-Update v1.6
This is Reworked version for try to make it more flexible an powerful.
Give a review is appreciated, and please report where it not work.
Tested with ONLY it on negative, and with other embedding for better compatibility.
Made some sample image with old v1.4 prompt for comparison.
-Update v1.4
reworked for better understanding prompt.
try to make it get clear line/color in cartoon-anime style prompt.
tested with and without some of my other embedding for make it not washing out other embedding.
worked for do better background consistence.
Some sample image made with my old model KKW Ideal and standard DPM++ 2M Karras sampler.
Tips
When use it in negative, and you want to weight it, just use these tips: after weight it, add it again without weight, like
(kkw-new-neg-v1.4:1.8) kkw-new-neg-v1.4
********
If in your model weight it giving bad result, try adding it multiple times without weight.
After many test, i reworked my negative for made a single one that not waste so much prompt and work great for all style you want.
The last 8 image attached are same prompt at different cfg(2,5,8,11,15,20,25,30) to show how this negative work.
Tested on my own model, let me know how work in other model, share result help me to do it better.
Description
New formula negative v1.0
FAQ
Comments (10)
Hi, where can I find the kkw.kkw-DevKkw model?
Not published yet. is a WIP merge of all my model, the base model is IDEAL.
Wow! Quite possibly the best negative embedding yet. It even pairs nicely with other negatives such as CyberRealistic_Negative-neg.
Thank you.
How do you measure that? I'm asking honestly, because I really want to know how to test these. Seems that negatives should remove "everything bad" from any style or prompt or do I have too high expectations? :P
@olivetty if i correctly understand what you mean with "measure", before release i test it in these way:
-made a wildcard with different prompt, style etc..(about 100-150 prompt)
-run generation with using only negative i want to test and for positive using created wildcard (for these i use a modified wildcard script for don't trigger same prompt- modified script aviable in my github page)
- repeat test from 1 to 30 cfg
- checking result and if is good start testing in same way but adding other embedding to see how it work mixing it.
-if seem good i publish it, having more user try it mean have more possible issue and work to make it better
Obviously "good result" is also subjective, and checkpoint used is also a variable we don't have to underestimate.
@devilkkw Exactly what I was looking for! :) So this is for anything, anime, photo, landscapes, people etc?
@olivetty yes, i have different prompt in wildcard, from simple to detailed, and different style.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.