I noticed Cullen Rutherford LoRAs for Pony were pretty non-existent for a time so here we go.<hr />V4 (clip skip 1) is my fourth attempt at adding a bit more detail to this LoRA. It was during a recent new playthrough of DAI that I decided to grab cleaner training images of Cullen to add so I got extra face shots and trained some additional concepts (like outdoors, snow, castle). It didn't do much for his mouth scar but...it is what it is.I still recommend putting "simple background" in the negative prompt to reduce the chances of getting a blank white background.V4 basic prompt tags are as follows:<pre><code>cullen_rutherford, brown eyes, 1boy, male</code></pre>V4 outfit tags:<pre><code>fur-trimmed cloak, pauldrons, breastplate, shirt, gauntlets, gloves, belt, black pants, boots</code></pre> trained style tags (particularly when using the Pony V6 checkpoint and no style LoRA added to the prompt):<ul><li>realistic (use this with 2D/2.5D higher detail shots)</li><li>photorealistic (use this for realism shots)</li></ul>All my of generation testing for V4 was done in Forge on CFG 4-5 (for SFW) and CFG 3-4 (for NSFW) using DPM++ 2M SDE - Karras at 20 steps and I was using an adetailer.<hr />V3 (clip skip 1) is my third attempt (after cleaning up the training tags) so I did a better job at identifying angles/shot frame. The faces generate better this time but I still recommend using an adetailer to clean it up (if the shot is further out). I also included better NSFW shots in the training data this time.I do recommend putting "simple background" in the negative prompt to reduce the chances of getting a blank white background.V3 basic prompt tags are as follows:<pre><code>cullen_rutherford, brown eyes, male</code></pre>V3 outfit tags:<pre><code>fur-trimmed cloak, pauldrons, breastplate, shirt, gauntlets, gloves, belt, black pants, boots</code></pre> trained style tags (particularly when using the Pony V6 checkpoint and no style LoRA added to the prompt):<ul><li>realistic (use this with 2D/2.5D higher detail shots)</li><li>photorealistic (use this for realism shots)</li></ul>All my of generation testing for V3 was done in Forge on CFG 4-5 (for SFW) and CFG 3-4 (for NSFW) using DPM++ 2M SDE - Karras at 20 steps and I was using an adetailer.<hr />V2 (clip skip 1) is my second attempt (after learning a LOT more about training tags) so I did a better job at identifying indoors/outdoors (on the training images) as well as labeling specific parts of the armor and such. The faces generate fairly well but I recommend using an adetailer to clean it up.V2 basic prompt tags are as follows:<pre><code>cullen_rutherford, brown eyes, male</code></pre>V2 outfit tags:<pre><code>fur-trimmed cloak, pauldrons, breastplate, shirt, gauntlets, gloves, belt, black pants, boots</code></pre>All my of generation testing for V2 was done in Forge on CFG 6-8 (the sweet spot tends to be CFG 6-7) using DPM++ 2M SDE - Karras at 30 steps and I was using an adetailer.<hr />V1 (clip skip 2) was my rough first attempt at training a Cullen LoRA. I tried my best training it where it'll actually get his mouth scar (it doesn't seem to generate all the time) and the outfit I trained was his armor. It's not perfect but it's still good on being mostly consistent with Pony/SDXL checkpoints.V1 basic prompt tags are as follows:<pre><code>cullen_rutherford, brown eyes, male</code></pre>V1 outfit tags:<pre><code>armor, gloves, black pants, boots</code></pre>All my of generation testing for V1 was done in Forge on CFG 4-7 (the sweet spot tends to be CFG 5-6 depending on the checkpoint but sometimes 7-8 work fine) using DPM++ 2M SDE - Karras.

Dragon Age Inquisition - Cullen - Pony

I increased the amount of training tags and changed the labeling on some of the armor pieces. I also changed the clip skip to 1 since I was noticing my realistic LoRA on clip skip 2 weren't turning out so well.

Files with SHA256: c6ac11612881dbc862d0b2a745f75aa3556c903785505e0db635db87a4f5791f