Fallstruck - Project Dashboard

🎯 Larger Goal

Build the creation pipeline: illustrations → 3D → ControlNet → 4K stereoscopic 60fps animation. Train Style + Character LoRAs for consistent, scalable animation production.

⚡ Current Focus

Active

Curate 15-25 illustrations for Style LoRA training

Validate workflow with Style LoRA before Character LoRA

📊 LoRA Training Phases

Phase 1: Style LoRA

In Progress

Capture LD's illustration aesthetic. Train once, use for all characters.

Curate 15-25 images
Set up cloud training
Run training
Test style match

Phase 2: Mada LoRA

Pending

Character LoRA for Mada. Generate new poses without illustration.

Collect 15-20+ Mada images
Generate pipeline renders
Curate final dataset
Train and validate

Phase 3: Additional Characters

Future

Character LoRAs for other Fallstruck figures as needed.

👥 Character Roster

Mada

Core Protagonist

17-year-old Storm Hunter marked by an inverted Touch sense. Survived the green lightning strike that killed her mother. Weighs 337 lbs due to gravity wave storm sickness — doesn't look heavy.

View full profile →

Vendra Vith

Major Supporting Character

19-year-old transporter (alias Lane Ghost) moving untracked goods for the Rainbow Market. Stage 3 sun sickness, helmeted, blunt. Saving for ZT-A zitizenship before the age-20 cutoff. Mada's traveling companion in Chapter 01.

View full profile →

❓ Open Questions

Mada's hunting style — how does she approach storms differently than others?
Shie's relationship — friend, rival, or ally in the capital?
The Shadowy Group — when does their interest in Mada become active?
First Blue Accord infraction — what triggers it?
Thunder jump progression — can storm sickness be controlled?
Zurn State arc — progression, control, or consumption?

✅ Accomplished

Done

Depth smoothing breakthrough — stereo micro-tear elimination validated

Script: tools/depth-estimation/smooth_composite.py

Done

LoRA training plan documented

Two-phase approach: Style LoRA first, Character LoRA second

Done

Sparse animation research completed

Approach A MVP viable after Character LoRA is ready

Done

Cloud GPU investment confirmed

LD willing to pay ~$1-3/LoRA for training

🔜 Next Steps

Pending

Curate Style LoRA dataset (15-25 of LD's best illustrations)

Quality over quantity — must represent consistent aesthetic

Pending

Set up RunPod or Vast.ai account for cloud training

Pending

Run first Style LoRA training