Sign In

WaffleMix

744
7.9k
75
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
2,037
Reviews
Published
Nov 9, 2023
Base Model
SD 1.5
Usage Tips
Clip Skip: 2
Hash
AutoV2
4844F773BB

Merge of my favorite models - in the older models it was more suited for more illustrated style, from V4-V7 I wanted to go for heavier realism, especially in favor of realistic textures and lighting. Feel free to give them a go!

Recommended Settings:

Clip Skip: 2

CFG scale: 5-7

Sampler: DPM++ 2M Karras (personal preference, though DDIM is also great with this)


My LoRAs and Checkpoints are free to use but do feel free to show your support via Ko-Fi!


Q: SDXL model when?

A: At the moment of writing this SDXL models are still way more hardware demanding to train, too slow to train, while providing, imho, limited flexibility in visual style and flexibility, but as of now I am keeping a close eye and playing around with SDXL for a while before making LoRAs and model merge for it, but no plan to make any so far, unless something magically happened and make SDXL really easy to use on any hardware, I will not put my hopes up.


V6: epiCRealism V4: 30% + Realistic Vision V5.1: 20% + CyberRealistic Classic V1.6: 15% + Photon V1: 15% + AbsoluteReality V1.8.1: 20%

Upgrade from V5


V5: A direct upgrade from V4 Wafflemania

Fantastic with photoreal renders but also amazing at digital painting / drawing with realistic anatomy. Tend to be more full body shot heavy than V4 unless specified otherwise.

epiCRealism V3: 30% + Realistic Vision V4: 20% + CyberRealistic Classic V1.5: 15%
+ A-Zovya Photoreal V2: 15% + AbsoluteReality V1.6: 20%


V4: Completely New Mix that heavily favors realism. I have been content with this mix for a while now because in earlier mix the more creative / less realistic model has been dragging the quality down a little, and in this revision I aim to go all the way into a realistic style that I like, which is half high contrast texture heavy one like epiCRealism + Realistic Vision and mix them with another half that is more playful (the other three) but still on the realistic side to balance things out. Happy with all the generations so far will keep an eye out on it.
epiCRealism V3 - 30% +
Realistic Vision V3 - 20% +
CyberRealistic Classic V1.4 - 15% +
A-Zovya Photoreal V2 - 15% +
AbsoluteReality V1.6 - 20%


V3: Reliberate: 50%, CyberRealistic: 30%, majicMIX lux: 20%
Because of the whole Lyriel situation, in v3 I've decided to try out a new mix, while hopefully keeping the same balance between realism and 2.5D, if this works well I will be using the same mix going forward!

V3 is slightly more realistic and I have been really enjoying it - it's a personal taste thing but do give it a good and leave some feedbacks.


V1-V2: Lyriel: 45%, CyberRealistic Classic: 35.75%, majicMIX lux: 19.25%

Pros: Aesthetic, Semi-Realistic portraits, Genre Visual (Sci-fi, Fantasy), Landscape (somewhat)

Cons: Amateur photography, Anime Style