<h2 id="sylvia-h:-princess-soldier-bastard">Sylvia H: Princess - Soldier - Bastard</h2>Dutiful and deeply conflicted bastard daughter of an Empress and Empire on the brink of total war. A decorated graduate of the Imperial Military Academy, she seeks to atone for the sin of her birth. Would you like to know <a target="_blank" rel="ugc" href="https://civarchive.com/articles/11461/a-world-of-characterswip">more?</a><h3 id="anima">[ANIMA]</h3>New dataset and new training method used for an Anima version of Sylvia. Depending on how popular this is, I may make more character loras with Anima. Sylvia is the trigger - enjoy!<h3 id="krea-2">[KREA 2]</h3>Realistic (fictional!) concept for Sylvia - new Krea 2 version. Trigger word is now just Sylvia - model mostly keeps her face consistent so prompting for short hair and brown skin can be useful for that extra push - interestingly, Krea had a harder time with her freckles and beauty marks than Klein and ZIT. Tested between 0.6 - 1.3 strength, typically using at higher strengths for face-focused shots and lower strengths if the character is subjected to style changes (hair, etc.). Enjoy!<h3 id="flux.2-klein-9b">[FLUX.2 Klein 9B]</h3>Realistic (fictional!) concept for Sylvia H - new Klein version using her softened appearance. Trigger word is now just Sylvia - model mostly keeps her face consistent so prompting for short hair and brown skin can be useful for that extra push. Tested between 0.6 - 1.3 strength, typically using at higher strengths for face-focused shots and lower strengths if the character is subjected to style changes (hair, etc.). Enjoy!<h3 id="z-image-v2">[Z-Image V2]</h3>Realistic (fictional!) concept for Sylvia H - version 2 softened her appearance a little and I experimented with freckles and beauty marks, as well as more body-focused data that helped capture the intended tall/lean/athletic frame she should have. Training happened twice, baked it harder with 15 epochs and got closer. Trigger word is now just Sylvia - model mostly keeps her face consistent so prompting for short hair and brown skin can be useful for that extra push. Tested between 0.6 - 1.3 strength, typically using at higher strengths for face-focused shots and lower strengths if the character is subjected to style changes (hair, etc.).<h3 id="z-image-v1">[Z-Image V1]</h3>Realistic (fictional!) concept for Sylvia H. Z-Image training went very well despite me not having a lot of experience with captioned Loras. SylviaH is the trigger - model mostly keeps her face consistent so prompting for short hair and brown skin can be useful for that extra push. Tested between 0.8 - 1.5 strength.Data was created through Nano Banana Pro - first a face was composited and edited, then training images were generated. I would consider this a prototype but the results are very good! I hope you enjoy.<h3 id="illustrious">[Illustrious]</h3>Same dataset as Pony[v2] but with all the added greatness of Illustrious. Training went very well, character is well defined with nothing but the Sylvia H trigger tag. Plays well with styles and actions and other LoRAs. Occasionally brown skin is useful.Tested mainly at 1.0 strength (in theory could lower it if anything causes issues) with both Euler A and DPM++ 2M Karras, but theory states it should work with whatever you personal preference is. Have fun and enjoy!<h3 id="flux">[Flux]</h3>Realistic concept of my OC Sylvia H LoRA. Face and skin are relatively consistent, which is nice considering this Flux version was developed off of a training dataset I tried (and failed) to use for a realistic SDXL version.Works well at around ~0.9 but anywhere from 0.7-1.0 work well. Tags like brown skin and short hair can keep it further on track.Hopefully people have as much fun with Sylvia and this new Flux model as I have!<h3 id="pony">[Pony]</h3>Working as usual, tags like brown skin and small breasts help keep it aligned. Lower strengths ~0.85 are what I usually use. Plays nice with other LoRAs, particularly styles, so have fun!<h3 id="creator-notes">[Creator Notes]</h3>Everything I make is for one purpose - fun. I don't mind not knowing much about AI (tbh don't even really understand how I managed to install A1111/Forge, Git is an enigma to me) , it allows me to create, learn, and grow. Personal projects are fountains of enjoyment for me, and in sharing them I hope to share some of that with the world - hopefully someone else feels inspired to create something!

✦ Z-Cast: Sylvia H. [OC]

Softer, more my style ZIT prototype. Different dataset methodology.

2750702_training_data.zip

SylviaR3.safetensors

ZIT v2

Anima

Krea 2

Flux.2 Klein 9B

ZIT v1

Illustrious

Flux v0.1

Pony v2

Pony v1

Files with SHA256: 8040e11a8bc600e8146a7b62b194ff46bfd7698ede69781a349f8d86dd65cea0