. You can make your own avatar, which copies your face and voice,

0:56or make a character to stand in for you. Your own avatar needs a phone to set up and isn't available in UK, the EU or Switzerland. So I'm using a character for this showcase. I picked a professional template, and Flow filled in the prompt for me. Then I clicked format button right there, which adds more details to the prompt.

1:12That's optional. So feel free to describe exactly how you want your character to look like. I just didn't want to think about it at this time. I switched the image model to Nano Banana Pro and kept the picture landscape for a regular YouTube video. Flow offered to make a full body picture too. But a talking head only needs the face, so I skipped it.

1:28Once it was created, he landed in the Characters tab next to my own avatar. I blurred my face because I don't like how my mouth looks in that picture. I have no idea why. I know I look way more handsome in person. Anyways, I named him Mr. Smith. Flow also lets you pick a voice for a character.

1:42But I left that alone and let Flow choose. For consistency, I should have picked the voice. So don't forget to do it, because the voice could change from clip to clip as you create those three consecutive clips separately. Now he needed something to say. So I had ChatGPT write a video prompt with his lines.

1:57But any chat AI works. The main rule I gave it was 10 seconds tops per clip. I discovered a better way to do this step, and I'll show you at the end. Back in Flow, I turned on Agent mode, Flow's built-in assistant. In settings, make sure it's set to one video. Otherwise, you'll end up paying for two for the same clip.

2:12Before it spends anything, the agent shows you the cost and waits for your okay. This one was 15 credits. In my first attempt for this demo, I made a mistake and forgot to add Mr. Smith to the prompt. So Flow used my own avatar, and I watched myself read his lines. It's funny, but it's also 15 credits gone. On a free account,

2:28that mistake leaves you enough credits for two clips instead of three. To avoid that, click the plus sign next to agent and add your character before you paste the prompt. Then paste the same prompt, approve the 15 credits, and let it run. This is how that one came out. It looks good, but it came out looking younger than the character I made.

2:44Then I noticed that captions were burnt into the video, which I never asked for. Video models sometimes add them on their own. For the next two clips, I edited the prompt for them to avoid adding captions or any text on screen, and that fixed it for me. This next part turns 10 seconds into 30, and it's where most free users quit.

3:00Flow won't extend these clips for you, so here's the manual workaround I use. And by the way, this workflow is before I created my Flow MCP, which actually does this automatically for you. You can find the link to that video down in the description. You may notice that there is an extend button, but right now it only works on clips from Google's Veo models,

3:17not Omni Flash, the ones that we are using, because these are more powerful. So the 10 second clips that we are making right now, we are choosing the model called Omni Flash. The idea is to take the last picture of one clip, which Flow calls a frame, and start the next clip from it.

3:30So the new clip picks up exactly where the old ones stopped. First, I asked ChatGPT to generate a prompt for the next 10 seconds, starting with Mr. Smith saying, "Oh, by the way, I almost forgot," and then one more point. The third clip would be his closing remarks. From this point on, you need to turn off Agent mode,

3:44because the regular prompt box is where you add a starting frame. To grab the last frame, go to All Media and open your first clip. Drag the timeline all the way to the end. Then click Save frame at the top right of the picture. Click Done, and the frame shows up in All Media like any other picture. Now add that picture as your starting frame,

4:00paste in the second clip prompt, and hit Generate. That's it. Second clip starts from the exact frame where the clip 1 ended, and the voice stayed the same, even though I never picked one. I guess I got lucky. The one thing that I noticed changed was actually the lighting, because clip 2 came out brighter than clip 1.

4:16I'll come back to that at the end as well. Clip 3 is the same routine: save the last frame of clip 2, paste in the closing lines, and generate from that frame. With all 3 clips done, it's time to put them together. Open your first clip, click the plus button after it, and add the second clip.

4:30I added clip 1 twice by accident, so watch for that. Naming each clip after they are done will help, so remember to do that. Then, trim the little pauses between the clips by dragging the clip ends, so it plays like one take. When it looks right, click the download arrow in the top right corner,

4:44and you get all 30 seconds as one file. Flow can also make vertical clips. So if you want a short or a Reel, pick vertical before you generate anything. Now let's take a look at how Mr. Smith came out. All 30 seconds of it. <i>Claude Code can scan your entire codebase,</i> <i>understand how everything connects,</i> <i>find problems, write the fix,</i> <i>and actually make the changes for you.</i> <i>And you can do all of this just</i> <i>by telling it what you want.</i> <i>Oh, by the way, I almost forgot.</i> <i>Claude Code can also run commands,</i> <i>test its changes, and fix what breaks.</i> <i>So it's not just writing code.</i> <i>It can actually work through the problem.</i>

5:14<i>And that's why Claude Code</i> <i>feels different.</i> <i>You're not just asking</i> <i>AI for snippets anymore.</i> <i>You're giving it a task,</i> <i>letting it work through the codebase,</i> <i>and reviewing what it changed.</i> So what came out is actually pretty much usable, and I would definitely use it, especially for creating vertical shorts, because it follows the exact same steps.

5:30You can see where it fell short, though. There are captions in the first 10 seconds, and the light jumps when clip 2 starts. When I use clips like this, I fix the lighting in my editor by matching the exposure across all three. You can also cover the joins with a b-roll or transitions, which actually makes going

5:44from one clip to another a lot smoother, with no distinct differences. If I did this again though, I'd change two things. First, I'd ask ChatGPT for all three clips at once, with no captions line in every prompt. Second, I'd have Mr. Smith end each clip still and looking at the camera, so that last frame is a clean place to start. I'll put the updated prompts in my free Builders Club community,

6:04and the link is in the description for that. So everything you just saw, I did by hand in Google Flow. Why did I say it was manual? Because there's an automated way to do it, and I built a tool for it. It's a free Claude tool that does all of it for you, without you ever touching the Flow app.

6:16The link to that video is on your screen right now. Watch that one next.