TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

TL;DW: Too Long; Didn't Watch Distill YouTube Videos to the Relevant Information

328 pointsby pkaeding3 months ago

53 comments

kragen3 months ago
What I&#x27;d really like is a service that edits down YouTube videos by removing all the stock footage and talking head crap, then speeding up the audio to fit over the remaining novel information—whether that&#x27;s new battlefield footage, electron micrographs, demonstrations of machining techniques, or just elephant toothpaste. The talking head filler seems like it would be easy to recognize, but stock footage recognition presumably would have a significant false negative rate, which is okay.<p>This would reduce some videos to just a transcription, which would be the ideal outcome, I think. The less of my limited time on Earth I waste watching some dumbshit reading a script at a camera, the better. Summarizing the transcript further like this site does might be occasionally useful, of course.
评论 #43056500 未加载
评论 #43068646 未加载
评论 #43057407 未加载
评论 #43068075 未加载
评论 #43057115 未加载
评论 #43064716 未加载
评论 #43071684 未加载
评论 #43062834 未加载
henry20233 months ago
What if we summarize all the information in the world into a few hundred volumes of human knowledge, then summarize those into a 10,000 pages book, then that into a 10 long form essays, then those into a 100,000 chars blog post, then that into a pamphlet and finally we summarize one more time into a single tweet.
评论 #43057742 未加载
评论 #43056989 未加载
评论 #43057008 未加载
评论 #43058328 未加载
评论 #43056516 未加载
评论 #43059823 未加载
评论 #43063702 未加载
评论 #43056713 未加载
评论 #43058095 未加载
评论 #43059810 未加载
评论 #43057211 未加载
评论 #43064565 未加载
stong13 months ago
Hi HN! I&#x27;m the author of this service. Thank you for your support.<p>There may have been some temporary downtime due to residential proxy running out of bandwidth. I have purchased additional bandwidth. (I run this service for free.)<p>There also may be some errors with particular videos because they are not accessible in certain regions. For now all requests to YouTube originate from United States, but open to change in the future to some kind of round-robin or fallback system.<p>I know it&#x27;s not perfect. I developed the tool originally for my own use. It&#x27;s open source and I&#x27;m open to any patches or pull requests.<p>Enjoy!
评论 #43062161 未加载
评论 #43062908 未加载
评论 #43064511 未加载
评论 #43062449 未加载
评论 #43060703 未加载
CM303 months ago
Tried it on 3 random videos I watched, and the results were... mostly good, albeit mixed.<p>On the one hand, it got my video about a Mario &amp; Luigi: Brothership glitch dead right, immediately listing where you&#x27;d need to die to get an item early and what you&#x27;d get out of it.<p>It also did an okay job summarising a Zelda dungeon analysis video by someone I&#x27;m subscribed to, with some info on why that dungeon was well-designed that clearly came from the video.<p>Unfortunately, it did a poor job at summarising a video about plagiarism in the YouTube speedrunning essay space, associating the problem with smaller creators rather than the person the video was about and leaving out far too many details to be useful.<p>This seems to confirm my assumptions about how an AI summariser would work in general; if the original media is a straightforward piece about one easily understandable topic, then it&#x27;ll do fine and work about as well as a human would. If it&#x27;s a longer piece with multiple points backed by various examples, then it&#x27;ll struggle to summarise it in a way that makes sense.
评论 #43058049 未加载
fullshark3 months ago
So what is everyone doing with all this free time they&#x27;ve now accumulated from not reading, watching, or listening to media?
评论 #43060348 未加载
评论 #43060561 未加载
评论 #43060576 未加载
zorgmonkey3 months ago
I tried a couple videos in both this site and Kagi&#x27;s summarizer, both were decent but each time Kagi did better.
评论 #43056057 未加载
gloosx3 months ago
Idea hackneyed since LLM&#x27;s appeared. Cool that implementation is open-source, though yt automatic captions are sometimes completely off-point, especially when people talking in the video don&#x27;t have a diction of a tv show host.<p>I wonder if an idea found it&#x27;s niche after all? Do you guys summarise you videos to short texts and that leaves you satisfied? For me video is video, I can relax, sit and watch&#x2F;listen to it. With text it is different, it is a mental exercise to read and process it, so turning video into text feels like an essential downgrade. I would prefer watching at 1.5&#x2F;2x speed instead of text summary if I want to finish it faster.
评论 #43056927 未加载
评论 #43060384 未加载
评论 #43060277 未加载
评论 #43056796 未加载
gus_massa3 months ago
I tried with a few Thunderf00t videos. He has good analysis, but the guys repeats everything too many times. Many are about silly impossible &quot;inventions&quot; &#x2F; scams, but this is an experiment that he published in Nature Chemistry:<p><a href="https:&#x2F;&#x2F;tldw.tube&#x2F;?v=LmlAYnFF_s8" rel="nofollow">https:&#x2F;&#x2F;tldw.tube&#x2F;?v=LmlAYnFF_s8</a> &quot;<i>High speed camera reveals why sodium explodes!</i> --&gt; &quot;<i>Coulombic explosion. (Sodium and water reaction)</i>&quot;
评论 #43056016 未加载
评论 #43057084 未加载
评论 #43056389 未加载
doright3 months ago
Seems to use gpt-4o under the hood. I wonder if using Deepseek would make any difference in quality.<p>I used to have nice and detailed summaries with this app lasting many paragraphs. It used to be totally free and you could submit as many links as you wanted. However they started forcing you to wait dozens of hours between summaries or pay for credits. I haven&#x27;t found a YouTube summarizer as high quality since.<p><a href="https:&#x2F;&#x2F;play.google.com&#x2F;store&#x2F;apps&#x2F;details?id=com.emote.youtube_summarizer">https:&#x2F;&#x2F;play.google.com&#x2F;store&#x2F;apps&#x2F;details?id=com.emote.yout...</a><p>I&#x27;m thinking one could replicate such a service pretty easily and be able to plug in your Deepseek API token instead. It would be convenient if such services let you &quot;bring your own API key&quot; so to speak, so you&#x27;d only have to manage one bank of credits. But it&#x27;s understandable that lots of people want in on a slice of the AI pie right now.
评论 #43058316 未加载
tikotus3 months ago
Something that could be interesting is processing all videos in some subgenre, like say &quot;game development&quot;, rank them by popularity, summarize them, and then analyze for patterns. This would be valuable information for anyone wanting to make a video with higher chances of being shown to people.
评论 #43056659 未加载
xivzgrev3 months ago
Pretty cool. The summary at first was helpful, but then delves into repeating itself. I tried it on John wick. Here’s the latter part of the summary.<p>The video calls out the cliché tropes and logical inconsistencies in &quot;John Wick,&quot; showing how they detract from the film&#x27;s emotional impact. The critique outlines the logical flaws in &quot;John Wick,&quot; focusing on cliché tropes that undermine the film’s emotional depth. The video critiques the logical inconsistencies and clichés in &quot;John Wick,&quot; highlighting how they reduce its emotional impact. The critique points out clichés and logical flaws in &quot;John Wick.&quot; The video notes clichés and flaws in &quot;John Wick.&quot; The video critiques &quot;John Wick.&quot;
ggasp3 months ago
I made a shortcut for iPad to do more or less the same thing and promised myself to keep “engineering” the prompt but work pretty well as it is.<p>Link: <a href="https:&#x2F;&#x2F;www.icloud.com&#x2F;shortcuts&#x2F;fbb5a315cb354fbf903bcfa4f409ba95" rel="nofollow">https:&#x2F;&#x2F;www.icloud.com&#x2F;shortcuts&#x2F;fbb5a315cb354fbf903bcfa4f40...</a><p>What the shortcut does is taking the transcript of the video and asking your ChatGPT app to extract main ideas, quotes, and facts.<p>The prompt comes originally from fabric CLI tool <a href="https:&#x2F;&#x2F;danielmiessler.com&#x2F;blog&#x2F;fabric-origin-story" rel="nofollow">https:&#x2F;&#x2F;danielmiessler.com&#x2F;blog&#x2F;fabric-origin-story</a>
energy1233 months ago
For specific questions I get the transcript from <a href="https:&#x2F;&#x2F;tactiq.io&#x2F;tools&#x2F;youtube-transcript" rel="nofollow">https:&#x2F;&#x2F;tactiq.io&#x2F;tools&#x2F;youtube-transcript</a> then copy paste to a LLM and append my question
albert_e3 months ago
Tried Andrej Karpathy&#x27;s latest video -- <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=7xTGNNLPyMI" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=7xTGNNLPyMI</a><p>It said &quot;too long video&quot;<p>Exactly, duh ;D
评论 #43066035 未加载
评论 #43067009 未加载
acrophiliac3 months ago
An error occurred: tuple indices must be integers or slices, not str<p><a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=Ks-_Mh1QhMc&amp;list=PLWDUzz3hCDJL313cxfWSnTh5X0ZtP1oVU&amp;index=3&amp;pp=iAQB" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=Ks-_Mh1QhMc&amp;list=PLWDUzz3hCD...</a>
评论 #43060018 未加载
Brajeshwar3 months ago
The sweet spot of most YouTube videos was supposedly under 10 minutes. What happened? These days, the typical YouTube videos are almost 30 minutes long. I&#x27;m fine with documentaries and the like. Is there a way to bring back the idea that under 10-minute videos work the best?
评论 #43059028 未加载
评论 #43057335 未加载
评论 #43057264 未加载
TNWin3 months ago
I tried it to summarize some lecture videos. And the summary ranged from average to bad. Nothing I couldn&#x27;t already get from the description. Even ChatGPT 4o spits out far better content.<p>So far my method has been to take the transcript and use an LLM with customized prompt for summary.
rapind3 months ago
This feels like the inevitable outcome of the youtube algorithm favouring longer videos.
评论 #43056401 未加载
评论 #43056190 未加载
prvc3 months ago
Getting a &quot;Too long video&quot; message in response to a query is frustrating to the user, and, well, redundant information, so contrary to the purpose of increasing information density.
tanewishly3 months ago
Interestingly, YouTube also lends itself for this sort of thing for movies. You can watch scenes from movies and get the gist, while only spending ~30min on a movie. It&#x27;s a great way to watch mediocre movies - they&#x27;re not so horrible (1), and you do get the entertainment value without being exposed to the shlocky-est parts.<p>(1) Eg, if you skip the beach scene, Terminator Dark Fate is... palatable! But yeah, the reasons for mediocrity typically still shine through a bit.
jisnsm3 months ago
I tried using it on a 2-hour video and it said that… the video is too long?<p>Kind of like hiring a painter to paint your house and him refusing the job because your house is unpainted.
评论 #43058079 未加载
评论 #43058771 未加载
crocowhile3 months ago
This is so fun!<p>Could you make it compatible with openrouter please? I see you use the openai binding that allows you to specify a different base_url so that would work out of the box and integrates well with the .env<p>``` from openai import OpenAI client = OpenAI( base_url=&quot;<a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;api&#x2F;v1" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;api&#x2F;v1</a>&quot;, api_key=&quot;&lt;OPENROUTER_API_KEY&gt;&quot;, ) ```
thdhhghgbhy3 months ago
Text summarisation already frequently misses things, sometimes gets them completely backwards. Can&#x27;t convince myself this will be any better.
jknoepfler3 months ago
I see the value in this, and thank you for posting it.<p>That said, it really makes me wonder at how insanely inefficient information transfer over the internet can become. This is like... doing OCR on screenshots of PDFs to send messages, rather than just passing around text. Nice, searchable, parseable, indexible, extremely compressible, editable, readily versionable text.
m1keil3 months ago
That&#x27;s pretty useful. I think yt itself is experimenting with similar feature as I see some videos have AI summaries in the official app.
评论 #43057233 未加载
winkeltripel3 months ago
It seems to have some reasonable length limitations. It refused to distill this epic analysis of dungeons across 3 games, since the video approaches 4 hours long: <a href="https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=PajArJbPfpE" rel="nofollow">https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=PajArJbPfpE</a>
nashashmi3 months ago
What is really cool about this is that my attention span is kept after reading the summary.<p>Like the following video: <a href="https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=JBkAKCJJuFQ" rel="nofollow">https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=JBkAKCJJuFQ</a><p>I am not getting emotionally invested to the point of losing the end focus of the speaker. Actually now I am looking forward to it.
singlepaynews3 months ago
My struggle with all these tools is that it’s been universally easier to type “summarize this video &lt;link&gt;” into chatGPT than to even keep track of the link to service. I desperately want a openAI-powered business to be viable, but it’s a snake eating its tail
nottorp3 months ago
Most videos would do well with a summary of the summary of the generated transcript. Should take 30 seconds to go through.
asah3 months ago
Haha maybe not.<p><a href="https:&#x2F;&#x2F;tldw.tube&#x2F;?v=A4kpVO56OBU" rel="nofollow">https:&#x2F;&#x2F;tldw.tube&#x2F;?v=A4kpVO56OBU</a>
JoeDaDude3 months ago
This is not the only service of its kind. There is also the Video Gist:<p><a href="https:&#x2F;&#x2F;www.videogist.co&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.videogist.co&#x2F;</a><p>I can&#x27;t speak to the differences between the two. Once casual observation is that the one posted by the OP seems to be a lot faster..
ryanmcbride3 months ago
I thought maybe I&#x27;d finally be able to through a Wendigoon video with this, since I&#x27;m generally interested in the things he talks about but can&#x27;t stand some of his linguistic tics. Unfortunately looks like most of his videos are too long for this, which I guess is ironic considering the name.<p>TL;DP
评论 #43070597 未加载
评论 #43056416 未加载
hrkname3 months ago
I use YouTube&#x27;s transcript feature or 2x speed for slow talkers.<p>Especially with political commentary many YouTubers stretch their content or talk slowly for more ad deliveries. It is annoying, because often there is an important five minute section hidden in a one hour video.
esafak3 months ago
Does it just parse the captions like every other service or does it analyze the audio too?
leovander3 months ago
&gt; An error occurred: tuple indices must be integers or slices, not str
评论 #43060030 未加载
renewiltord3 months ago
There&#x27;s also a funny one like <a href="https:&#x2F;&#x2F;tldw.tech&#x2F;" rel="nofollow">https:&#x2F;&#x2F;tldw.tech&#x2F;</a> which makes everything into a The Verge video.
ddoice3 months ago
Cool! I had thought about doing the same but in a Chrome extension that would use Llama 3.2 with WebGPU, but when I tested it, it was very slow and sometimes crashed the browser.
czk3 months ago
Got a chuckle from that recent commit message. I feel you
评论 #43056672 未加载
observer9873 months ago
Nicely done! Thanks for open sourcing it as I&#x27;ve been able to run it locally and experiment with the prompt.
Awtem3 months ago
I like the idea, but I get the message: &quot;Too long video&quot;... So I am wondering about its application if it only works for short videos
someothherguyy3 months ago
I would like to see one of these that produces an outlined summary then has timestamps as citations for the outline&#x27;s individual entries.
评论 #43057655 未加载
the__alchemist3 months ago
Will I still get to learn about Brilliant?!
garbagepatch3 months ago
It would be nice if it also gave the videos better titles. Then make it an extension to replace the titles automatically.
评论 #43056499 未加载
andai3 months ago
Change the prompt to &quot;in bullet points&quot; to increase the value by about 80%.<p>(And add &quot;by section&quot; to get to 100%!)
TheArcane3 months ago
Bit of a waste to host this locally and still need an openai api key. This would be great with Ollama
prashp3 months ago
How is this different from using Gemini to summarize a video given a YouTube URL?
bli9405053 months ago
I tried a video 1:42 hours long and it told me it was too long. Increase the limit?
atommclain3 months ago
Curious how well this compares to the same functionality built into Gemini.
Ancalagon3 months ago
Can’t wait till futurama style streaming-into-my-brain is available
dankobgd3 months ago
too long even this site cannot &quot;watch&quot; it
joshuaturner3 months ago
From some light testing, the results here are pretty lackluster. I&#x27;ve used videosummarizerai.com in the past via ChatGPT and the ability to specify the format of the information you&#x27;re looking for is a huge plus.<p>Here&#x27;s the summary of this video [<a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=W1WkG8WuRcg" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=W1WkG8WuRcg</a> (AoE2 video)] from tldw.tube:<p>&gt; The video discusses the most powerful team combinations in the &quot;10x Shared Civ Bonus&quot; mod for Age of Empires II. It highlights how certain civilizations, when combined strategically, can create overpowering unit types and tactics. The focus is on four-player team compositions that leverage unique benefits: for instance, the combination of Franks or Incas for free castles, Japanese and Tatars for rapid trebuchet fire, and Celts for enhanced onagers. The video also explores various combinations for cavalry and infantry, including the Goths for swift unit production and a range of secondary civilizations to amplify strengths. Commenters contributed ideas for clever combos, emphasizing the limitless potential for exploiting the mod&#x27;s mechanics to develop broken strategies that can quickly overwhelm opponents by accelerating unit production or enhancing existing attributes, thereby showcasing the humorous and chaotic nature of the mod&#x27;s gameplay.<p>and here&#x27;s the summary of the same video from videosummarizerai.com via ChatGPT with the prompt &quot;Give a summary of all civ combos mentioned in order&quot;:<p>&gt; Here’s a summary of the civilization combinations mentioned in the video in order:<p><pre><code> Castle Spam Team: Incas: Free castles and massive villager armor bonuses. Teutons: Adds +30 range to castles. Celts: Castles fire at four times the usual rate. Sicilians or Spanish: Increased castle build speed (castles built in 30 seconds). Ultimate Trebuchet Team: Japanese: Faster fire rate and instant packing&#x2F;unpacking for trebuchets. Tatars: Adds +20 range to trebuchets. Britons: Adds splash damage and perfect accuracy with Warwolf. Saracens or Celts: Additional attack bonuses or faster firing trebuchets. Best Onager Combo: Celts: Increased fire rate and durability for onagers. Mongols: Super-speed movement for onagers. Koreans: Adds +10 range and removes minimum range. Slavs: Makes onagers free. Scorpion Team: Khmer: Adds range and additional projectiles. Romans: Reduces cost and increases attack rate. Ethiopians: Adds torsion engine tech for better area damage. Chinese: Adds huge damage to projectiles for almost quadrupled effectiveness. [goes on to list remaining 4 team combos]</code></pre>
wopwops3 months ago
&quot;Too long video&quot; lol
computerthings3 months ago
I tried it on music videos, and with most of them I got the distinct impression it&#x27;s drawing from the description or external info, maybe PR, because it was very generic, the funniest I saw being <a href="https:&#x2F;&#x2F;tldw.tube&#x2F;?v=kxstMJY2Q28" rel="nofollow">https:&#x2F;&#x2F;tldw.tube&#x2F;?v=kxstMJY2Q28</a><p>I was about to give up because after all, this isn&#x27;t for music videos, tried something it could not <i>possibly</i> do well with, and I&#x27;m impressed, against my will.<p><a href="https:&#x2F;&#x2F;tldw.tube&#x2F;?v=7X62OJG6Whg" rel="nofollow">https:&#x2F;&#x2F;tldw.tube&#x2F;?v=7X62OJG6Whg</a><p>This isn&#x27;t just an acceptable summary, this is really good, considering how much could be missed (like it did with <a href="https:&#x2F;&#x2F;tldw.tube&#x2F;?v=6k7kk1lUnis" rel="nofollow">https:&#x2F;&#x2F;tldw.tube&#x2F;?v=6k7kk1lUnis</a> for example).