HD Horizon at a Glance:
Welcome to HD Horizon, where image resolution is reimagined. This tool is my personal venture into enhancing how we experience SD Inferencing, making every pixel count.
What Sets It Apart:
HD Horizon specializes in multi-resolution inferencing, smoothly transitioning from 256x to an impressive 1536x. Its standout feature? The ability to achieve high-resolution inferencing natively, without the need for additional fixes like HRFix. This means more natural, sharper images at any size.
Why It's a Game-Changer:
For artists, designers, and anyone passionate about digital imagery, HD Horizon offers the freedom to create without resolution restrictions. Your visions can now be as detailed on screen as they are in your imagination - crisp, clear, and true to your original idea.
Your Companion in Creative Exploration:
HD Horizon invites you to explore the frontiers of resolution. It's not just about seeing the bigger picture; it's about seeing every tiny detail that makes up that picture. Join me in this journey of discovery and let your creativity reach new heights.
Welcome to "The Resolution Frontier" - where clarity meets creativity.
The Project, and Where It Goes From Here:
This is a first attempt proof of concept.
If you'd like to help build this out, I can certainly use assistance with image sourcing. I require a very large, very diverse set of images across as many concepts and styles as possible. Images should be 12MP+.
To contribute please message in the GreyModel Discord.
Description
This is a first attempt proof of concept.
If you'd like to help build this out, I can certainly use assistance with image sourcing. I require a very large, very diverse set of images across as many concepts and styles as possible. Images should be 12MP+.
To contribute please message in the GreyModel Discord.
FAQ
Comments (22)
I'm pretty shocked you only have one review (mine) on here. This really is a pretty big deal for those of us who like to make detailed images and be able to blow them up to 40"x60" prints. Well, for what it's worth, I appreciate the work you put into this. I've been using it most of the night since I downloaded it and have been pretty impressed.
Well, glad you're enjoying it. Truth be told I was mostly making this for my own project, but made sense to share. It's far from perfect... but that's what 2.0 is for, learning from mistakes and improving :D
That's a nice marketing blurb (ChatGPT?), but what does it do, and how?
It was indeed ChatGPT, but it does what it says... allows native (no HRFix) inferencing at multiple resolution targets from 256 through 1536... and how is... by way of Lora? I trained a lora at said resolutions in a way to try and minimize the influence beyond the desired goals.
Please stop using ChatGPT to write this stuff, or at least edit it to remove all the regurgitations. I can't take any more "frontiers" or "companions" or "explorations" or "journeys." Ugh.
I'm using this Lora on MageSpace and I'm noticing a marked improvement in my generations without evening needing to use Hiresfix. Great work! 🙌
How are you able to use this on mage?
@Picky_Pickles I have Pro+ on Mage and you can import any 1.5 or SDXL Model, TI or LoRA. I have imported over 100 things so far and loving it.
how do we use it exactly ? what do you mean we don't need a hires fix ? Do I just generate as usual a 512x768 image and it would look better with this or is there something to be done with resolution also ?
You simply use the lora, and then choose the resolution you wish. You may be able to use significantly higher resolutions.
@GreyModeler you mean my memory errors I was getting before would be gone ?
@patientx737 not if you have a low vram gpu, generating a 1024*1024 img will take more time than a 512*512 as always, but you can use this tool to actually generate at 1024*1024 natively with SD 1.5, which was not optimal before. This will use more resources than generating a lower image obviously
for some reason this doesnt work for me AT ALL, i keep getting blurred out images just like the most recent pic in the gallery
Are you trying on more than one model? As I've mentioned several places, this is not equally compatible on all models.
@GreyModeler oh really, i didnt know that. ok which models can it work on?
@discodiffuse Apologies, didn't see this. It's trained on base, so that, but there's no way I'm aware of to know which models work best. I've tested with good results on GreyModel, and a few other s that worked well. I believe my samples link a couple of the models.
Was there a v2.0 shortly ago? Actually, I got an alert that v0.9 was a new version....so I'm lost. Is v1.0 older than 0.9?
Apologies, there was no new version (but it is slated as my next training project) -- what must have triggered that was publishing a draft version (0.9) that had inadvertently not been published, and had several user generations attached to it.
@GreyModeler Ah, gotcha! Well, on the plus side it reminded me this exists so I started using v1 and it's pretty amazing how much it can change an image for the better. I'm on mage so there's no size changing for me, but it absolutely ups the image quality and prompt adherence. No idea how you even made this, but well done!
In case anyone else was wondering: I tested a bunch, and my conclusion is that this Lora and:
are wholly complimentary to / with each other, I get consistently better results running both simultaneously (at the same strength, usually around 0.4) than I do with either one of them on their own.
thx for the tip
real game changer