JV-MIX (MERGE)

VERSION 2.0 Revison 1: New Models, Improved Generations. Different Sampling settings

(TESTED and CONFIRMED): This model works wonderfully with the LCM Sampler and LCM Lora. Make sure when using the LCM Sampler that you have your steps set to at a minimum of 6.

Merge of 3 Checkpoints:

• MIX-Pro-Remaster (629f56d8d5)

• Sudachi (9b170ebadc)

• tmndMix-VII (e16273c410)

Model Parameters:

• Sampler: DPM++ 2M SDE Karras or any other DPM model (Except 3M)

• Sampling Steps: 20 - 40 (30 Recommended)

• Hires. Fix: Enabled

• Upscaler: 4x-UltraSharp or 4x-AnimeSharp

• Hires Steps: 15

• Denoising Strength: 0.4

• Clip Skip: 2

• VAE: MoistMix

Recommended Textual Inversions:

• veryBadImageNegative

• Negative-hand

PLEASE DON'T USE ANY OF THE MODELS I HAVE MERGED OR CREATED FOR COMMERCIAL PURPOSES

If you have any questions feel free to comment

描述:

Added tmnd-mix as tertiary model, weighting throughout each model: 0.4.

Added Sudachi as secondary model.

Removed Mix-REV4.

Supports Inpainting, Img2Img

Best ever lora support

训练词语: