Image LoRAs

Small adapters that teach an image model specific aircraft and set pieces it can't draw on its own.

Role
Creator
Areas
AI, VFX, Tools
A generated C-123 Provider in green and tan camouflage parked on a concrete apron, seen from the side, with AH on its tail.

A LoRA (low-rank adaptation) is a small add-on to an image model. It teaches the model one subject it otherwise draws badly, without retraining the model itself. We trained these seven for SDXL 1.0 with a custom variant of Kohya_ss and published them free on Civitai.

The subjects have something in common: a general model has seen too few clear pictures of them to get them right. It knows "military cargo plane" but not the shape of a C-123. It knows "motorcycle" but not the particular bike a show's audience would recognize at a glance. Previsualization, concept art and pitch imagery often need exactly that kind of specific object, so each LoRA closes one of those gaps.

Each one is called up by a trigger word in the prompt:

LoRA Trigger word What it adds
Daryl Dixon's motorcycle TWDDDBike The customized Honda CB750 from The Walking Dead
C-130 Hercules AIR_C130 The four-engine Lockheed transport, in flight and on the ground
Alexandria wall TWDAlexandriaWall The sheet-metal walls of the Alexandria Safe-Zone
C-123 Provider AIR_C123 The twin-engine Fairchild transport, in camouflage and bare metal

Likeness studies

Three more were an experiment in reproducing real people's likenesses: characters from The Walking Dead, as played by the actors who portray them. They are for education only, not for commercial use and not meant for production. Each page carries a takedown contact for the actors' representatives.

LoRA Trigger word Played by
Daryl Dixon TWDDaryl Norman Reedus
Negan Smith TWDNegan Jeffrey Dean Morgan
Carol Peletier TWDCarol Melissa McBride

Every example image on these pages was generated with the LoRA it illustrates.