AmalgamationXL V2-X-Lightning2
Made a lightning version of V2-X-Tau, using the top lightning models and suffice to say I'm blown away by it myself.
For anyone who likes this model I highly recommend trying it out, I feel you will be very happy with the results. Happy Diffusing! :)
For best results I found
CFG - 1 - 2 (1.5 I found to be the best)
Steps - 4 - 8
Sampler - DPM++ SDE
Scheduler - Karras
AmalgamationXL V2-X-Tau
As you can expect I've merged the new JuggernautXL_X_RunDiffusion model into AmalgamationXL. Had done a few merges before hand as well which I hadn't uploaded hence the jump to V2.
AmalgamationXL V1.1-Tau-Lightning
Electrifying Speed and Quality
The AmalgamationXL V1.1-Tau-Lightning model represents the latest fusion of top SDXL models and the new SDXL-Lightning models. This iteration takes the robust foundation of AmalgamationXL and infuses it with the unparalleled agility of SDXL-Lightning models, resulting in a version that redefines the limits of image generation. V1.1-Tau-Lightning is designed to cater to creators seeking not just excellence in visual fidelity but also unprecedented speed in bringing their visions to life.
I appreciate if people want to test out the model's capabilities and help run testing
<3 Have at it!
I've made a SD 1.5 Amalgamation Checkpoint as well
https://civarchive.com/models/363285/amalgamation-sd15
AmalgamationXL V1.0-Mu
After many merges of some new amazing models, I've finally reached V1.0-Mu.
When I have time I will go through a more detailed breakdown of the merges that led to this process but for now enjoy this amazing model. Improved in all aspects of realism, coherency and prompt adherence.
AmalgamationXL V0.7 Theta: The Recursive Merge Evolution
AmalgamationXL-V0.7-Theta stands as the zenith of recursive model enhancement, a fusion of art and technology that has traversed several transformative phases. Each recursive merge has been a careful curation, integrating the strengths of specialized Stable Diffusion models to form a versatile and powerful tool for image generation.
AmalgamationXL-V0.7-Theta
This model is not just an iteration; it is a gateway to a multiverse of imagery. AmalgamationXL-V0.7-Theta is your artistic partner, capable of rendering images with a level of detail and realism that breathes life into your creative vision. It is a culmination of relentless innovation, standing as a versatile, robust, and adaptive model for any creator seeking to push the boundaries of image synthesis.
This model (V0.7-Theta) seems to have better overall cohesion, realism and prompt adherence.
Generation Guide:
Steps: 20-50
Sampler: DPM++ 2M SDE
Scheduler: Karras
If you like what I do, feel free to have a look at my ko-fi or patreon where I have a bunch of ComfyUI workflows and other Stable Diffusion related things.
I also offer custom AI services and custom model merging on my Patreon
Description
As you can expect I've merged the new JuggernautXL_X_RunDiffusion models (Both SFW and NSFW versions) into the model.
FAQ
Comments (13)
Is the newest version still Lightning, or is it better used as a regular SDXL checkpoint?
No it's a regular SDXL checkpoint, I will make a lightning version when I can find the time :)
@IndrasMirror excited for when it comes out! saw u make a comment about this on reddit and looks like just what ive been trying to find
@rime11 Thanks, hope you like it, I honestly haven't used another model since I made this merge haha definitely my go to. :) Save a ton of HDD space too :P
@IndrasMirror I tried it out and I absolutely loved it!! I was training my Loras on sdxl base and the likeness was horrible, I even tried training on just juggernaut and it didn't come out well so i decided to train it on this and the results were 100x better even with far less epochs! Definitely my new favorite
@rime11 Awesome, I haven't trained a Lora in a while and haven't been having great luck since my first SDXL ones, are you using Onetrainer or Koyha, I'd love if you could maybe share some setting with me, I want to get back into it :)
@IndrasMirror neither im broke lol, I use civit ai lora trainer. A tip from me is no matter how many images u have, always use 1 repeat and use the optimizer prodigy. Set epochs to 100 and choose the 100th one.
@rime11 Cool thanks, I tried one Lora recently on Civitai, wasn't too bad, I'll give your settings a shot :) Are you doing anything special for Dataset and tagging?
@IndrasMirror I use Gpt 4 image captioner from Leosam
Sorry, i am new here. I see everyone using Juggernaut XL and people keep coming up with merger of JuggernautXL claiming it's better.. I am confused if i shd use Juggernaut XL or this one. Pls enlightem me.
Its really up to you my man, this is a merger of ALL the top SDXL (in my opinion anyway, but definitely contains most of the top tiered ones), so it has features and like influences that extend well outside of JuggernautXL, I personally use it over any checkpoint at the moment, main reason I made it was so I didn't have to use different ones all the time.
@IndrasMirror Thanks for clarifying the doubt bro. I will try the model :)
If your planning on doing anything with the model for commercial purposes use Juggernaut, because according to the license which most people don't bother to read you can only generate with this model and do nothing with it.