Dragon Age Inquisition - Cullen - Pony

Details

Model description

I noticed Cullen Rutherford LoRAs for Pony were pretty non-existent for a time so here we go.


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:

cullen_rutherford, brown eyes, 1boy, male

V4 outfit tags:

fur-trimmed cloak, pauldrons, breastplate, shirt, gauntlets, gloves, belt, black pants, boots

trained style tags (particularly when using the Pony V6 checkpoint and no style LoRA added to the prompt):

  • realistic (use this with 2D/2.5D higher detail shots)

  • photorealistic (use this for realism shots)

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.


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:

cullen_rutherford, brown eyes, male

V3 outfit tags:

fur-trimmed cloak, pauldrons, breastplate, shirt, gauntlets, gloves, belt, black pants, boots

trained style tags (particularly when using the Pony V6 checkpoint and no style LoRA added to the prompt):

  • realistic (use this with 2D/2.5D higher detail shots)

  • photorealistic (use this for realism shots)

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.


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:

cullen_rutherford, brown eyes, male

V2 outfit tags:

fur-trimmed cloak, pauldrons, breastplate, shirt, gauntlets, gloves, belt, black pants, boots

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.


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:

cullen_rutherford, brown eyes, male

V1 outfit tags:

armor, gloves, black pants, boots

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.

Images made by this model

No Images Found.