Introducing: MHordekAI, an AI image generation app, and more

Hey everyone!

I don’t know about you but all the channels appear to be particularly noisy this week.

And AI being the thing of the hour, I couldn’t resist making a little app in that space.

MHordekAI, an GUI frontend to AI Horde

Now it’s nothing terribly special or necessary, but I find it fun.

AI Horde is (in their words) a crowdsourced distributed cluster of Image generation workers and text generation workers..

So what does MHorkekai do? Basically it submits requests to their API, waits for the resulting job to finish and shows the results. In other words, you can have some fun creating images by textual prompt.

If you try it, be aware that results will vary greatly depending on options you choose, and especially which Model(s).
None of this is documented in the app, nor will it be, but there are plenty of guides out there on how to achieve good results.

Also be aware that downloaded images are always in WebP format, and Gallery/Tracker doesn’t support that (depending on SFOS release). This means these images may not appear in Gallery until you convert them.

Supported Features: (at the time of this post)

  • Submitting image generation requests (prompts)
  • Tracking request and generation status
  • Viewing and downloading generated images
  • doing all of that using either an anonymous, or a personal API key
  • … and some more

Watch the Releases for updates

Planned Features (no promises):

  • Supporting Alchemy (Interrogation)
  • Supporting Scribe (Text generation)
  • automatic download and conversion of results.

Watch the Milestones to track progress.


Plenty, probably.

  • Some are in the space of WebP support, which stock SFOS python does not have. Currently I work around this by calling ffmpeg, but that could be flakey.
  • UI is not great, rather functional (I hope).

You are invited to find and file more.

If you have problems with the Chum-published app, please try the one from Chum Testing.
If you have problems with the Chum Testing-published app, please try the one from Chum.
If both of the above fail, maybe try master from the repo.


Pull requests of any kind are always welcome.

  • Packages: Alpha-grade packages are available from Chum and Chum Testing. View it on Chum Web
  • Source Code: at Codeberg. Languages used: very bad QML, very bad Javascript, terrible Python


@mal may have read your post : Comparing master...jb62112 · sailfishos/python3-imaging · GitHub and is working on upgrading Python imaging package, enabling WebP in the process. Thanks @mal.

PS : thanks also @nephros for this interesting interface.


Yea I mentioned it on IRC also yesterday. Nice!

I guess I shall look into installing OS-version specific helper scripts once that PR goes through but that’s no biggie.

1 Like

Well, this sounds interesting! Thanks!

This app is very interesting. I think nothing can replace working on a desktop for this purpose but nevertheless is nice to be able to make some testing on the phone.
Thanks for sharing!

Looks like qt5-qtdeclarative-plugin-qmlstatemachine is missing from Requires, attached page won’t load without it

1 Like

Ah, thanks! Will add that.

1 Like

I have decided to publish v0.3, Beta 2, which adds a lot of things.


  • “Horde or Not?” - rate images on the Horde to gain Kudos. IMO, this is fun.
  • Support for adding “prompt styles”, i.e. preconfigured prompts and models to Dream
  • Basic support for Alchemy, image interrogation
  • Very Basic support for Scribe, text generation
  • Swedish translation via @eson, thanks! (And sorry this is incomplete for now as I added tons of stuff on top of your v0.2.11 submission.)

More details in the Release Changelog: "Fuck it, lets go public" Beta Release 2 - nephros/harbour-aihorde -

Available on Chum:Testing, coming soonish to Chum.


No problem, I’ll update some rainy day. :wink:

1 Like

v0.3 Beta3

First working version of QR Code generation. We are probably the first/only app to pick up this feature (announced 2024-05-20) apart from db0’s own Lucid Creations,

Note that getting this to produce anything is hard. Don’t get frustrated.
Note that when creation fails, the creation job will show up as “Done” with zero images, not as failed. This is due to #61

This is probably the last Beta Release for 0.3, coming up: a proper Scribe client which behaves like a chat app. By which we will call this 0.4 or 1.0 or something. The curious are invited to look at the devel branch.

No screenshots for this release, but a teaser for the next:


There is something fishy with Sailfish and QR codes…
The ones on link above for instance. The witchy one is okay, but the icy one doesn’t work at all. Probably something colour related.
I know it’s of topic, but anyway… :wink:

You have to say though the icy one is very distorted. The brainy one does work well too. (Tried with CodeReader).

Yes, brain and witch works well, with CodeReader and stock camera both.

Thanks a lot for this app!

I once tried it and it worked. Now, when I want to start a job for creating an image, I always get a banner saying “Input payload validation failed” whenever I try any other prompt than the predefined one with the robots.
Am I doing something wrong? I get this no matter whether I have set a API-key or not.

1 Like

How do you get backpage when rating? Just updated to see if I got a broken build but:

Once I get into rating there is no coming back for me???
Same when choosing any model, no way back to switch to another, but maybe that’s by design?


No, going back is just not implemented.

But you can open multiple instances of the app.

The idea being you can do Dream and Scribe in parallel in two windows for example. (Also it makes it easier to handle things like the request queue internally.)


Hm, not sure why that would be.

Did you change any other things apart from the prompt?

Oh and I think you need to use Enter for the prompt to be applied really, just switching focus away after clearing it might have the effect that an empty prompt is being sent.

1 Like

Oh sorry I meant to reply to this post that shows rating interface with a dot in top left: Introducing: MHordekAI, an AI image generation app, and more - #9 by nephros but maybe it’s post generation, thanks, thought my install was somehow buggy, switching the three animated tiles in opening screen to just stack instead of replacing pagestack should be fairly ez, will have a look, thanks

You can remove models using the little X icon near the model button.

The model choosing dialog indeed just adds models to the list, and doesn’t support deselection well.