Marsaili is an Autumn Muse in her mid 20's.

Marsaili embodies the fiery essence of fall through her vibrant presence. This Textual Inversion (TI) was crafted to capture her striking beauty, characterized by curly vibrant red chin-length hair and emerald green eyes that radiate warmth and intensity. Her pretty freckled face, paired with a soft rosy complexion, gives her an air of natural charm and vitality. Marsaili’s features include high cheekbones, a delicate jawline, and an elegant balance of sharp and soft contours, reflecting her bold yet approachable aura.

Marsaili TI Changelog:

v1 - Initial experimental version.

An experimental step in refining my AI image generation techniques. By focusing on her autumn-inspired essence, this TI provides a starting point for creative exploration while avoiding common pitfalls like dataset entanglement or overly rigid prompts. This balance allows Marsaili to shine as a unique and adaptable muse in your artistic toolbox.

Developed using the embedding merger tool, this TI focuses exclusively on facial features and basic body shape, leaving backgrounds, and art styles to the user's creative interpretation. The approach avoids over-fitting and ensures that Marsaili’s core traits are adaptable across a wide range of scenarios, from rustic autumn landscapes to modern and fantasy settings.

The drawbacks: Version 1 uses about 2x as many tokens compared to other face TI's that are trained on images. This model was not trained on actual images and will have fairly random faces aside from the key details mentioned above in bold. If you find a particular face you like, stick to that seed and adjust the prompt. I created this TI as a way to be referenced using wildcards of my favorite faces.

v2 - Plans - More facial feature consistency.

Building on the Snöfrid v2 process, the following steps are planned to refine Marsaili’s representation:

Image Preparation:

• Generate outputs from Marsaili v1 and cherry-pick images that best capture her defining features and style.

• Merge selected images using ReActor, upscale with Hi-Res Fix, and downscale to standard training size for training.

Training Process:

• Begin with a midpoint TI trained on the SD 1.5 base model for initial stability.

• Continue training with a blend of my own two stylistically distinct models—one for illustrated features and one for semi-realistic elements—to generalize her traits and improve compatibility across model types.

Merging and Refinement:

• Combine the final TI with v1 and the midpoint TI to balance continuity and refinement. Merge factors will be adjusted iteratively to achieve consistency and versatility.

Goals:

• Ensure Marsaili’s TI remains adaptable across varied styles and checkpoints while preserving her unique identity and aesthetic.