Alice SDXL

SDXL Version of Alice LoRA

I created the character Alice using a large collection of images generated with the Imagen 3 model. Because the ImageFX platform made it easy to maintain character consistency within a single thread, I produced more than 100 images there of a woman with carefully defined visual traits.

I refined her appearance with the help of a personalized GPT model that guided me through building the face—hair color, eye shape, eye color, and so on. Alice’s look is entirely my personal preference, and I emphasize that she is not based on any real person.

Here is part of the appearance prompt that was used to build Alice in Imagen 3:

“a seductive woman with long messy wavy brown hair pushed to one side, green upturned fox-shaped eyes with sharp black eyeliner, thick curled lashes, dark expressive eyebrows, full red glossy lips slightly parted, playful seductive smirk, photorealistic”

Initially, I supplied several dozen images to train the Alice LoRA for the FLUX models. This allowed me to generate a much larger and highly consistent new iteration of characters, which in turn became the basis for Alice SDXL.

Why the name Alice? Simple :) That’s the name of my AI assistant who helps me in my creative work.

As a LoRA for the SDXL model, it delivers the best results in a three-stage workflow:

1. Stage 1 – Build the overall image using a well-trained SDXL checkpoint with the LoRA attached (≈ 27 steps, though this can vary by checkpoint).

2. Stage 2 – Run it through a trained SDXL Refiner checkpoint with the LoRA attached and perform an Upscale (≈ 10 steps).

3. Stage 3 – Apply face detection and FaceDetailer with the main SDXL checkpoint and the LoRA attached (≈ 10 steps).
   This final stage is crucial for keeping the face consistent, even when the full character is shown at a distance.

Recommended strength: 1.0