--- title: "I Built an AI App That Edits My Famous Dogs' Videos" description: "An AI video editing app for pet content creators can trim real footage, add captions, and keep approval human. See how Freddi Studio works." url: https://ziplyne.agency/blog/i-built-an-ai-app-that-edits-my-famous-dogs-videos published: 2026-10-14T16:02:15.114Z updated: 2026-10-14T16:02:15.114Z author: "ZipLyne Editorial" publisher: ZipLyne --- # I Built an AI App That Edits My Famous Dogs' Videos > An AI video editing app for pet content creators can trim real footage, add captions, and keep approval human. See how Freddi Studio works. ## Key takeaways - In Horowitz's Gemini analysis of 662 of the account's past reels, run before a line of Freddi Studio's code was written, about three quarters of the reels posted since 2024 turned out to be a single take with no cuts. - A Claude agent orchestrates the entire editing decision inside Freddi Studio, calling Apple Intelligence to find usable footage, the Mac's own speech recognition to time captions, Gemini to write hooks, ElevenLabs to generate music, and native Mac rendering to produce the final file, with every output routed back to Horowitz for manual approval before anything is final. - The agent can review clips, write hooks, set trims, and render two or three versions of each clip, per Horowitz's account of the workflow, for him to compare, but it cannot publish anything. ## Can AI Edit Your Pet Videos for You? Yes. An AI video editing app for pet content creators can edit real camera-roll footage instead of generating synthetic clips from a text prompt. Isaac Horowitz, founder of ZipLyne, built Freddi Studio, a native macOS app, to edit real footage of his own dogs for the @frankiandfreddi account, and nothing posts without his approval. An AI video editing app, in this sense, is software that takes footage a creator already shot and turns it into a finished, captioned, publishable clip, rather than inventing video from scratch. Horowitz built Freddi Studio for one specific account: @frankiandfreddi, which features his English Bulldogs Franki and Freddi and has grown to more than 1 million followers across Instagram and TikTok, by Horowitz's own count. Franki has since passed away, and between running ZipLyne and everything else, Horowitz hadn't had the time to edit and post the way the account used to. The operator lesson sits underneath the dog story: a narrow agent wired to specific tools, with a human still approving every output, can take over a real production job instead of just demoing one. That pattern applies as much to a founder's content pipeline as it does to any repetitive workflow a business wants off someone's plate. ![Diagram: I Built an AI App That Edits My Famous Dogs' Videos](https://pub-aee74429e0604b2da81fc4f8bd15a430.r2.dev/ziplyne-agency/i-built-an-ai-app-that-edits-my-famous-dogs-videos-a4f0bff9ec81c8ed.webp) ## What 662 Analyzed Reels Taught Isaac Horowitz About What Actually Works In Horowitz's Gemini analysis of 662 of the account's past reels, run before a line of Freddi Studio's code was written, about three quarters of the reels posted since 2024 turned out to be a single take with no cuts. That finding came directly from studying the account's own back catalog, not from general editing advice borrowed off the internet. The same analysis surfaced a second pattern: the account's biggest reels open on a short text hook rather than a flashy intro, and heavy cutting, zoom effects, and sound effects did not help performance. Clips chopped into multiple pieces and layered with effects consistently underperformed the plainer ones. **The data said less editing wins, so Freddi Studio does less editing.** That 662-reel analysis became the product spec. Instead of chopping footage into something busy, the app trims, captions, and scores clips the way the account's best-performing posts are already built: minimal cuts, a strong opening line, and nothing extra pulling attention away from the dog. This is also where Freddi Studio splits from most AI pet video tools on the market, which are built to generate flashy effects fast rather than study what one specific account's audience already responds to. Ready to see how ZipLyne can help? [Let's build something real](https://ziplyne.agency/contact) Watch ### How I Made a Dog Dance with AI 😱 | Viral AI Dog Dance Tutorial 2026 (Free Tool) From GR Editz on YouTube ## How Does the AI Agent Decide What to Edit? A Claude agent orchestrates the entire editing decision inside Freddi Studio, calling Apple Intelligence to find usable footage, the Mac's own speech recognition to time captions, Gemini to write hooks, ElevenLabs to generate music, and native Mac rendering to produce the final file, with every output routed back to Horowitz for manual approval before anything is final. No single model makes the whole call; each tool handles one narrow task. The sequence runs like this: 1. Apple Intelligence reads frames on the Mac to determine which camera-roll videos actually show Freddi, so the agent skips empty background footage before any editing starts. 2. The Mac's built-in speech recognition times every spoken word in the clip, which drives karaoke-style captions timed to the audio instead of generic subtitle placement. 3. The app checks camera metadata and scans for burned-in captions to separate raw footage from videos that were already edited and posted, so nothing gets re-edited that's already live. 4. Gemini reviews each surviving clip and writes the on-screen hook in the account's own voice, trained on examples pulled from past high-performing posts. 5. ElevenLabs generates the music bed for the edit. 6. Rendering happens natively on the Mac. FFmpeg handled the heavy lifting during the back-catalog analysis that shaped the whole system. 7. The Claude agent ties every step together through Freddi Studio's own MCP server, calling each tool in order and packaging the result for Horowitz to review. That's what [real AI app development](https://ziplyne.agency/blog/what-real-ai-app-development-looks-like-in-production) looks like once it's built for production instead of a demo: narrow tools, each doing one job, chained together by an agent that can actually call them in the right order. ## Keep reading - [How We Built an AI Podcast That Writes, Voices and](https://ziplyne.agency/blog/how-we-built-an-ai-podcast-that-writes-voices-and-publishes) - [What Real AI App Development Looks Like in Production](https://ziplyne.agency/blog/what-real-ai-app-development-looks-like-in-production) - [How Founders Should Brief an Agency for an AI Workflow](https://ziplyne.agency/blog/how-founders-should-brief-an-agency-for-an-ai-workflow) - [Custom AI SaaS Development for Nontechnical Founders](https://ziplyne.agency/blog/custom-ai-saas-development-for-nontechnical-founders) ## What Still Needs Approval Before Anything Posts? The agent can review clips, write hooks, set trims, and render two or three versions of each clip, per Horowitz's account of the workflow, for him to compare, but it cannot publish anything. Every post to @frankiandfreddi still requires a manual tap from Horowitz before it goes live, and that line doesn't move no matter how capable the agent gets. That boundary is deliberate, not a limitation left unresolved. A lot of current AI content hype describes systems running an account end to end with zero human involvement. Freddi Studio isn't built that way, and Horowitz says he doesn't want it to be. The agent can ask. It can't decide. ⚠️ Watch out Any pet video tool that claims it posts on full autopilot is skipping the one step that protects an account: a human checking the output before it goes public. The same principle shows up in other production AI builds, where [chaining AI workflows together cuts real admin work](https://ziplyne.agency/blog/built-ai-system-replaced-20-hours-weekly-admin-work) without removing the human checkpoint that keeps the output safe to ship. Agents handle the repeatable grind; a person still signs off on anything that leaves the building. ## What Horowitz Is Telling the Agent to Do With His Dog's Next Video Horowitz's next move with the agent follows one simple instruction: pull up the last video shot of Freddi, review it, write a hook in the account's voice, set the trims, render a couple of versions, and wait for a tap of approval before anything posts. The workflow runs through the same stack built for the rest of the account: Apple Intelligence to confirm the dog is in frame, Gemini for the hook, ElevenLabs for music, and native rendering on the Mac, all called by the Claude agent through Freddi Studio's MCP server. The actual task list looks like this: 1. Pull the most recent camera-roll clip where Apple Intelligence confirms Freddi is in frame. 2. Review the footage and flag the usable sections. 3. Write a hook in the account's voice, pulling from past posts that performed well. 4. Set the trim and cuts, leaning toward a single take over heavy editing. 5. Render two or three versions, for example a different hook on one and no music on another so a trending sound can be dropped in later. 6. Ask for approval before anything goes near @frankiandfreddi. Results from this pipeline are coming out pretty good, according to Horowitz, which is reason enough to keep running real footage through it instead of letting it sit unused on a phone. The same lesson applies past dog videos: in [AI automation for traditional businesses](https://ziplyne.agency/blog/ai-automation-traditional-businesses-step-by-step-transformation-guide), a narrow agent wired to the right tools, with a human still checking the output, can take over work that used to eat up real hours every week. That's Horowitz's rant for today. If there's a workflow buried under manual editing, admin work, or repetitive steps somewhere in a business, the fix usually isn't more hustle. It's building an agent that does the boring part and still asks before anything ships. ## Frequently asked questions ### What is Freddi Studio? Freddi Studio is a native macOS app Isaac Horowitz built to edit real camera-roll footage of his English Bulldogs for the @frankiandfreddi account. It pulls videos directly from Photos, runs them through a Claude agent that calls Apple Intelligence, Gemini, and ElevenLabs, and renders the finished clip natively on the Mac. It was built for one specific account, not sold as a general tool. ### Does Freddi Studio post pet videos automatically? No. The agent inside Freddi Studio can review clips, write hooks, set trims, and render multiple versions, but it cannot publish anything on its own. Every post to @frankiandfreddi still requires a manual tap of approval from Horowitz before it goes live. That checkpoint stays in place no matter how capable the underlying models get. ### Can AI turn a pet photo or text prompt into a video? Not in this system. Freddi Studio edits footage a creator already shot on their camera roll instead of generating synthetic clips from a photo or text prompt. Apple Intelligence first confirms the dog actually appears in a given clip, then the agent trims, captions, and scores that real footage rather than inventing new video from scratch. ### Which AI tools does the editing agent call on, and what does each one do? A Claude agent orchestrates five separate tools instead of relying on one model for everything. Apple Intelligence scans frames to confirm the dog is in the clip, the Mac's built-in speech recognition times karaoke-style captions, Gemini writes the on-screen hook in the account's own voice, ElevenLabs generates the music bed, and rendering happens natively on the Mac. ### Is Freddi Studio available for other pet content creators to use? No, it was built as a one-off tool for a single account. Horowitz created Freddi Studio specifically to edit footage for @frankiandfreddi, his English Bulldogs' page, rather than as a packaged product for other creators. The app is tuned to that account's own posting history and voice, which makes it a custom build, not a general release. ### How big is the @frankiandfreddi audience that this editing system is built for? The account has grown to more than 1 million followers across Instagram and TikTok combined, by Horowitz's own count. It features his English Bulldogs Franki and Freddi, though Franki has since passed away. That audience size is why studying 662 past reels mattered: the editing system was built around patterns that already worked for that specific following. ## Ready to see how ZipLyne can help? [Let's build something real](https://ziplyne.agency/contact) ZipLyne Editorial Editorial Team Editorial desk for ZipLyne. --- Originally published at https://ziplyne.agency/blog/i-built-an-ai-app-that-edits-my-famous-dogs-videos