One-Shot Character Customization Pipeline

Overview

Built a complete one-shot character customization pipeline for Nomi by Toyumi, a London-based startup building gamified self-help products for Gen Z. The pipeline enables rapid character generation through custom LoRA models trained on Flux Dev, integrated into a ComfyUI-based production workflow.

Scope of Work

Custom LoRA Training — We trained character-specific LoRA models optimized for one-shot customization, enabling users to generate consistent character representations from minimal input. The models achieve high fidelity while maintaining the stylistic flexibility required for a consumer-facing product.

Pipeline Architecture — We designed and built the end-to-end workflow from training through inference, with a focus on reliability and ease of integration into Nomi’s existing product infrastructure.

Model Selection & Strategy — Guided the Nomi team through a strategic pivot to Flux Dev as their backbone model, replacing a less capable prior solution. This included evaluating backend customization options, benchmarking model performance, and walking the team through every stage of the selection and integration process.

Client-Facing Controls — Collaborated on the development of user-facing generation controls, translating the technical capabilities of the underlying models into intuitive product interactions. Bridged the gap between model capability and product design.

Delivered With

Developed in collaboration with Araminta K through Promptcrafted LLC.