I notice you’re not actually writing this message from lemmy but piefed.
Piefed has many features lemmy does not so ensure you choose the right fediverse software for your own needs.

Make sure you have some proper protection for the signup would be my main feedback, those that allow completely free registrations quickly get infested with spam bots and then their instance get denied federation by the other instances.

You will need to setup a reverse proxy that handles ssl and letsencrypt in front of the syncstorage-rs docker and make sure you don’t bind to http://localhost/ but https://example.com/.
Firefox clients refuse to connect to a sync storage that doesn’t use a signed certificate unless it is on localhost.

To use it on firefox mobile you will need to activate the developer mode by going to Settings -> About Firefox and press the logo multiple times (I think five?). Exit the settings menu and enter it again. Now you will see Sync Debug at the top of the settings menu and in that menu you can setup custom servers.

Not recommended as it will be a constant maintenance and upgrade cycle to keep it working with Firefox but you can actually setup your own Firefox server stack (not just sync, but profiles and account everything):
https://github.com/jackyzy823/fxa-selfhosting

I’ve heard a lot of good regarding linkwarden.

Here’s how piefed handles it using the browser.

While I agree that open-weight is better than nothing, it is not the same as open-source.
Having a truly open-source model would be wonderful and it would also allow us to verify what data it is trained on and whether it has the right to use that data.

The Open Source Initiative’s definition of a fully open-source AI model includes the training data as a key element. Some developers wonder, however, how feasible it is to distribute the enormous datasets required.

• https://www.pbs.org/newshour/science/whats-the-difference-between-closed-open‑source-and-open-weight-ai-a-researcher-explains

He says Stanford HAI has staked out a specific bar: alignment with the Linux Foundation’s Model Openness Framework at its top tier, the “Open Science” class – meaning the code, the training data (or a thoroughly documented, auditable account of it), and the tooling are all released, not just the weights, alongside a real way for outside researchers and communities to download, run, study, contribute to, and modify the work. Anything short of that, in his view, forecloses entire lines of research: You can’t, for example, study how a model’s training data shapes its cultural assumptions, audit it for how fairly it represents different populations, or test new architectures at the scale where genuinely new behavior shows up. You can only observe a finished system from the outside and guess at why it does what it does.

• https://hai.stanford.edu/news/open-weight-models-arent-enough-we-need-truly-open-source-ai-models-for-science-and-society

I’d say Element is the one to go for. A list of clients is available here:
https://matrix.org/ecosystem/clients/

I really wish the culture on lemmy would evolve to include a short summary of the content and a reason for sharing a link instead of just posting a link without context.

I’m fortunate that nearly all of my close friends own something I’ve made. But online? In comment sections? In all the loud posturing about the impending AI takeover?

I keep coming back to the same question:

When was the last time you bought something directly from an artist?

While I understand the sentiment I think focusing the buy to be directly from an artist is a bit weird.
Does me visiting the local theatre not count?
Does me buying a vinyl with music not count (none of my musician friends has the tools to make vinyls, that’s for sure)?
Does me buying an older oil painting from the local auction not count?
Does me buying a 4k UHD movie not count?
Do I not support an artist when I buy the poster they designed as a commission from the company that sold it? (Kirsten Tinds posters of edible plants and berries and so on sold through Koustrup & Co. are gorgeous: http://www.kirstentind.dk/Plakater_og_Kortmapper.html)

Gemma 4 12b unified as a q6 gguf fits with about 24k context inside your vram. koboldcpp+sillytavern works well on the software side.

Congrats! Then you have a guide for an easy reverse proxy right inside your router. :)
Try it out and see if that doesn’t fit you better than the nginx proxy manager.

The main point was that if your mini-pc running the proxy is also behind a windows docker desktop install it too will have incoming connections appear as 192.168.65.1.

A solution then is to switch to a linux docker host where you can get rid of that hidden NAT. There are many webuis for docker such as dockge and portainer on linux without having to use the terminal.

My preffered solution would be to switch to a router/firewall with reverse proxy support such as opnsense with haproxy and skip the nginx proxy manager completely.
Having the proxy configured in the same UI as vlan, port forwarding, nat rules and so on is nice. https://opnsenselab.com/posts/opnsense-haproxy-reverse-proxy/

Long shot but you might’ve solved the nginx x-forwarded header multiple times without seeing it if the nginx proxy is behind the same bridged docker desktop system setup on your minipc then the visitors IP will show as 192.168.65.1 to the nginx proxy too so you forward that same ip to jellyfin even when it works.
An easy check would be to spin up nginx as a vm instead of a nginx reverse proxy docker image and see if it works.

No real solution, but a lot of chatter regarding that:
https://github.com/NginxProxyManager/nginx-proxy-manager/issues/674

You can always contact techaro.lol and ask if they’re willing to setup a secondary domain for you.

Or you can use others endorsements of the product to give it weight.
"The Linux Foundation uses Anubis to protect git.kernel.org and lore.kernel.org and seem very satisfied with the result.
For the low cost of 50USD/Month we get a custom branded version of it.

• https://social.kernel.org/objects/1bb84190-f207-4adc-ac0e-8eb87a8a60f9"

I imagine Anubis looking like that is because professional settings really should sponsor when using open source.
https://anubis.techaro.lol/docs/admin/botstopper/
https://github.com/sponsors/Xe

https://docs.docker.com/desktop/features/networking/#how-containers-connect-to-the-internet
I think your discord friends are right.

Outbound traffic from the container is sent through Network Address Translation (NAT) using a virtual adapter (typically with an internal IP such as 192.168.65.3). You can view or change this with the Docker Desktop settings.