mok
@mokshith_v
Followers
770
Following
738
Media
60
Statuses
451
ai x video @sievedata
San Francisco, CA
Joined April 2018
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Brazil
• 960436 Tweets
Elon Musk
• 658004 Tweets
O Brasil
• 653507 Tweets
Chelsea
• 526347 Tweets
Alexandre de Moraes
• 469678 Tweets
Osimhen
• 281095 Tweets
Sancho
• 250066 Tweets
Sterling
• 211520 Tweets
Bluesky
• 178597 Tweets
Xandão
• 145112 Tweets
Politico
• 137848 Tweets
Napoli
• 135594 Tweets
Adeus Twitter
• 134207 Tweets
#مولد_ولي_العهد
• 104647 Tweets
Threads
• 97576 Tweets
Tchau
• 69685 Tweets
Labor Day
• 68256 Tweets
Eddie
• 64393 Tweets
Temple
• 58088 Tweets
VPNs
• 42449 Tweets
CONGRATULATIONS RM
• 39916 Tweets
#الجناح_مطلب_اهلاوي
• 31233 Tweets
CONGRATULATIONS NAMJOON
• 29878 Tweets
Michigan State
• 22145 Tweets
Oklahoma
• 18851 Tweets
ジェシー
• 18223 Tweets
Caitlin Clark
• 16194 Tweets
#ウルトラマンアーク
• 14893 Tweets
野菜の日
• 13053 Tweets
#شكرا_الوليد_بن_طلال
• 11702 Tweets
Shelton
• 11104 Tweets
Evades
• 10492 Tweets
Tiafoe
• 10116 Tweets
Introducing Tasmania - a better YouTube video search engine.
Every result links to a specific moment in the video, so you can pinpoint the exact part of a video that made it surface in the results!
24
70
716
1/ Today we’re excited to unveil Sieve (
@sievedata
), a cloud native platform for processing, searching, and running all sorts of AI models on video.
53
44
218
After months of feedback from early customers, we’re launching Autocrop in 1.0.
With this release, developers have a dead simple way to super charge their video platforms for today’s age of short-form social content.
Some of the story behind why we built this:
Meet Autocrop 1.0. An API for social video editing.
Speaker tracking. Auto reframing. One developer-focused API powered by the Sieve platform.
1
5
32
1
1
77
Excited to announce that
@sievedata
is a part of
@ycombinator
's W22 batch 🎉
We're building core video analytics infra for the next generation of smart apps in industries from security, to media, supply chain, and agriculture.
Super stoked to be on this journey with
@thpicy
:)
3
3
39
These are new waters for us as a company but I'm finally excited to share the news! The world is transforming right beneath our feet because of generative AI. We think this partnership will bring that transformation to the world, 10x faster.
Read more:
6
2
38
Thrilled to be releasing the 1.0 of our Dubbing API. It’s the first AI Dubbing solution purpose-built for developers. High quality, comes with a ton of knobs, and packaged with friendly pricing.
It’s another step we’re taking to building production-ready solutions on top of the
Excited to launch Sieve’s Dubbing API in 1.0 — the first AI dubbing solution purpose-built for developers!
✅ Speaker Style Preservation
✅ Multiple Speaker Support
✅ 29 popular languages
✅ Customizable voice engines, and more!
More details in 🧵(1/6)
3
8
123
0
1
40
@DeveloperHarris
this is because AWS and GCP are now just hardware renting businesses. it's up to the startups to build experiences developers truly love. and that's not a bad deal for Amazon or Google as a business since they still get a rake + outsource product development and marketing!
3
0
22
Many companies tend to store petabytes of data just in case. There's way too little attention put towards how to make use of that raw data effectively. I wrote a little about it recently on Towards Data Science :)
0
4
24
Transcribe 1h of audio in 21s for $0.05, with best-in-classy accuracy on
@sievedata
🦧. That's 𝟓𝐱 𝐜𝐡𝐞𝐚𝐩𝐞𝐫 than the next cheapest transcription API.
Powered by
@OpenAI
's Whisper-Large-V3, with many optimizations on top. Cheap, accurate speech-to-text is here. 🧵
2
0
24
pretty soon i don't think we'll have to train image or video classifiers at all for anything. was pretty fascinated with moondream (shoutout
@vikhyatk
) so i made a thing that can classify any video in under 10s.
2
1
23
5/ We'd also like to thank our amazing investors including
@MatrixVC
, ,
@YCombinator
, AI Grant (
@natfriedman
and
@danielgross
), and a host of other angels.
1
0
19
trying out early versions of a new thing we're working on, it's still a super rough experience but something about it tells me we're barking up the right tree
3
0
17
3/
@sievedata
's API lets developers create a project with a workflow, push video to it, integrate into a web app or a dashboard, and submit feedback to improve models over time. All without setting up a separate backend or having to think about infrastructure.
1
1
18
it is cool to see that many AI products that took top notch research teams ~years to refine and actually ship back in 2020 can be done in ~months or ~days now by a single engineer due to awesome OSS.
0
0
17
The entire backend is powered by an alpha embedding search feature in
@sievedata
of course and the frontend is hosted on
@vercel
with
@nextjs
. Reach out if it's something you'd like to play around with!
1
0
17
2/ 80% of the internet's IP traffic is video, TikTok will have 1.8B users by the end of this year, smart cameras are being installed in every industrial setting, and we're being ushered into a new generation of creative tools.
1
0
17
Building blocks all the way down. State of the art audio enhance by just combining a couple OSS models.
Pretty insane that open source models when combined together can reach production grade quality on audio enhancement. Enjoy the YouTube GOAT with high fidelity.
6
12
44
1
0
15
Open source models are getting really, really good. Kind of surprising to see how something like this could be built over a couple days.
Excited to show off a Sieve workflow the team worked on over the weekend: AI generated Twitter bots!
With a few open-source models, we built a bot that responds to your questions with a custom, realistic talking head avatar, all deployed on
@sievedata
.
2
3
24
2
0
15
These are great improvements. We just updated Describe on
@sievedata
to use this latest model when `visual_detail` is set to low. Almost on par with the other VLMs we're using that are heavier :) It get's the text on his shirt + more details of the scene!
New moondream release out today! Mainly focused on improved OCR and captioning. If you're using moondream for image captioning definitely worth checking this one out!
14
15
241
1
1
14
i’ll be honest, this started out as a toy demo project for
@sievedata
but seeing the cost, quality, speed trifecta here makes me think this would be genuinely be useful to so many people. i'm blown away by the quality something so simple brings. great work by
@gaurang_bharti
!
🙈 Introducing Describe Beta. The most descriptive audiovisual summaries for videos generated with AI.
- Process videos up to 12x faster than realtime
- Costs <$0.01 / min of video
- Combines visual and audial components
Try on your own videos now:
6
8
38
1
1
13
🌴We recently released a feature that embodies the way we think about AI at
@sievedata
: The Job Tree (1/X)
2
4
14
What could we do to make this better?
1. Index more videos (not just 5800)
2. Take into account video metadata, comments, etc.
What else?
Introducing Tasmania - a better YouTube video search engine.
Every result links to a specific moment in the video, so you can pinpoint the exact part of a video that made it surface in the results!
24
70
716
4
0
13
4/ We've been working closely with software startups redefining their products through video AI in private. Today we're opening up Sieve for any developer to try. Thank you to all our customers who've helped us reach this milestone.
1
0
13
Thank you all for the love today! A bunch of you sent us questions along with workflows you want to try building on top of
@sievedata
:) Can't wait to see those demos live and will probably share some cool ones with the broader community on our account.
1/ Today we’re excited to unveil Sieve (
@sievedata
), a cloud native platform for processing, searching, and running all sorts of AI models on video.
53
44
218
0
0
12
Gearing up for a launch next week! DM me if you're interesting in tinkering with video + AI stuff, and would like an early preview 🍪
1
0
13
we were inspired by
@pika_labs
“sound effect” feature so we made our own version. 𝐭𝐡𝐢𝐬 𝐯𝐢𝐝𝐞𝐨’𝐬 𝐬𝐨𝐮𝐧𝐝𝐬 𝐚𝐫𝐞 𝐜𝐨𝐦𝐩𝐥𝐞𝐭𝐞𝐥𝐲 𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐞𝐝 𝐮𝐬𝐢𝐧𝐠 𝐀𝐈.
1
3
13
man we generated this last night thinking
@sama
would still be at microsoft. bro keeps switching on us.
We’ve recently been working with lipsyncing models and wanted to share a message on behalf of Satya. The stakes couldn’t be higher.
Check out the links below to learn more and try it yourself for free!
3
3
26
1
0
12
6/ This is just the beginning, we can't wait to see what you all build. Just get started for free by going to !
0
0
11
Note: it's not your traditional search and isn't meant to find the most topical videos. Instead it helps you pinpoint an exact "thing" you might've been looking for within clips.
1
0
11
sharing is caring 💏
We’re excited to launch GPU sharing on Sieve. Split any of our GPU offerings into up to 8 components to run models more efficiently with higher GPU utilization 🫂
0
2
8
0
0
11
@alexandr_wang
why LLMs specifically? why not this trend of awesome joint, multimodal latent spaces that represent more than just text?
2
1
10
while i don’t come from a creative background, building
@sievedata
has let me to interact with so many people pushing the forefront of video creation with AI. it got me curious enough about the history that i wrote a blog post on it :) hope you enjoy!
0
2
11
Banger video
@MadeWithCapsule
and banger app madeWith
@gaurang_bharti
Introducing Highlights Beta for developers. Find highlights in long-form video automatically.
- Specify any search term (i.e. “most viral worthy”)
- Auto-generate titles for the clips
- Auto-score each clip based on relevance
Try on your own content now:
4
3
19
1
2
11
Really excited to unveil our first case study! It’s been amazing working with the
@ZightApp
team, especially Sachin and
@nsphin
, and really awesome to see them tell the story of using
@sievedata
for Zight’s AI capabilities.
2
1
11
We're hiring. Unfortunately we don't work from speedboats :( We work in glorious SF.
Live shipping from the speedboat, because 5G 🚤 ❤️ 🚤 ❤️ 🚤 ❤️
From the classic series:
19
2
247
0
0
9
VoiceCraft by
@PuyuanPeng
is probably the first open source TTS that's passable in zero shot voice cloning. here's
@fredagainagain1
's voice...cloned.
Announcing 𝐕𝐨𝐢𝐜𝐞𝐂𝐫𝐚𝐟𝐭🪄
SotA for both speech editing and zero-shot text-to-speech, Outperforming VALL-E, XTTS-v2, etc.
VoiceCraft works on in-the-wild data such as movies, random videos and podcasts
We fully open source it at
30
140
686
1
1
9
Awesome work by the team. Cool to see how much you can gain by speeding up ffmpeg ops + parallelizing over burst compute!
Introducing fast-asd: a new open-source method of doing active speaker detection on video that’s 90% faster than existing solutions.
1
3
13
0
0
10
We just updated audio enhance on Sieve. Now it uses some "non-AI stuff" like adaptive leveling and loudness normalization on top of some diffusion which to my ear sounds much better than a lot of audio enhance solutions i've seen. Simple solutions ftw.
0
1
9
@gorkemyurt
i don't think this is in any way specific to data / ai. PMs working at cloud providers just aren't good at building delightful developer products. whether it's ML infrastructure, web hosting, financial services, security, etc.
0
0
9
people who have used opencv, librosa, moviepy, etc for projects. anything you hate a lot about them? or anything you love so much about them?
0
0
8
it amazes me how many diamonds in the rough there are when it comes to open source models. here's a two year old model that does "active speaker detection" on video. pretty unoptimized code but it works great.
1
0
9
why is it so hard to create an azure account and why is it asking me to log in with linkedin??
1
0
8
something about apple notes on mac. i like to type in it more than i do in notion or anywhere else.
1
0
8
best video editing tool out there. sorry everyone else working on video editing tools you stand no chance.
2
0
7
this is amazing. there could not be a more exciting time to be working with computer vision :)
Releasing moondream2 - a small, open-source, vision language model designed to run efficiently on edge devices. Clocking in at 1.8B parameters, moondream requires less than 5GB of memory to run in 16 bit precision.
71
209
2K
0
0
8
dubbing is such a fun use case to do with ai. its such a clear example of ai models seeming more like building blocks versus the end experience / product.
1
0
7
sometimes you just know an email you've written is a masterpiece. super short, but super packed with some good shit.
0
0
8
can't wait for a future where custom legal work for startups is fully automated. let's have two bots going back and forth on contract redlines!
2
0
8
any good resources on the latest when it comes to GPU availability? really wish someone built a tool that gave more transparency into which regions + clouds have what sort of availability at a given moment. crazy that even cloud sales reps don't have full transparency into this.
2
0
7
the art of sales. never expected myself to appreciate it as equally as engineering.
0
0
7
Going model shopping on GitHub is pretty fun, especially when you see that the only inference code available is one that evaluates on the entire benchmark dataset
0
0
7
fwiw im a celtics fan and i like kyrie
In honor of a ☘️ sweep today, here's Kyrie telling Boston he plans to "re-sign" four years ago, now dubbed in Chinese.
Developers, get ready for next week. We've been cooking 👀
1
0
45
2
0
7
been trying to figure out how to say something like this for a few days and this does a pretty good job. chatbots or text prompts exist because that’s how the underlying models work. but is every user of the internet just typing into a terminal? no. so much more to be imagined.
Question-answering systems and chatbots have inadvertently created a false narrative that AI can solve any problem with a simple prompt, including complex creative tasks like storytelling and filmmaking. The ability to generate video from language has been misinterpreted as the
11
23
133
0
1
7
the best customers are those that offer super candid feedback, and trust you will do right by them. it's the kind of relationship that pushes product forward the most.
0
0
7
I read somewhere recently that only 40% of the MAUs on Figma are designers (other portion is likely PMs, engineers, etc). What are some other examples of products with similar usage characteristics (maybe in a different domain)?
2
1
6
so satisfying when something you think is so simple just wows a customer. makes you realize that 99.99% of the world doesn’t have any understanding of how half this stuff is even possible…
1
0
7
The OAI / MSFT relationship is weird in that they are competitively selling the same product?
I just got access to what I think is OpenAI's "limited access" voice engine within 1 day of request. MSFT calls it "personal voice".
1
0
6
Read more about how and why here:
1
0
4
everyday i wake up and think about how much i appreciate black boxes and the magic they let me wield
1
0
6
@mathemagic1an
@BookCameo
This is awesome! We made something similar and launched it yesterday that takes the Wav2Lip output to make it look even better.
1
0
6
@tdsone3
@modal_labs
@LatchBio
@SeqeraLabs
my impression is that there are more problems to solve than just the compute in bio. visualization, plotting, hyper specific optimizations around particular workflows, etc. curious to hear you take on those.
2
0
6
Scarlett Johansson voice over,
@3blue1brown
style math videos.
@gatekeep_labs
make it happen!
I am impressed by 4o's subtly flirtatious voice. To duplicate that I'd use
@sievedata
to add context to the situation. Poor kid learning geometry shouldn't have a flirty teacher, so the voice should adjust by including Sieve analysis. Market is infinite, specialize in the nuance.
0
0
1
1
1
6
Some underrated computer vision use-cases:
1. "camera in a box" solutions in factories, construction, agriculture, etc.
2. more video-first platforms like
@tiktok_us
+
@Whatnot
3. competitive analytics (sports + esports)
4. Analyzing the world of Zoom (i.e.
@Gong_io
,
@GoRewatch
)
2
1
6
@HipsterSmoothie
@DescriptApp
now lets get
@DescriptApp
to auto-edit those clips into the right aspect ratios!
1
0
6
who wants to see ilya sounding chill? run up the likes and i'll go find a clip :)
2
0
6
hot take: in the long run, there will be no AI company that makes it big by serving proprietary models as an API
Every AI company is freaking out rn
They're all trying to lock people with yearly plans as new tech is released every day
Would you accept the lock?
79
5
190
0
0
6
since when did all the best sf coffee shops disincentivize sitting there with a laptop by having no wifi or outlets? function over form damn.
2
0
5
Yes! Except I think video is interesting in that it's not just bounded strictly to the public web. So much utility exists in being able to process and search camera feeds, private logs generated by robot fleets, and more.
Just wait for
@sievedata
in October :)
Internet users consume way more video than text, yet the web's core infrastructure is still built around text.
What does the web of video look like? At a minimum, we need better search engines.
This is one of the most interesting domains to apply AI breakthroughs today.
55
123
1K
1
0
5
The pattern of combining models is here to stay.
"𝘛𝘩𝘦 𝘱𝘳𝘰𝘤𝘦𝘴𝘴 𝘪𝘴 𝘤𝘰𝘯𝘵𝘳𝘰𝘭𝘭𝘦𝘥 𝘣𝘺 𝘢𝘯 𝘓𝘓𝘔 𝘢𝘴 𝘵𝘩𝘦 𝘢𝘨𝘦𝘯𝘵, 𝘸𝘪𝘵𝘩 𝘵𝘩𝘦 𝘷𝘪𝘴𝘶𝘢𝘭 𝘭𝘢𝘯𝘨𝘶𝘢𝘨𝘦 𝘮𝘰𝘥𝘦𝘭 𝘢𝘯𝘥 𝘊𝘓𝘐𝘗 𝘴𝘦𝘳𝘷𝘪𝘯𝘨 𝘢𝘴 𝘵𝘰𝘰𝘭𝘴."
VideoAgent
Long-form Video Understanding with Large Language Model as Agent
Long-form video understanding represents a significant challenge within computer vision, demanding a model capable of reasoning over long multi-modal sequences. Motivated by the human cognitive
5
38
235
0
0
5
computer vision workloads are one example of something that's GPU bursty in the sense that there are many CPU operations sometimes intertwined with their GPU counterparts (i.e. reading frames, other transforms, etc). what other ai inference workloads are like this?
0
0
5
as one of the few real contenders to AWS, it would cool if cloudflare could acquire huggingface
0
0
5
Anyone that offers hosted ML deployment, meaning just saying "hey you give me a model, I give you a direct endpoint".
Is that not just a commodity? Open to hearing if folks feel otherwise.
1
0
5
yolo world v2 is everything i ever wanted when playing with object detection models back in 2016. that was the bound of my imagination. it seems trivial given all the other models we have out today but it is amazing that we have this.
0
0
5
Every AWS product line can be built 10x better.
A bunch companies that sell specialized compute and get really efficient at using cloud infra + building an objectively better product.
And when these products become big enough to roll their own compute in-house...rip AWS.
0
0
5
every software company really is just selling compute credits with their own value-add markup
1
0
5
The team at Kaiber is building something really special
Join us tomorrow at 4pm PST.
We reintroduce
@KaiberAI
and chart our path for this new chapter of creativity.
0
0
14
0
0
5
bullish
0
0
5
what most think to be tiny tiny corners of broader fields tend to be large lakes some folks spend their entire lives swimming in. pretty cool when you get to speak with said folks.
1
0
5
the video editing landscape is such an interesting one. literally every single player now has AI top of roadmap each with there own mini-take on what's important and how to do things.
this wasn't the case at the beginning of last year.
0
0
5
Really cool to see the gap closing between open and closed TTS models.
@elevenlabsio
is still the clear winner for production applications but curious which OSS models people are actually using in production if any.
1
0
5
7000 languages existing sounded like such cap i needed to google it
Toucan TTS: MIT licensed Text to Speech in 7000 languages! 🔥
The most multilingual open-source TTS model out there ⚡
Step 1: They built a text frontend that can turn text in any language from the ISO-639-3 list into language-agnostic articulatory features.
Step 2: Then,
15
227
957
1
0
5
anyone that would find really good video captions useful? something that accounts for both audio + visual components.
2
0
4
Literally creating a whole new slow on the cost / speed tradeoff versus the rest of the market.
1
0
4
many such cases
0
0
4
this is the man i wish existed back in high school. no disrespect to pyimagesearch but this guy is pyimagesearch rebooted.
0
0
4
you know facebook really has their content targeting on lock when ur cofounder sends u an instagram short like this
0
0
3
and it's only 100 lines of code!
used moondream on
@sievedata
and gpt4. probably could use phi-2 or gemma with a bit more prompting for this task tho which would make this fully OSS dependent + lightweight.
0
1
4