Sieve

Excited to launch Sieve’s Dubbing API in 1.0 — the first AI dubbing solution purpose-built for developers! Our API features: ✅ Speaker Style Preservation — preserve emotion & tone ✅ Multiple Speaker Support — distinct voices for each speaker ✅ 29 popular languages supported with voice cloning ✅ Background Noise Addition — more natural dubs with original noise included ✅ Language styles — specify styles such as "informal french", "shakespearean english", etc. ✅ Faster than realtime ✅ Customizable voice engines, safewords and more! Try it out with your own content here: https://lnkd.in/gKCVGNCC Check out some of our examples and learn more about how we built dubbing on Sieve's platform in our blog: https://lnkd.in/g6jX6gZC Looking forward to seeing what you build!

In honor of a ☘ sweep today, here's Kyrie telling Boston he plans to "re-sign" four years ago, now dubbed in Chinese (Mandarin). Developers, get ready for next week. We've been cooking 👀

Fresh paint 🎨 by the best artist in town Abhi U.!

Some people have asked us how we implemented a high-quality auto-cropping algorithm that’s also FAST when leading open-source models that detect active speakers accurately only run at 300% the length of an input video on a T4 GPU. Well low and behold, autocrop is powered by a powerful implementation of active speaker detection that we open-sourced a while back! Under the hood, it’s a parallelized combination of yolov8 and talknet-asd. You can read about it here: https://lnkd.in/e3-aDAVE Or check out the open-source code: https://lnkd.in/g3FWMNJN

Last week, we announced Autocrop in 1.0. We thought we’d share a bit more of what developers have been doing with it. By far, the most popular use case seems to be around repurposing long-form content (like a podcast) into shorter social clips. The idea is as follows: 1. Transcribe a video 2. Pass the transcript into an LLM to find just the most important or potentially viral moments 3. Autocrop them into different aspect ratios 4. Render captions onto the clips Now you’ve got 50+ edited clips, ready to go. Finding good “highlights” reliably requires some experimentation though. Thankfully, we wrote a blog about just that, relatively recently. Give it a read! Finding Highlights: https://lnkd.in/eJZTq8pB And if you haven’t checked out Autocrop 1.0 yet, read about it here! https://lnkd.in/d6PCi2Ta

Meet Autocrop 1.0. An API for social video editing. Speaker tracking. Auto reframing. One developer-focused API powered by the Sieve platform. ⏩ 3x realtime processing without rendering 💰 cost-effective pricing designed for developers (sievedata.com/pricing) 👪 powering 10+ video platforms already Try it out via our interface or through API here: https://lnkd.in/dXvKis2J Learn more about how we built it: https://lnkd.in/d6PCi2Ta

Meet Autocrop 1.0. An API for social video editing. Speaker tracking. Auto reframing. One developer-focused API powered by the Sieve platform. ⏩ 3x realtime processing without rendering 💰 cost-effective pricing designed for developers (sievedata.com/pricing) 👪 powering 10+ video platforms already Try it out via our interface or through API here: https://lnkd.in/dXvKis2J Learn more about how we built it: https://lnkd.in/d6PCi2Ta

Launching tomorrow, a new way to edit videos automatically with AI…

The video editing landscape is SO different from just a year ago and we have such a fun up-close view here at Sieve. Almost every team building a video editing product now has AI top of roadmap. Why? If you want to create a good video, it requires a lot of taste. But behind the scenes it’s also a very menial process. For example imagine you’re MKBHD about to absolutely trash the Humane AI Pin. What does your process look like? 𝐑𝐞𝐯𝐢𝐞𝐰 𝐚𝐧𝐝 𝐬𝐭𝐨𝐫𝐲𝐛𝐨𝐚𝐫𝐝 You first test the device and note down your thoughts. You then do your research on the company and the launch to contextualize your review for your viewers and tell a story. 𝐑𝐞𝐜𝐨𝐫𝐝 You now plan the shots you want and record yourself along with the device in various conditions. You probably do a bunch of retakes and end up with up to a TB of raw video footage. 𝐎𝐫𝐠𝐚𝐧𝐢𝐳𝐞 Most of the video footage you recorded just now is useless. You need to sift through hours or footage to find the right shots that have just the right talk tracks that flow well in the video. 𝐄𝐝𝐢𝐭 Now time to pull out all the signature MKBHD moves. Silky smooth intros, animations, text, transitions, coloration and more. 𝐃𝐢𝐬𝐭𝐫𝐢𝐛𝐮𝐭𝐞 You now post on YouTube but oh wait…what about Twitter, TikTok, YouTube Shorts, Instagram, Snapchat, Reels, and the bazillion other platforms people use to watch content. Time to re-edit to just the right aspect ratio and distribute. I know I’ve oversimplified the process but a lot of the what I’ve described isn’t necessarily creative. It’s menial. Product teams in video creation are starting to change this so creators get more time to be creative. Throughout this week, I’ll talk about how 🙂

We’re excited to unveil our case study with Zight, a leader in the video communication space. Zight uses Sieve to power all of the functionality that’s a part of Zight AI, their AI suite. In this post on the Zight blog, we interview Phin (Séraphin Hochart), Zight’s Head of Product, who goes through why AI is important to them and why they picked Sieve to power it all. Read the blog: https://lnkd.in/gwmQ8vP2 Or see Phin talk about the partnership below 👇