I… replied to a comment vs OP. Doh!

Ollama’s version is distilled with Qwen or Llama depending on parameter size, so it’s going to behave very different than the original model, since it is very different.

Very cool.

I have been playing with Omnigen (a new open source image-based AI) for unblur and exposure adjustment. Works pretty darn well, and have considered creating a plugin for it for GIMP if no one else does.

No, concede is to admit defeat, cede is to give up something. You could concede the battle, and as a result cede land to the victor. I meant cede.

In strict definition, true, but to those in the US cede has the connotation of giving up something after a lost battle. Just the way we’re taught.

Please update the title to the full title of the article. This is super misleading and makes it sound like the founder doesn’t want this.

True, but a 3 year old TV with original firmware would have been pre-adpocalypse. My never-connected LG boots pretty quick when it was last on an HDMI port before turning off.

Maybe give https://nowsci.com/only-sensor a shot? I built the guides/schematics/models for ESPHome devices as a learning experience for myself.

@SkyNTP@lemmy.ml I felt the same way. Now I just keep making new things for it, currently on garage door opener, blinds opener, and may even automate turning on my DIY solder fume extractor.

For now. https://www.yahoo.com/tech/nvidia-transitioning-open-source-gpu-152323688.html

Pretty sure they use Whisper, which is what FUTO Keyboard already uses on Android to keep it local to the phone.

I use Heliboard as a keyboard, then FUTO Voice connected to the mic button.

Would be interested to know of you can connect, enable, then block it’s access to the internet. That’s what I did with Wyze cams.

That was actually what made my decision. My drive is encrypted at rest, and E2E makes it much harder to do other things I may want to do with my library. For instance, I have a bunch of photography sessions imported as local libraries, so they’re just folders on disk I can modify outside of Immich as I want.

https://immich-app.github.io/immich-charts/

I’d put the client ML as a disadvantage.

Having the AI processes run on a beefy server is much nicer in my book, but of course can only be done without E2E encryption.

Be sure to try Immich, too. If you don’t need E2E it’s better in my book. Especially with local-AI powered search, which Ente doesn’t have.

What? Now you’re just plain wrong.

Immich is more feature rich, has self-hosted AI, is 3 years old, has 52K+ stars on GitHub vs Ente’s 16K (with 2 less years no less), almost 4 times as many forks, and has a public roadmap.

How is that “very early development”? Want to give you the benefit of the doubt, but this feels very “Ente insider” to me. Outside of E2E encryption, what specifically does Ente offer that Immich doesn’t?

Put this in the other comment, but why?

But it doesnt have all the local-AI image recognition features, does it? Being able to search for “black car interior” is as random as it is effective in Immich.

I see people suggest Ente, but Immich has been fantastic for me, and is funded by FUTO which means it’s less likely to enshittify.

In other words:

Try Immich!! It’s worth trying.

But it doesnt have all the local-AI image recognition features, does it? Being able to search for “black car interior” is as random as it is effective in Immich.

Pretty sure YouTube’s beat scientists to this.

I don’t know anyone in the US who feels this way.