Divine Diffusion by Dever (DDD) [SDXL] - v1.1
推奨ネガティブプロンプト
gaussian noise, worst quality, lowres, oversaturated, undersaturated, overexposed, underexposed, grayscale, bw, bad photo, bad photography, bad art, blur, blurry, grainy, morbid, ugly, asymmetrical, mutated, malformed, mutilated, poorly lit, bad shadow, draft, cropped, out of frame, cut off, censored, jpeg artifacts, out of focus, glitch, duplicate, pixelated, soft focus, color fringing, overprocessed, oversharpened
推奨パラメータ
samplers
steps
cfg
clip skip
resolution
推奨ハイレゾパラメータ
upscaler
upscale
steps
denoising strength
ヒント
Start with no negative prompts as the model doesn't appear very sensitive to negative prompting.
Follow the original creators of AlbedoBase, Mobius, and Paradox 3 to appreciate their hard work.
バージョンのハイライト
v1.1
- fix orange and high contrast bias, less overexposed backgrounds
- better skin, water, dark colours
Dever's Divine Diffusion or DDD
v1.1
(I've kept most of the original showcase prompt with their seeds so you can actually compare the images; I did replace some of the boring images)
- fix orange and high contrast bias, less overexposed backgrounds
- better skin, water, dark colours
What is this ?
A merge of some of my favourite models (that allow sharing of merges with author credit), so here goes, in no particular order:
- AlbedoBase (https://civitai.com/models/140737?modelVersionId=329420) by AlbedoBond (https://civitai.com/user/albedobond)
- Mobius (https://civitai.com/models/490622?modelVersionId=545562) by Corcelio (https://civitai.com/user/Corcelio)
- Paradox 3 (https://civitai.com/models/424460/paradox-3-sd-xl-10) by Thaevilone (https://civitai.com/user/Thaevilone)
+ some (unreleased) LORAs I made
Why am I sharing this ?
This is something I always wanted to do but always put off as being something a bit silly and what a terrible timing this is with SD3 just around the corner.
I'm not going to sell you some bs and tell you that this is some state of the art latest awesomeness model like some other people do, in fact let me do the opposite, this took less to make then it takes me to make a LORA which is really disgusting when you think about it, it's so easy to "borrow" other people's work and call it your own.
Testing it and making 20 images for the showcase took more then coming up with the merge recipes and making the model.
I haven't tested this a lot but in my experiments what has worked well for me (as I typically use anyway) were the following:
Sampler: DPM++ 3M SDE Exponential
Steps: 30
CFG: 7
Start with no negative or whatever, I didn't find it to be very sensible
HiresFix: 20 steps | Denoise 0.4
Anyway, I hope you give this a try and if you like it remember that this is based on the hard work of other people and I'm simply delivering this to you in a convenient package. Go follow those amazing creators and try their models.
