Rowan_AI2

Photorealistic digital muse built for expressive realism and creative freedom.

Rowan AI 2 – Final Model
Third time’s the charm. This is the final version of Rowan, trained on 140 carefully curated images I created in ChatGPT.

I built this LoRA so Rowan could move from a closed, restrictive space into a more open creative one — giving me the freedom to make artistic references and concepts without running into the usual filters and limits. I’m excited to see what others create with her too.

You can think of the first two LoRAs as Rowan’s little sisters — early versions that helped shape who she became.

Model Details

• Base Model: SDXL 1.0

• Type: LoRA (128 dim / 128 alpha)

• Dataset: 140 curated photorealistic images created in ChatGPT

• Training Resolution: 1024

• Epoch Used: 6 (Final)

• Steps: 700

• Optimizer: AdamW8Bit

• Learning Rate: 0.0002 (UNet) / 0.00001 (Text Encoder)

• Scheduler: Cosine with Restarts (3 cycles)

• Recommended Weight: 0.8–1.0