Decided to share this merged comic book style model that I originally made just for myself. I have been trying to get a stable mix of different art models, and I feel like this is finally one of my best attempts so far!
Updates:
v4: More consistent and less busy look, experimenting with some Lora merges for a flatter look
v3: Updated with better curated models and a more consistent look
v2: Merged with updated initial models, and added a wider variety of models for better styles
Please do share your images, I am super excited to see them!
Good-To-Knows
Here are some thing to know/keep in mind about this model:
You can get detailed images out of relatively short prompts (I'll keep updating the sample images as and when I try more stuff)
Works really well with embeddings like EasyNegative.
Sometimes has trouble with male bodies; adding feminine in negative prompts tends to help for that.
Model handles NSFW quite well imo, but do note that its not a dedicated nsfw model. So you might have better success for only nsfw prompts elsewhere.
Reasoning
I find myself coming back to this model a lot. While other comic models get the 'texture' of a page and lines better, I gravitate towards clean lines, flat areas, and visual fidelity a lot - the kind found in modern comic book workflows. I feel like this model does that quite well.
I'm still quite new to the checkpoint game, having only trained a couple of loras and dreambooth checkpoints myself so far. I will be experimenting and hopefully versioning up this model as I find more stable and interesting models to incorporate!
Ingredients
Its a merge between ICoMix, Babes, OccidentalMix, Western Animation Diffusion, and a sprinking of Aniflatmix (overall I wanted something weighted towards non anime style faces). Additionally, I'm applying a LoRA that is this one very sensitive comic book style model that I have at extremely low weight (0.1).
Actual merge board recipe for the checkpoint can be found in the model info with v2 onwards.
Thanks to lostdog, alexds9, SpicyTree, Lykon, Duskfallcrew, and CyberAlchemist for their work on the initial models!
If you like this, check out my other models!
Description
More varied training with a lot more curated art style models
# Recipe
= icomix_V05 + babes_20, 0.5, ckpt
= __O1__ + occidentalmix_v20 + model-v1.5, 0.2, ckpt
= __O1__ + westernAnimation_v1 + model-v1.5, 0.2, ckpt
= __O2__ + __O3__, 0.5, ckpt
= aniflatmixAnimeFlatColorStyle_v20 + poodaBeepMix_illustrationComic, 0.3, ckpt
MCBS = __O4__ + __O5__ + model-v1.5, 0.1, safetensors
# Variables(System)
__O1__
__O2__
__O3__
__O4__
__O5__