Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: VideoGist – Useful YouTube video summaries (videogist.co)
72 points by nliang86 10 months ago | hide | past | favorite | 46 comments
Hi all! I put together a website that summarizes youtube videos.

Enter any YouTube URL, and it will give you an overall summary, individual chapter summaries, along with key video frames.

Here are a few examples:

https://www.videogist.co/videos/tips-for-technical-startup-f...

https://www.videogist.co/videos/openai-devday-opening-keynot...

https://www.videogist.co/videos/cbs-evening-news-full-episod...

https://www.videogist.co/videos/the-best-easy-miso-salmon-re...

I'd love to hear feedback - if it's useful (or not!), what could be improved, etc. Thanks for taking a look!

p.s. it can take a few seconds for summary content to show up - I am working to speed it up




This is nice! One thing that bothers me is the web page's Title which is "VideoGist". I'd appreciate it if it could be named "VideoGist - <youtube video title>" or something like that.


Noted! Will get this out tonight.


This is absolutely fantastic, I love it!

I think I've found a (minor) bug, there seems to be an inaccuracy with the time stamps. I assume you're taking the chapter start and end times from the video themselves, but they don't align with the summary in this video:

https://www.videogist.co/videos/eevblog-102-diy-constant-cur...

For example, chapter 5, Heat Sink and Power Dissipation starts at 5:18 in your summary, but in the actual video it starts at 12:15.

EDIT: I'm just now seeing that your summary only has 7 chapters, but the YouTube video is segmented into 13 chapters, so it appears you're not using them from the video after all. Are you doing the segmenting with the LLM as well? How do you get timestamps from that? EDIT END

Anyway, thanks for this fantastic product, I was about to build something similar, but with a different focus:

Assume I've already seen a video and I want to look up a detail from that video, say how to do thermal calculations like in the example video, but I remember neither the video name nor the time stamp. I'm trying to build an app that generates embeddings for chunks / chapters of videos which I can then search semantically.

Do you have something similar planned? Because that's something I'd pay good money for.

Also, would you mind sharing a little bit about the tech stack? I'm assuming you're using yt-dlp to download videos with chapters and running whisper for a transcript, then something like gpt-3.5-turbo for summaries? Because that's how I'm doing it right now :D


I got an error with this video:

https://www.youtube.com/watch?v=Uf-kDNFMCD0

https://www.videogist.co/videos/panasonic-s-rapid-response-t...

Or maybe it's still pending? I don't really understand if I'm supposed to refresh the page or if it will update automatically when processing has finished, so I hit refresh and got this error instead of the loading page:

  We're sorry, but something went wrong.

  If you are the application owner check the logs for more information.


Apologies - I am hitting a rate limit on a third party service. Working on it now.


I (think) I was able to get a rate limit increase. Try again?


Kudos for the landing page, this is how AI products should be presented.

1) Direct access to the product

2) Multiple demo examples

I've seen so many launches here recently that have no demos, require sign up and basically gate all the usefulness of the product behind walls.


Feedback: Browser Extension would be much more useful in practice. I usually end up at a video, and THEN want the summary.


Great feedback - I will look into what it would take to make a browser extension client.


Thanks!


Picked one from my recent history: https://www.videogist.co/videos/10-things-you-didn-t-know-ab...

The text summary is very, very good. Kudos!

(As a nitpick, it has 'Wesan' instead of 'Wunderschön', but David himself doesn't pronounce it very clearly, so can't really blame the speech recognition there.)

The frame grabs next to the text appear to be sampled at random times, and not for their content. In the video, David presents screenshots and photos and what not. It happens to work out for Chapter 4, 6 and 7. In 5 of the other chapters it shows just the talking head. If a human had picked them, I think they'd show the most relevant imagery for the particular part of the video, and not the talking head.


Hilarious to see a Legion of Skanks video in the immediate history list.

But it was surprisingly useful. The screen grabs throughout the summary were especially nice.

I'd like it if I could jump to the specific timestamp a summary relates to on YouTube itself from the gen'ed text.

Thinking about it more though I'd almost rather this backfill breakpoints in podcasts I listen to and let me jump to them.

I also wouldn't worry about perf too much. I'd be ok with waiting on the order of minutes to get a quality improvement.

Final nit, the tone of the LLM being used sucks. Maybe try a different one or give the user options with open router?


I love this so much. I’ve got a backlog of videos to comb through and this might just be the solution. To me, immediate usefulness is a great informal indicator of a righteous hack. As a first text I threw a rather dense talk I gave to Legal Hackers DC chapter early last year at it and got a very solid set of summaries [1] broken down into chapters that I didn’t even realize existed but in retrospect totally agree with. And the most convenient part is that you support static links to the summaries so I can reshare it. One small question that hit me immediately is the copyright all rights reserved at the bottom of each summary. I’m not necessarily poking criticism of that but do wonder if it’s really what you want? Especially with the built in share feature it seems like at least some rights to reuse and such are expected. Anyway, overall this is stellar work, easy to use, and immediately helpful. Thank you!!

[1] here’s the summary I was describing: https://www.videogist.co/videos/generative-ai-for-law-1021


I meant “first test” not first text. Though it was the first text it output so that too, I guess.


Thank you! I will look into whether I actually want that - if it's not necessary I'll remove.


First I get an UNKNOWN ISSUER from Firefox as a security warning, if I bypass that I get.

Then my corp network reports

FortiGuard Intrusion Prevention - Access Blocked Web Page Blocked

You have tried to access a web page that is in violation of your Internet usage policy. Category Newly Observed Domain URL https://www.videogist.co/

:/


It might be due to the .co endpoint and your corporate policy? AFAICT the SSL cert, etc. is g2g - let me know if there's anything I missed though!


I'm looking for a good video summarizer! Sometimes a really interesting video pops up in my feed that I don't have time to watch but I want to skim a text version of it.

The ones I've tried have had a lot of issues. I just tried VideoGist on https://www.youtube.com/watch?v=tajhx6oTXnY and it printed an overall summary and the summaries of three "chapters" that it seems to have created (of lengths 18 seconds, 53 seconds, and 1:36), and now it seems to be hanging. It's an over 18 minute video. I don't know how long that's supposed to take, but I think being able to do the process relatively quickly is key, plus giving good feedback about the process (like a progress bar), and not being buggy/glitchy.


Apologies! I'm working on the reliability tonight. If you go back and try that video again, you'll see a full summary (I just regenerated it).


The summary text is descriptive and accurate (so far, need to test more). Thumbnails for each segment could be better--less repetitive, eyes open, etc. But overall as good or better than what I've seen from commercial video mam vendors over the last 20 years.


Very cool! Curious about the technical details of the LLM side. Can't imagine this is cheap to run.

Tiny suggestion: the timestamp in the generated text block could link to the corresponding timestamp in the video.


I have a similar (less polished) side project I threw together in a weekend [0][1]. I use ChatGPT's API and it costs ~0.5c / article generated. So IMO very reasonable.

[0] https://gitea.va.reichard.io/evan/VReader/src/branch/master/...

[1] https://vreader.va.reichard.io/


It depends. If it is using YouTube's transcript API, it shouldn't be that bad. The problem with that is that it tends to mess up when there are multiple people in the video.


Wouldn't the expensive part of this be the LLM summary, not the video transcription? I can do transcription pretty easily with my low end graphics card and a fast version of whisper.

I was wondering though, if this uses the YouTube transcript if available, and falls back to do the transcription itself if not.


Has it been HN-ed? WheN I tried to summarised a video I got "We're sorry, but something went wrong. If you are the application owner check the logs for more information."?


Apologies - I didn't anticipate hitting rate limits on a third party service so quickly. I was able to get them increased - try again?


Nice, was able to get a summary for my recording https://www.videogist.co/videos/scaling-beyond-microservices...

How are you defining chapters though?


Very cool. I've been using a similar approach to summarize videos on my website https://ray.run/videos/112-playwright-installation-2022 Doesn't cost much and getting great feedback about it.


Very cool. We have something similar at https://videotap.com, although a bit more focused on repurposing your video content into other mediums (written blogs, twitter threads, etc.)


That is an awesome use case!


The cooking one looks spot-on, is it using GPT4 and transcript or with Vision ?


This is great. I asked it to summarize my showreel (which I understand is not the kind of video you want to summarize) and the results are both surprisingly accurate and hilarious. Well done :)


Great work! I got a question though, what's the processing time for longer videos? Like lecture videos that usually are somewhere around 1h-2h? Can it handle longer vids reasonably fast?


Should be in the ballpark of 30s to 1 minute. You should start seeing earlier parts of the summary faster than that though (< 10s). I am also working on improving chapter summary quality for long videos - sometimes too much content gets rolled into a single chapter.


Great work. Can you tell us about how you built it? Is it open source?


There's plenty of websites like this one, but I really like this one. Saved in the bookmarks.

Good summary and great breakdown with timestamps, would be even better if they were clickable.


Could you name them? Because I'm looking for something just like that, ideally self-hosted / running locally.


Can we "download" a slide deck from this -- with chapter summary in slide notes :)


Every video I try I just get: Something went wrong


Is the service hugged to death? I've tried it on a couple of videos, and it's stuck on "Loading content" ...


Apologies - the app was hitting a rate limit on a third party API which I think I just resolved.


is there bookmarklet available?


I love it. I never tried any summary website but it's phenomenal. What technology do you use?


Nice!

Feedback - linking to the timestamped sections was the first thing I tried to do on your examples.


Do you mean from a video playback perspective, or something different?


I mean that if I find a summary section useful, I might want to watch the video clip around that point in time. So being able to click a link to go to YT to watch it at that point in time or even better, embed the video with that timestamp instead of the image, then I can just click play instead of see the image.

Thanks




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: