Helluvaverse – Multi-Concept LoRA (NoobAI / Illustrious)

A character-focused LoRA trained to encapsulate the Helluvaverse cast in a single place.

This project aims to progressively cover all major Helluvaverse characters under one unified LoRA while keeping prompting friction low.

Current Character Coverage

• Stolas

• Octavia

• Stella

More characters will be added in future dataset expansions.

Training Philosophy

This LoRA was captioned deliberately so that:

• Character name alone should carry most of the identity weight

• Minimal auxiliary tags required

• Cleaner prompts

• Reduced tag clutter

That said:

⚠️ Recommended tag: colored sclera
This helps prevent white sclera on characters without a white sclera, such as Stolas/Octavia/Stella.

Dataset & Training Details

Total Images: 200

• 50 images per character

• 1 multi-character folder

• Each character has its own structured folder

Epoch Structure:

• 200 steps per epoch (1 pass per folder set)

• Uploaded checkpoint: Epoch 38

• Total steps: 7,600 steps

Optimizer: Adafactor
Network Dim: 64
Alpha: 32

NOTE: Avoid excessive style LoRAs stacked at high weights (can distort facial proportions), some styles work, others just don't, at least as of V.1.0.

Feedback on character accuracy, eye fidelity, and silhouette consistency is welcome.