How to enable hardware-Accelerated Video Decoding in Brave on Linux: Smoother Playback and Better Battery

Sun, 14 Dec 2025 01:35:15 +0000

Hardware-accelerated video decoding offloads playback from CPU to GPU, improving performance, reducing heat, and extending battery life—especially for high-resolution videos. On Linux, this works in Chromium-based browsers like Brave, but often requires flags. No more relying on h264ify extensions or downloading videos for MPV!

Benefits

Smoother high-res (1080p+) playback
Lower CPU usage
Better battery life on laptops

Prerequisites (Ubuntu-Based Distros, e.g., Mint)

For Intel GPUs (common on laptops):

sudo apt update
sudo apt install intel-media-va-driver-non-free vainfo

Verify with vainfo (should list supported profiles like VP9/H.264).

AMD/NVIDIA: Ensure Mesa/proprietary drivers are installed.

Enabling Flags Support on Ubuntu-Based Distros

Brave’s Debian package doesn’t read brave-flags.conf by default (unlike Arch’s AUR package). Create a wrapper:

Fix permissions:

sudo chown --reference=/usr/bin/brave-browser-stable.original /usr/bin/brave-browser-stable
sudo chmod --reference=/usr/bin/brave-browser-stable.original /usr/bin/brave-browser-stable

Create new launcher:Bash

sudo nano /usr/bin/brave-browser-stable

Paste:

#!/bin/sh
: $$ {XDG_CONFIG_HOME=" $${HOME}/.config"}
unset -v flags_conf
flags_conf="${XDG_CONFIG_HOME}/brave-flags.conf"
if [ -f "${flags_conf}" ]
then
 unset -v flags
 flags="$$ (sed 's/#.*//' < " $${flags_conf}" | tr '\n' ' ')"
 set -- $$ {flags} " $$@"
fi
exec /usr/bin/brave-browser-stable.original "$@"

Divert the original launcher:

sudo dpkg-divert --add --rename --divert /usr/bin/brave-browser-stable.original /usr/bin/brave-browser-stable

Now create/edit ~/.config/brave-flags.conf for flags.

Recommended Flags

Add to ~/.config/brave-flags.conf (one line, restart Brave):

Wayland (often default/best):

--enable-features=AcceleratedVideoDecodeLinuxGL,AcceleratedVideoDecodeLinuxZeroCopyGL,AcceleratedVideoEncoder

Xorg/X11 (or fallback):

--enable-features=VaapiVideoDecoder,VaapiIgnoreDriverChecks,Vulkan,DefaultANGLEVulkan,VulkanFromANGLE

Verification

Play a video (e.g., YouTube 1080p+).
Ctrl + Shift + I → Three dots → More tools → Media.
Check Decoder name: VaapiVideoDecoder = GPU accelerated (avoid FFmpegVideoDecoder).

Credits

This amazing commentary on github: https://github.com/brave/brave-browser/issues/2300#issuecomment-2718755680

As always, the Arch Wiki is an invaluable resource for all things Linux. If hardware video acceleration still doesn’t work, check the wiki directly for the latest flags and troubleshooting tips. https://wiki.archlinux.org/title/Chromium#Hardware_video_acceleration

How to use AI privately at Brave browser (2 methods)

Mon, 08 Dec 2025 23:15:59 +0000

Introduction

Brave Browser 1.69 introduced a game-changing feature: you can now connect Leo (Brave’s built-in AI assistant) to any model you want — including fully local models or third-party APIs. This means you get an always-available AI sidebar with zero subscription and total control over privacy and cost.

Here are the two best methods I’ve tested (one ultra-private, one smarter but cloud-based).

Method 1 – Maximum Privacy: Run a Local Model with Ollama (No GPU Needed)

You can run a surprisingly capable model completely offline, even on very modest hardware. The current sweet spot is Google’s Gemma 2 2B — it’s tiny (~1.4 GB), runs great on CPU, and works perfectly even with just 4–6 GB of RAM free.

Step 1: Install Ollama

Download and install Ollama from the official site: https://ollama.com (It has native packages for Windows, macOS, and Linux.)

Step 2: Download Gemma 2 2B

Open a terminal and run:

ollama pull gemma2:2b

(If you use Docker: docker exec -it ollama ollama pull gemma2:2b)

Step 3: Verify it’s running

Open http://localhost:11434 in your browser. You should see “Ollama is running” — that’s all you need.

Step 4: Add the model to Brave Leo

Open Brave → Settings → Leo
Click Add new model
Fill in the details exactly like this:

Label → anything you want (e.g., “Gemma 2 2B Local”)
Model Request Name → gemma2:2b
Server Endpoint → http://localhost:11434/v1/chat/completions
API Key → leave empty

(Optional but recommended) Set this model as your default for new chats.

Done! You now have a fully private, offline AI inside Brave that uses almost no resources.

Method 2 – Smarter Answers (Cloud): Use CryptoTalks.ai (Pay-as-you-go, No Subscription)

If you want access to the absolute best models (GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Flash, Llama 3.1 405B, etc.) without creating accounts at OpenAI/Anthropic/Google, CryptoTalks.ai is currently the best option. You pay only for what you use and can fund the account with Bitcoin or Lightning.

Step 1: Create an account & get your token

Go to https://cryptotalks.ai/signup → sign up → copy your API token (keep it safe!).

Step 2: Add a tiny amount of credit

Deposit any amount via Bitcoin or Lightning. Even $1–2 lasts a very long time for personal use.

Step 3: Add the model(s) to Brave Leo

Same process as before, just different values:

Label → e.g., “Claude 3.5 Sonnet”, “GPT-4o”, etc.
Model Request Name → exact model ID from their docs, examples:
- openai/chatgpt-4o-latest
- anthropic/claude-3.5-sonnet
- google/gemini-flash-1.5
- meta-llama/llama-3.1-405b-instruct
Server Endpoint → https://cryptotalks.ai/v1/chat/completions/
API Key → paste your token

You can add as many models as you want and switch between them instantly in the Leo sidebar.

Pro tip: Check current model rankings at https://artificialanalysis.ai/models to pick the best one for your needs.

Real-World Use

With Leo + your own model you can:

Summarize long articles or YouTube videos in one click
Explain complex code snippets
Draft emails or messages
Translate on the fly
All without ever leaving the browser and without sending data to big tech

When I’m on battery and want zero extra power draw → I switch to the local Gemma 2 2B. When I need maximum intelligence → I switch to Claude 3.5 Sonnet or GPT-4o via CryptoTalks.

Conclusion

Brave just turned every browser into a private, customizable AI workstation. Pick Method 1 for 100% privacy and zero cost, or Method 2 when you want the absolute best answers available today.

Either way — welcome to the future of browsing.

Thanks for reading! 😊

Brave on Aquasp's blog