Did you miss me? Well I'm back..
And with me I bring you the newest tool and toy to make all your darkest fantasies come true..
MYTHWEAVER SDXL v1.0!
Mythweaver is a dynamically trained model designed to help you craft mesmerizing fantasy artwork and designs, enabling you to bring every world and character from your imagination to life.
Create epic DnD character designs or intricate NPC portraits for your players, and then design the fantasy realms for them to explore. Mythweaver seamlessly bridges the gap between different art styles, whether itβs classical graphic novel paintings, comic-style illustrations, or stunning photorealistic depictions. It comprehends hundreds of artists and art styles, offering unparalleled flexibility.
Whether you're visualizing a high-fantasy world or designing unique characters for your tabletop adventure, Mythweaver delivers rich, immersive images that not only captivate the eye but cater to all your fantasy art needs.
The model works well with Loras, and is surprisingly dynamic, detailed, and variable.
It's also consistent when you need it to be too.
All the images are 1:1 and should be able to be produced with A1111 and Forge without any issues.
If you have any questions, feel free to ask and I will do whatever I can to aid.
BEST SETTINGS:
SAMPLING: DP++2M KARRAS
STEPS: 20-60 (AVERAGE 20-30)
HIRESFIX ON/LATENT NEAREST EXACT/STEPS 20-50/DENOISE 0.45-0.6/UPSCALE 1.5
WIDTH: 800 (720-840)
HEIGHT: 1200 (1024-1280)
CFG: 4 - 7.5
This models examples were produced with Clip Skip 2 but Clip Skip 1 works in some cases as well, so experiment and try both!
This may or may not require SDXL-VAE-FP16-FIX You can find it here: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main
-ENJOY!
Description
V1.0 Clip Skip 2
FAQ
Comments (5)
Having tried multiple SDXL checkpoints, this one is still one of the best to use as a base for style Loras. Most others randomly tend to drop a style on some seeds, no idea why, but this checkpoint works perfectly on every type of style Lora. Would recommend.
I created it with great care and tested it on everything I could get my hands on. I mean the better majority of the gallery here is my renders. I would say the results speak for themselves.
Even stock.. without anything it's a very concise and variable, mature model. If you give it good prompts it.. it will generate good results. Loras just add everything else you could think of to give it focus. ;)
Thanks alot for your review and please add some of your images to the gallery here I want to see what you and others make with it <3
I had issues with both the vanilla VAE and the fixed VAE. Both would cause the same issue - desaturated/washed-out images full of white splotches. Switching to the SharpSpectrumVAEXL fixed it. Both the FP16 and FP8 versions work fine:
https://civitai.com/models/455138?modelVersionId=506692
Fixed vae located here should work with zero issues. https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main. Rename it to: sdxl-vae-fp16-fix I can't see why it would not? You may have memory issues or something else? Unsure. But yes that VAE is a 'slightly' brighter / saturated version made for SD1.5 long ago based off the same principle.
adding contrast:0.5, or vibrant colors, to any prompt / model worth it's salt should fix your issues though generally and if it doesn't.. something else is wrong.
Would you say it's capable of performing well with sci-fi too?
I've really liked the style I see, so I'm gonna test it out anyway.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.