White Album 2 LoRA

A LoRA model trained on characters and scenes from White Album 2, specialized in character voice and visual representation.

Training Details

• Total Steps: 13,000

• Training Duration: 3 days on RTX 4090

• Status: Training halted due to slow progress and high costs

• Dataset: 250 images + 250 videos

• Training Framework: ai-toolkit

Current State

⚠️ Note: This LoRA is still in an underfit state and would benefit from additional training. Due to the high cost of cloud GPU training (3 days on 4090), further training is not planned in the immediate future, though I may continue training later if feasible.

Character Triggers

Required: Always start your prompt with WhiteAlbum2

Character Descriptions

• Setsuna (雪菜): Orange-brown hair, amber eyes. Voice: Soft, melodic, trembling.

• Kazusa (冬馬): Black hair (hime-cut), blue eyes. Voice: Cool, low, intense.

• Haruki (春希): Short dark hair. Voice: Steady, desperate.

Strengths

🎤 Voice Characterization: The model's strongest feature is voice representation. Using prompts like "Haruki says [dialogue]" will capture the specific vocal characteristics of that character.

Usage Tips

⚠️ Detailed prompts are essential. This LoRA requires highly detailed prompts including:

• Character descriptions

• Composition details

• Scene elements

• Voice/dialogue specifications

Vague or minimal prompts will likely produce poor results.

Example Prompt Structure

WhiteAlbum2, Kazusa, black hair with hime-cut, blue eyes, cool expression, [composition details], [scene description], Kazusa says "..."