cloudflare-r2 on S Anand

Editing Workshop Videos

Sun, 14 Jun 2026 08:27:16 +0530

I sometimes use Google Meet, Teams, Zoom, etc. to record workshops and talks. These record the entire session, including before and after the actual talk, and save it as large MP4 files.

I use ffmpeg to trim the video to just the talk, and then compress it for sharing. I’m sharing the options that work for me, discovered by trial-and-error.

To trim it, I use the following command:

ffmpeg -ss 00:10:00 -to 02:10:00 \
  -i "original.mp4" \
  -map 0 \
  -c copy \
  -avoid_negative_ts make_zero \
  -movflags +faststart \
  new.mp4

Arguments:

-ss 00:10:00: Start reading from 10:00 into the source file. Placing -ss before -i makes this fast because ffmpeg seeks instead of decoding from the start.
-to 02:10:00: Stop at 02:10:00 in the source timeline. So this keeps roughly the section from 00:10:00 to 02:10:00.
-map 0: Keep all streams from the input: video, audio, subtitles, metadata streams, etc. Use -map 0:v:0 -map 0:a:0 for just the first video and audio streams.
-c copy: Do not re-encode. Just copy the existing encoded streams. This is very fast and preserves original quality, but cuts only cleanly near keyframes.
-avoid_negative_ts make_zero: Rewrite timestamps so the output starts cleanly near zero. This helps avoid weird negative timestamps and duration/reporting issues in some players.
-movflags +faststart: Move MP4 metadata to the beginning of the file. This helps playback start faster in browsers and streaming contexts.)

To compress it I use:

ffmpeg -i 'trimmed.mp4' \
  -map 0:v:0 -map 0:a:0 \
  -c:v libvpx-vp9 \
  -b:v 0 \
  -crf 42 \
  -deadline realtime \
  -cpu-used 8 \
  -row-mt 1 \
  -tile-columns 2 \
  -frame-parallel 1 \
  -threads 8 \
  -pix_fmt yuv420p \
  -vf "fps=16" \
  -c:a libopus \
  -b:a 32k \
  -ac 1 \
  -ar 48000 \
  -af "highpass=f=80,lowpass=f=12000" \
  "compressed.webm"

Arguments:

-map 0:v:0 -map 0:a:0: Use the first video stream and first audio stream from the input
-c:v libvpx-vp9: Compress video with VP9 codec
-b:v 0: Use constant-quality mode instead of targeting a fixed bitrate
-crf 42: When testing on samples, CRF 42 was small enough and clear enough for me
-deadline realtime: -deadline good is too slow for me though that’s better for production
-cpu-used 8: 8 is the maximum allowed for VP9 for now
-row-mt 1: Enable row-based multithreading for faster VP9 encoding
-tile-columns 2: Split each frame into tiles so multiple threads can encode in parallel
-frame-parallel 1: Allow frames to be decoded in parallel; useful for VP9/WebM playback
-threads 8: Use up to 8 CPU threads during encoding
-pix_fmt yuv420p: Use the most compatible pixel format for browsers and video players
-vf "fps=16": Reduce frame rate to 16 fps to shrink file size; good enough for slides/talk videos
-c:a libopus: Compress audio with Opus, the standard audio codec for WebM
-b:a 32k: Use low audio bitrate; enough for speech
-ac 1: Convert audio to mono; good for voice and halves stereo audio data
-ar 48000: Use 48 kHz, the standard sample rate for Opus
-af "highpass=f=80,lowpass=f=12000": Remove very low rumble and very high noise; keeps speech frequencies

I deploy these on CloudFlare R2 on a custom domain: media.s-anand.net. R2 costs 1.5c / GB-month. Storing 10 years’ worth of ~500 talks of ~0.5 GB each is under $4 / month - quite affordable. (If it increases, I can shift. Statis files are easy to move.)

I serve them via a <video> tag. Here is an example video:

<video preload="metadata" width="1920" height="1080" controls style="max-width: 100%; height: auto;">
  <source src="https://media.s-anand.net/2026-05-23-ai-unboxed-context-engineering.webm" type="video/webm; codecs=&quot;vp9, opus&quot;">
</video>

Tracking redirects minimally

Thu, 07 May 2026 16:28:22 +0800

Everyone needs a tracking URL shortener.

Why tracking? I want to know if they opened my email and clicked the link.
Why shortener? I want them to know what the link is about. For example, https://r.s-anand.net/edge-remote-debugging.html is so much more meaningful than https://chatgpt.com/share/68528565-0d34-800c-b9ec-6dccca01c24c

I’ve used redirection services in the past - like t.co, bit.ly, goo.gl, ow.ly, and others. They tend to vanish, start charging, serve ads, etc.

Here’s my solution: use static HTML for redirection.

For example, https://r.s-anand.net/example.html redirects to https://example.com/. Here’s the code:

<script data-goatcounter="https://sanand0.goatcounter.com/count" async src="//gc.zgo.at/count.js"></script>
<meta http-equiv="refresh" content="0; url=https://example.com/">

The first line sets up tracking with GoatCounter, which is my current favorite analytics provider. It might vanish, but I can export the data if required and move to another. While it lasts, I can check how often the link was used.

I can create personalized links. For example, https://r.s-anand.net/example.html?ref=user1 tags user1. I can create unique links for each recipient.

The second redirects to the target page immediately.

I have deployed the files on a CloudFlare R2 bucket with a custom domain, which is practically free and fast. But I can switch at any point to any other hosting provider (e.g. GitHub Pages, etc.) by just copying the files and changing the DNS settings.

Advantages: No hosting, no fees, no maintainance, no vendor lock-in.

Things I Learned - 17 Nov 2024

Sun, 17 Nov 2024 00:00:00 +0000

This week, I learned:

Anthropic has single-plage docs for LLMs. Condensed version and Full version
Malcolm Gladwell on the importance of self-correction
- Belonging to multiple social worlds is a good way to defend against no longer being good at what you used to be. Diverse values and social groups help.
- Self handicapping explains a lot about the world. You study late for a maths test - so you can fail for lack of trying, not aptitude. Ecosystems (e.g. sports teams) mitigate self-handicapping.
- You don’t have to be good in athletics to get the benefits. A slow runner gets the same discipline, pumping up, etc that a fast runner does
- Mono cultures are good to accomplish a known mission. Diversity is good to pivot during uncertainty. So, localize mono cultures
- Diversity helps only if there are sufficient numbers, or if they have enough power to change the organization’s thinking.
Use a standardized password strategy, e.g. use the month like GramNov2024 (via Namit)
Gemini has an OpenAI compatible API. Gemini Docs
Ethan Mollick says Claude is solving MBA case studies well. x.com
LLMs pay a lot of attention to the first 6 tokens. Ref
This is an interesting article on “UI in the age of Gen AI”. Ref
Google Open sourced Alphafold 3. Repo
Cloudflare R2 has the same API as S3 but is cheaper
Prefect.io is a good alternative to Airflow / cron. Can use for synchronisation tasks, e.g. Drive to server. But no Auth, UI params or config.
Gemini transcription does not give accurate timestamps. Whisper does. But the quality of transcription is similar.
Pass a complex data structure to Claude.ai and have it create an app to visualize it. It does well. Simin Willison
Tech Council Ventures and Sunicon VC invest in early stage startups, and aloso provide them technology support (via Naveen)