Sign In

Vibrant Horizon - Turbo XL

350
2.0k
52
0
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
1,158
Uploaded
Jan 14, 2024
Base Model
SDXL Turbo
Usage Tips
Clip Skip: 1
Hash
AutoV2
ABD3DB6820
0
0
0
0
0
This Stability AI Model is licensed under the Stability AI Non-Commercial Research Community License, Copyright (c) Stability AI Ltd. All Rights Reserved.

Vibrant Horizon -Turbo XL (V2)

What's new you ask?

  • Slightly reworked the merges

  • Outputs should come out cleaner

  • Hands should be better

  • More versatility

  • Now works with DPM++ 2M SDE Karras

New favorite settings:

  • Bump up that CFG a little to 3.5

  • Bump up those steps a little to 10

My workflow:

  • Gen at 768 x 1024

  • Hires Fix. at 1.5 up to 2.0 / 8 steps / Denoising at 0.25

  • Adetailer for both faces and hands

Just PNG Info any of the showcase pics to get the settings in A1111 but other than that, just play around with the settings and see what you prefer.

Have fun!

Vibrant Horizon -Turbo XL (V1)

VHTXL is a first dip for me into the world of XL. This is for now a simple merge of some of my absolute favorite Turbo/LCM XL models.

This model being a Turbo/LCM mix, it allows for some amazing quality gens at very low steps and CFG, while also allowing you to use your favorite extras such as Hires Fix or Adetailer.

I did not originally build this merge to release it, but I have been enjoying it so much that I thought I would share it.

⚠️ I would like to stress that contrary to my other merges, this model is NOT focused on NSFW male pictures generation meaning no peen for you, at least for the time being. As the technology evolves and more content becomes available, I will revisit the idea. ⚠️

This merge includes the following models:

These models are all incredible in their own way and I really invite you all to check them out as well as the rest of the work from their amazing creators.

SETTINGS

This is a Turbo/LCM model, meaning that the settings are very important and that it is best for you to ensure you are using what follows. You are of course free to experiment, but the settings below are what generally works best:

  • Sampling Method: DPM++ SDE Karras

  • Sampling steps: 8

  • Width: 576

  • Height: 1024

  • CFG Scale: 2 > 3.5

  • Clip Skip: 1 or 2

Experiment, see what works best for you.

These are the basic settings you can use to generate your pictures.

VAE

I like to use the original sdxl_vae.safetensors VAE - Other VAEs kinda wash out the outputs imo.

HIRES. FIX

If your hardware allows and if you want to take things further, you can add the following:

  • Hires. fix: On

  • Hires steps: 8

  • Denoising Strength: 0.25 (or whatever you like)

  • Upscale by: Up to you, but the higher the more taxing. I like 1.5.

  • Upscaler: Up to you, depends on the style you are trying to achieve. For realistic, my go to is 8x_NMKD-Superscale_150000_G. For Cartoon/Anime, I use R-ESRGAN 4x+ or R-ESRGAN 4x+ Anime6B. Play around with it and see what you like.

ADETAILER

Further to this, you can also use Adetailer to fix faces, just like you would with any 1.5 model. You can leave the settings as are or you can experiment.

And that's pretty much it. The model itself is capable of a lot of different styles and concepts and will give you great gens, even with simple prompts.

As this is based on XL, remember that you need to rethink the way you prompt. XL is smarter and understands what you want better. You however need to ask for what you want correctly.

Anyway, hope you enjoy, and once again, please do have a look at the models included in this merge as they are really incredible.

Have fun!