The First Descendant - Viessa's Summer Paradise w/ Head Skin

Viessa's Summer Paradise w/ Head Skin

V2 -

1. removed baked in makeup cosmetic.

2. Swimsuit is near perfect now, matching in game skin

3. This is Viessa's 3rd Swimsuit Skin released in-game last summer, if anyone was curious.

• Tags: smrvssa, black hair, long hair, swept bangs, light-blue wide eyes, lips, earrings, eyewear on head, blue one-piece swimsuit, strap chain, jewelry, high heels, bare shoulders

• check example images for other tags, depending on the model you use, the pre-tags can make a huge difference.

checkpoints:

• nova3DCGXL_illustriousV60

• realisticBlendXL_v40

• waiIllustriousSDXL_v150

• perfectdeliberate_v50 (stick with V5.0 or V4)

  • V6 still has weird issues just overall not just this LoRa.)

• 832x1216

• 896x1152

• Lora Strength: .75 - 1 (when using different poses, sometimes you lose swimsuit detail, so use a higher strength like 1 if that occurs.)

• DPM++ 2M SDE, DPM++ 2M, Euler a, (most success with the first two)

• CFG: 4.5 - 5 (4.5 sweet spot)

• Steps: 25-35

• ADetailer for Face.

Hires. fix: Lanczos, Upscale by 2, Denoise: 0.3 to 0.4, Hires CFG Scale: 4.5, (I didn't mess with upscaling too much in this version, feel free to test and let me know what works!)

some of the training area from in-game is baked in still but it can be overwritten easily, this is my first lora still, so any tips is much appreciated!

P.S. Viessa's eyes are technically a light-grey according to in-game screenshots running max Cinematic settings.

---

V1

This is my first lora and is far from perfect.

These were generated in Forge-Neo, so your experience will differ in Comfyui.

My in-game screenshots had her hair dyed blonde but you can simply change it to the default look by adding 'black hair.'

Checkpoints that I've tested and work well:

• perfectdeliberate_v40.safetensors

• perfectdeliberate_v50.safetensors

• waiIllustriousSDXL_v150.safetensors

• prefectIllustriousXL_v3.safetensors

• ilustmix_v9.safetensors

• nova3DCGXL_illustriousV60.safetensors

• illustriousAnimilf_v10.safetensors

• Resolution: 832 x 1216

• Steps: 25-35

• CFG: 5-7

• Sampling: DPM++ 2M, works great for PerfectDeliberate, or Euler A can work as well.

• Hires. fix: Lanczos, upscale by 2, denoise 0.25 to 0.4, hires CFG scale 5.

• Strength somewhat low around 0.6 - 0.8 seems to be the sweetspot, otherwise the background seems to get baked in. This is my fault I didn't crop my images.

Trigger Words:

smrvssa, blue eyes, red lips, competition swimsuit, yellow-framed eyewear, strapless, bare shoulders, eyewear on head

if you're having trouble with the correct style swimsuit, try 'single strap' or 'one strap' instead of 'strapless' after 'competition swimsuit'.

use ADetailer for Face when needed.

Any tips for future lora training are encouraged. Thanks!