How I Built an AI Video Factory That Creates 20 Videos a Week While I Sleep.

Hey guys, it’s Ritz here.

I know it’s been a while since I sent my last newsletter. There’s been a lot of advancements in AI lately — DeepSeek version 3.1 just got released, new image and video generation models have emerged, Roboflow released their RF-DETR model that beats YOLO on real-time object detection, and OpenAI updated their voice features with more expressive options. With great technology comes great possibilities, and I’m excited to show you what we’ve been working on behind the scenes at Augmented AI.

How We’re Automating Video Production for ARIA: Our AI Parenting App

For those who don’t know, I’ve been working on an AI for parents called ARIA. If you’re a parent reading this, you should definitely check it out. It’s essentially ChatGPT but backed by science, much more personalized, and uses a cool agentic framework for switching between a RAG model and a conversational chat model.

_{We trained a custom LORA to get consistent characters for Animatic.}

The challenge was creating content to promote it. I didn’t have it in me to be the influencer in front of the camera, which led me to explore video generation models. I was blown away by what’s possible right now, which sparked an idea: what if I could create content automatically to promote the ARIA AI Parenting App?

From Idea to 4 Videos Daily: The Birth of Our Automated Video Production Line

This led me to create Animatic, a technology that might seem simple at first but is capable of generating many videos in just under 10 minutes — which is crazy! I’m trying a bulk approach where we generate over 20 videos a week (4 videos per day) to see how we can scale marketing organically for the ARIA AI app.

We started by testing animation videos in the style of DreamWorks. It looks pretty cool, right? We can create content in many different styles — low polygon, Disney 3D, Samurai Jack style, chibi anime, lego movie, photorealistic, or cinematic styles — you name it. I’ve been particularly drawn to the DreamWorks style because it’s eye-catching.

How Animatic Works

Here’s the process behind Animatic:

1. Topic Generation: I scraped science-backed and expert-backed sources from the internet and created over 760 potential parenting topics in JSON format.
2. Storyboarding: The system creates prompts for voiceover scripts, image generation, and video sequences. Each storyboard has about 10–15 scenes, with each scene lasting 5 seconds.
3. Image Generation: We convert each scene into images using carefully engineered prompts. After testing many models, we found Flux 1.1 worked best for our needs. We also tried Google Imagen, but its safeguards limited our creative freedom, preventing us from generating DreamWorks or Disney-style images. The Reve models also yielded some great results.
4. Video Generation: We experimented with various models, including Google’s Veo2, Minimax, and Kling 1.6, which gave us excellent results.
5. Audio Generation: From the storyboards, we extracted the scenes and converted each script into speech. We tested ElevenLabs, which was the best at the time, but they all sounded somewhat robotic. Recently, OpenAI released their new text-to-speech model with more expressive capabilities, which I really like as it gives greater freedom over the narrative.
6. Combining Elements: We match videos with audio, ensuring they’re synchronized properly. If audio is longer than video, the video pauses — something we’re still working to improve. Perhaps a boomerang effect could solve this in the future.
7. Subtitling: We add captions using a local downloadable model. While it would’ve been easier to use the original script text as subtitles, aligning it with the audio was challenging. Instead, we use a transcription model and then a smaller language model (Gemini Flash 2) to correct any discrepancies between the storyboard text and transcribed audio.
8. Distribution: Out of sheer laziness, I created a script to automatically upload videos to YouTube, and I’m working on solutions for TikTok and Instagram once I figure out their APIs.

Where to from here with Animatic

I’m excited about this automated production line for videos. Right now, the ARIA parenting YouTube channel has around 13 subscribers and under 20 followers on other platforms, but I’ll be running this test for the next six months to see how many subscribers we can gain.

I’m adamant about this approach because automating our organic marketing channels means more customers for the ARIA AI app, which means more revenue. But Animatic itself has already proven valuable — we’ve been able to use this technology to generate videos for other companies and a university.

Another company, Meratutor from India, is creating an AI course creation platform and will be using our service to generate full courses from a single prompt, which is pretty amazing. Being in the course creation business, I know it used to cost thousands of dollars and take months to create videos. Now, with this technology, we can create them much quicker and more cost-effectively.

How You Can Create Professional Videos with Animatic While You Sleep

We plan to make Animatic self-serving, with a front-end interface for users to generate videos on their own terms. We’ll also offer an API to automate the process, just like we’ve done internally.

If you’re interested in this type of solution, we can create bespoke videos for your company. If you need videos fast and cost-effectively, let me know — I’d be happy to generate some free samples to demonstrate Animatic’s power.

At Augmented AI also offer an option to host the entire codebase (a more expensive solution where you would own part of the IP), or we can create bespoke videos for you. The self-serving version for the masses will be much cheaper, with pricing based on API calls.

That’s what we’ve been working on, and I’d love to hear what you think about this technology. The models are getting better every month, and I’m excited about how easy it is to generate high-end professional graphics for a fraction of the cost.

Looking forward to hearing from you, and I hope you enjoyed this update. See you in the next one!— Ritesh Kanjee