Gray Edwards - Final Fantasy Spirits Within - Pony

Here is my attempt at Captain Gray Edwards from Final Fantasy: The Spirits Within since I only saw Aki LoRAs but no Gray.

---

V3 is my third version (1st one done via local training rather than CivitAI) and is my attempt to clean up the training tags and improve the quality of generations. I especially wasn't happy with how the hair was generating on V2 (too long/messy). My test results with V3 were fairly decent.

Recommended LoRA strength: 0.6-0.7

V3 main tags are as follows:

g_edwards, male, black hair, very short hair, brown eyes, stubble

V3 outfit tags are as follows (2 outfits to choose):

grey shirt, t-shirt, black pants, black footwear

military uniform, black footwear

V3 optional tags:

weapon, gun

The settings I used for my V3 generation testing were 4-6 CFG at 20 steps using Euler A - Automatic using Forge. I also used an adetailer for the face.

---

V2 (clip skip 1) is my second training attempt and adds a slight amount of versatility when combined with art style LoRAs. I just added more NSFW images to the training set so the AI would be more consistent with 2D/2.5D style LoRAs (especially the face). I tested this with V1 and it definitely didn't work as well when using style LoRAs.

When using realism/2.5D checkpoints I recommend adding muscular to the positive prompt or he'll be a bit on the leaner side on body type. Leave it out for 2D if you want him to be less bulky. I also suggest trying with AND without the realistic tag as it'll change up your results.

V2 main tags are as follows:

g_edwards, male, black hair, short hair, brown eyes, stubble

V2 outfit tags are as follows (2 outfits to choose):

grey shirt, t-shirt, black pants, black footwear

military uniform, black footwear

V2 optional tags:

weapon, gun

The settings I used for my V2 generation testing were 4-6 CFG for SFW and 3-4 CFG for NSFW, 20 steps, DPM++ 2M - Karras using Forge. I also used an adetailer for the face.

---

V1 (clip skip 1) is my first training attempt and generates fairly well. The backgrounds are a bit limited because I used entirely movie screenshots and there's only so much I could do with that. I attempted to train more than 2 outfits but the armor/combat suit didn't train well so...you get 2 options instead of 3.

Also, I use an adetailer in my generations for this LoRA.

Optionally: You can use background tags like hangar or laboratory as those are what I tagged for the 2 types of backgrounds seen in my movie screen grabs.

V1 main tags are as follows:

g_edwards, male, black hair, brown eyes, stubble

V1 outfit tags are as follows (2 outfits to choose):

grey shirt, black pants, black footwear

military uniform, black footwear

V1 optional tags:

weapon, gun

The settings I used for my V1 generation testing were 4-6 CFG for SFW and 3-4 CFG for NSFW, 20 steps, DPM++ 2M - Karras using Forge. I also used an adetailer for the face.