Frieren Cosplayer CC ++ [Ilus] (r557)

IllustriousXL:

My second consistent character cosplayer. In this attempt, I expanded my character's capabilities to work with as many Illustrious types and versions as possible, including checkpoints trained on v0.1, v2.0, and the realistic fine-tunes. This time, you should be able the generate Frieren across many styles of models.

Overview:

• The model consists of three outfits: Main Dress, Nightgown, and Nude.

• Masked training was used with OneTrainer for more flexible backgrounds.

• Regularization was used with a large set of images.

• LoRA trained (Cosine Prodigy with Min_SNR_Gamma loss) using three separately trained LoRA models and then merged with DARE-TIES.

The training was tested using the MGAS comparison method to find the best epoch. The introduction article can be found here.

Usage:

1. Trained on: CyberRealistic-v3.8 & IllustriousXL-v0.1/2.0. Works well with most IllustriousXL models, from realistic, anime, to semi-real. Check the previews for examples.

2. Size: 1:1, 2:3 XL, Steps: 25-35, Sampler: Euler-a, CFG scale: 4-6, Strength: 1.0. Also works well at highres, e.g. 1024x1536 & 1248x1824 for 2:3.

3. [Optional] Adetailer can help. Preview images were only highres fixed with a denoise of 0.4 and upscale of 1.5x.

4. All outfits are tied together by the common tags, {frid7dren, photo of a woman, white twintails, pointy ears}.

5. The main triggers for the individual outfits:

   • Main Dress - {white tunic and capelet with gold trim, striped undershirt, leggings, tan boots (or brown boots)}

   • Nightgown - {white nightgown}

   • Nude - {nude}

6. Variations and helper tags (i.e. not always needed/wanted):

   • For all outfits: green eyes, red earrings.

   • Main Dress: white sleeves, belt, dark leggings, ornate staff.

   • Nightgown: navel, white panties, barefoot.

   • Nude: breasts (small, medium, and large), navel, pubic hair, pussy, barefoot.
     Camera angles: full body, cowboy shot, close-up

==========================================================