This is a tutorial to achieved the same style from the images posted here.
This requires some LORAs (linked at the end) to work.
A1111 Short Tutorial :
1) Generate an image with one of my models (MIX 8/11/12 LF or not LF) without LORAs. When you have a good one :
2) Send it to ControlNet to make a depth map of the image (depht midas is used here) with a strength of 0.5.
3) Add the LORAs list to the prompt and hit generate. Try incrementing seeds if first result isn't great until finding a nice one.
4) Upscale your image with the same prompt (x1.5, denoising around 0.5).
Other suggestions :
Navigating by incrementing seed instead of going random.
You can edit the depth map with Krita to help the AI going in the way you want
Editing with Krita between different steps to decrease chances of having bad images
Because of the LORAs, it will be more difficult to get uncommon concept, you may have to raise the value of some words
When up-scaling, to have better outputs, you can decrease the impact of the LORAs by doing this : ("LORAs list"):0.6 for instance (I would recommend to put in the bracket every LORAs but not the add detailer one because he has a good impact on the composition.
For ComfyUI, it is the same idea, you'll find in downloadable files this workflow to do the trick :
This is a simple workflow but effective, you could adjust it with your own knowledge on comfyUI. Be aware that there is a specific way of calculating the right values with the LORAs stacks that isn't like in a1111 so it is a bit more tricky.
With all this LORAs, there is an infinity of styles you can have and even more by adding other LORAs from your choice.
In fact there are some downsides to this :
It is necessary to use a depth map from a good image, otherwise, the composition will look to much like what are the LORAs trained on and to few like the original models.
It will be very worse with hands and a Krita intervention is recommended to fix hands.
It tends to do thicker character.
It brings more often weird stuff than without LORAs.
Based on some NSFW/Hentai LORAs, so it's good for hot and nsfw images that may be pegi 16 to pegi 18.
But on the aftermath, we end up with very pleasant images to look at and a style that is going further every traditional images generated by AI. In my point of view, this is insane and revolutionary, hopping that some of the downsides will disappear with SD 3.0...
I also hope you will enjoy those crazy unlimited styles.
Here, three styles suggestion to start using the trick :
Style W1 :
<lora:add_detail:0.8>, <lora:Platinum Noir The Other Side (NSFW) LoRA:0.8>, <lora:Kase_Daiki:0.4>, <lora:oil2.0_2-512LORA:0.4>, <lora:blue-senpai:0.4>, <lora:Flower_Prison:0.2>, <lora:naruko_hanaharu_SD1.5_dim32:0.2>
Style W2 :
<lora:add_detail:0.6>, <lora:Platinum Noir The Other Side (NSFW) LoRA:0.7>, <lora:Kase_Daiki:0.5>, <lora:oil2.0_2-512LORA:0.5>, <lora:blue-senpai:0.5>, <lora:Flower_Prison:0.3>, <lora:naruko_hanaharu_SD1.5_dim32:0.3>
Style W3 :
<lora:add_detail:1.0>, <lora:Platinum Noir The Other Side (NSFW) LoRA:0.9>, <lora:Kase_Daiki:0.3>, <lora:oil2.0_2-512LORA:0.3>, <lora:blue-senpai:0.4>, <lora:Flower_Prison:0.2>, <lora:naruko_hanaharu_SD1.5_dim32:0.2>, <lora:make25d_type5_v100:0.3>
LORAs links :
<lora:add_detail:1.0> https://civitai.com/models/58390/detail-tweaker-lora-lora
<lora:Platinum Noir The Other Side (NSFW) LoRA:1.0> https://civitai.com/models/371653/platinum-noir-the-other-side-nsfw-lora
<lora:Kase_Daiki:1.0> https://civitai.com/models/207589/kase-daiki-style
<lora:oil2.0_2-512LORA:1.0> https://civitai.com/models/17455/shiny-oiled-skin-20-lycorislora
<lora:blue-senpai:1.0> https://civitai.com/models/140555/blue-senpai-or-style
<lora:Flower_Prison:1.0> https://civitai.com/models/155005/winged-cloud-sakura-dungeon
<lora:naruko_hanaharu_SD1.5_dim32:1.0> https://civitai.com/models/391162/naruko-hanaharu-style-sd15
<lora:make25d_type5_v100:1.0> https://civitai.com/models/311348/convert-2d-to-25d-type5-25d-type5