Z-Image LoRa // GLZD_Porcelain
This LoRa designed to transform subjects into life-size, high-gloss porcelain figures with a focus on intricate, fired-in decorative elements.
Core Characteristics
Materiality: The primary effect is the conversion of skin into pure, uniform, RAL-style ivory or white porcelain. The surface features true ceramic realism, including wet high-gloss specular reflections, a fine "orange-peel" glaze microtexture, and microscopic glaze pooling at the edges of raised details.
Anatomical Realism: Despite being made of hard ceramic, the LoRA often depicts subjects with "porcelain elasticity," where the material yields and deforms realistically to show weight, gravity, and soft-tissue compression, particularly in curvy or voluptuous figures.
Recurring Decorative Elements
Floral Relief: A signature feature is the inclusion of dense, high-relief floral appliqués. Common colors include pink, blue, purple, and yellow flowers, often with gold paint accents.
Intricate Tattoos: Subjects frequently feature complex tattoos that appear glazed or fired directly into the porcelain, often in cobalt blue lattice patterns.
Gold Accents & Jewelry: The LoRA heavily utilizes gold filigree scrollwork, gold jewelry (rings, earrings, necklaces), and polished gold grills on the teeth.
Facial Makeup & Paint: Stylized features include painted eyebrows with visible brushstroke textures, colorful lips (blue, black, or purple), and star or heart sticker decals.
LoRA training dataset prepared with Debiddow works.
Description
Version 1
FAQ
Comments (6)
this would be a dope wan 2.2 lora
thanks
Did you know that FLF2V (First + Last Frames to Video) can work with your image references? It's somewhat effective for capturing details like this. Using lora, especially for general ceramic styles, would be even better I agree. I believe it has some real potential without lora. I would explore someday but not 2.2 and need more actual video footages of it to expand.
@midnight1111 - I have not mastered fflf myself. I have, however, trained a couple hundred wan 2.2 loras. This could be done quickly and easily and definitely does not require video. I may do it if I find some time.
Debiddow!
Winner chicken 🐣 and I have tested with Grok Images. It turned out capable of producing these works. Like he used Midjourney with it, you could train lora with the Grok images dataset.
@midnight1111 Agree! good job.




