Here is my attempt at Captain Gray Edwards from Final Fantasy: The Spirits Within since I only saw Aki LoRAs but no Gray.<hr />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: 0.6-0.7V3 main tags are as follows:<pre><code>g_edwards, male, black hair, very short hair, brown eyes, stubble</code></pre>V3 outfit tags are as follows (2 outfits to choose):<pre><code>grey shirt, t-shirt, black pants, black footwear</code></pre><pre><code>military uniform, black footwear</code></pre>V3 optional tags:<pre><code>weapon, gun</code></pre>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.<hr />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:<pre><code>g_edwards, male, black hair, short hair, brown eyes, stubble</code></pre>V2 outfit tags are as follows (2 outfits to choose):<pre><code>grey shirt, t-shirt, black pants, black footwear</code></pre><pre><code>military uniform, black footwear</code></pre>V2 optional tags:<pre><code>weapon, gun</code></pre>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.<hr />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:<pre><code>g_edwards, male, black hair, brown eyes, stubble</code></pre>V1 outfit tags are as follows (2 outfits to choose):<pre><code>grey shirt, black pants, black footwear</code></pre><pre><code>military uniform, black footwear</code></pre>V1 optional tags:<pre><code>weapon, gun</code></pre>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.

Gray Edwards - Final Fantasy Spirits Within - Pony

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.

Files with SHA256: efcdf8eb6472220894e428db3b910a39ce4fe526c085c9117971746b9dd8be43