Sirius (Azur Lane) — Character LoRA

This is my Sirius (Azur Lane) LoRA, trained on HassakuXLIllustrious v22, with the main focus being on reproducing Sirius's recognizable appearance and character design.

This LoRA isn't specifically designed for 2.5D realism, but it can produce some pretty good results with the right prompting. Experimenting with different prompt combinations, outfits, poses, lighting and compositions can make a noticeable difference.

Trigger word: Sirius

Base Model: HassakuXLIllustrious v22

Recommended Generation Settings

• Sampler: DPM++ 2M Karras

• Steps: 28–35

• CFG Scale: 5–6

• Resolution: 832 × 1216 (portrait) or 1024 × 1024 (square)

• LoRA Strength: 0.7–0.9

• Clip Skip: 2, if supported by your setup

These are starting points, not strict requirements. Adjust the LoRA strength depending on how strongly you want Sirius's character features to appear, and experiment with prompting to get the style and composition you want.

Tips for Better Results

• Include Sirius in your prompt to activate the LoRA.

• Describe the outfit, pose, expression and background rather than relying on the trigger word alone.

• If the character's appearance starts drifting, increase the LoRA strength slightly or simplify the prompt.

• If the output becomes repetitive or the character features look exaggerated, lower the strength.

• Use the same HassakuXLIllustrious v22 checkpoint used during training for the most consistent results.

The LoRA is intended for character illustrations, portraits, cosplay, alternate outfits and different visual settings. It may take some prompt tuning, but you can get some really nice results with a little experimentation.

Have fun generating!