Dominic Santiago | Gears of War - CivArchive (CivitAI Archive)

Dominic Santiago | Gears of War - v1.3

NSFW

I made Dominic Santiago or known as Dom in Gears of War community. This loveable man was Marcus Fenix's best friend and right hand. I think Dom deserved to be made into a LoRA so you can made some artwork based on him. If you want to use Dom with his big beard like in Gears of War 3, you can visit this link instead: Full Beard Dom. Enjoy!

Trigger Word:

dominic_santiago

Here's some example of common words of typical Dom render:

dominic_santiago, muscular, short hair, dark skin, (beard:1.1), chubby, standing, armor

//EXPLANATION

This prompt has it's trigger word 'dominic santiago'. This prompt has several words that support of Dom's appeareance:

1. muscular --> making Dom muscular. You can add some weight to determine how muscular Dom's gonna be. Syntax: 'muscular:<weightNum>', which <weightNum> is number of weight with default of 1.0. 
2. short hair --> making Dom has short hair
3. dark skin --> making Dom has darker skin
4. beard --> making Dom has beard
5. chubby --> making Dom looks chubbier
6. armor --> making Dom donning his COG armor

The model also support NSFW tagging too.

Settings I used:

Weight: 0.75 - 0.9
Sampling Method: DPM++ 2M SDE Karras
Sampling steps: 20 - 24
Resolution: 768x768 or 960x960 (768x960) or 640x768
- You can use HiRes + Upscaler too. I suggest to start from 640, though any resolution will work (don't go too high because of tiling problem).
CFG Scale: 5 - 7
Checkpoints
- Airfuck's Brute Mix
- Realistic Vision --> I recommend this bit closer to the intended model
- Mature Male Mix
- Seiza Mix

Description

Updated image quality. Using only 75 images.

v1.1:

I noticed major problem in previous model that the model were extremely rigid and I struggling to give other clothes. So, I updated this model quite shortly after this LoRA released to public. Here's some changes:

Increase image quantity in dataset from 50+ to 100+ images.
- The image increase includes replacing any bad images to much better images.
- I added much more variety in terms poses and clothings (including stripped version of Dom for sake of flexibility)
Increase LoRA size.
Decreased epoch by 1 from 15 to 14.
- I noticed that the model that used 15 epochs tends to overfitting just a little bit. So, I used the model that has 14 epochs.
Increased steps in training model hyperparameter
Deprecated 'dom_santiago' trigger word. You can still use this, tho but I won't mention it. Future update will completely removed.

v1.0:

Here's some info about Dom in this LoRA:

Trained using 50+ images
Epochs 15 and 10 repeats --> 4425 steps

v1.2:

I updated this model with adding more dataset and eliminates any bad images of this model. It's increased its flexibility and more easier to work (armor easily stripped off and change to various clothing). Here's some updates of this model:

Increased image that I used to construct this model from 100+ to 120+ images. This including elimination of bad image.
- The image that I used consists of Dom's using other clothing, NSFW, and much better quality COG armor
- Poses are much more varied
Increased steps during training
Experimented optimizer that I used. I change Prodigy to AdamW8bit sith Stable Diffusion 1.5 as its base.
Fully deprecated 'dom_santiago' activation tag for simplicity reason

LORA

SD 1.5

by doomguy11111

Download (Beta) View on CivitAI