reGen v17, New Alchemy Features come to the horde

Tazlin

Tazlin

— 3 min read
Share

I have been hard at work for quite a while, across many of the horde projects. Much of this time has been spent in the weeds, working on setting the foundation for more visible components. For example, I had resolved myself to clean up and improve the horde-sdk, as I saw that library as the best home for many of the previously-duplicated (or non-existent) functionality on both the client and server side of the horde. I also spent considerable time redesigning horde-model-reference to serve as a more useful store of the horde's model meta data. It is now also now what runs https://models.aihorde.net/, which will become the canonical source for horde model information, over the github.

The worker changes are currently in beta but the new models (qwen, zimage) and other listed features are available to users today. If you would like to join us in beta testing the worker changes noted here, join us in discord.

Worker improvements

In the past month or two, I have been able to turn my attention to horde-worker-reGen. Those of you active in the discord server may be aware that I have been running a beta of the newest worker version. That newest version includes, on top of many other fixes and features, two important features: First, a one click .exe installers and a one-line installer (for server/terminal installs). Secondly, a new textual-based UI, which works both in browser (the default launch mode) and in terminal.
I have had a long standing belief that users often considered installing the worker as "too technical", due to the variety of - admittedly complicated - steps required to install it in prior incarnation. My hope is that any one willing to run a worker will feel comfortable enough with the new experience, which is meant to reduce the amount of highly-specific knowledge about python or command line operations.

Alchemy returns

Another feature, long overdue, making a return to the standard image worker is Alchemy. This is supported as a stand-alone mode, including with CPU-only if you have no suitable GPU, or can be run concurrently with image generation. Just be sure to set a name for both worker types. I do not recommend running captioning concurrently with image generation right now unless you have a powerful setup.

I have also added a basic alchemy page to the aihorde.net homepage as a simple demonstration of the horde capabilities.

New Alchemy Modes

Along side the addition of alchemy, I am introducing support for several new upscalers, a couple of newer generation FaceFixers, and 4 new alchemy forms to showcase its potential. Try them out at https://aihorde.net/alchemy.

  • New Upscalers: 4xNomos8kSC, 4xLSDIRplus, 4xNomosWebPhoto_RealPLKSR, 4xNomos2_realplksr_dysample, 4xNomos2_hq_dat2, 2xModernSpanimationV1
  • New FaceFixers: GFPGANv1.3 and RestoreFormer (v1)
  • New Forms:
    • vectorize is a new alchemy form. Using a pure python (cpu driven) library, the alchemist converts a raster into an SVG ('vector') format.
    • describe creates a BlurHash for a given image; a short string which uniquely represents a placeholder for the image, as text.
    • palette generates the predominate color palette used in the given image, as a list of hex codes.
    • aesthetic scores the image as 0-10 based on LAOINs aesthetic scorer.
      • Image generations now also return a score as a GenMetadata entry for all images. This is possible because the safety process already calculates the image "features" (a tensor) required for this to work. The actual scoring is a tiny model that adds a trivial amount of processing time compared to what is already being done with respect to safety.

We will continue to evaluate what other alchemy forms and models that we can add.

Horde Status Page

I have also introduced a status page for the horde services. Those of you in the discord know that we have service alerts in a notifications channel, but for a quicker glance at the service health, we now have a dedicated web page.

In closing

As usual, I would like to emphasize that our mission is to provide generative AI and other frontier technologies to the public, free of charge. We can only do this through volunteer effort and donations. At the time of writing, we are in need of image and alchemy workers.

We understand that there is effort and real costs associated with running a worker and we fully appreciate everyone who does run one. If you have not looked into it lately, we think that its never been easier to run a worker. If you ever have issues, join us on our discord community or post an issue in the correct GitHub repository.

You can also make a financial contribution at https://haidra.net/donate/.

Read more