Horza

Context for LLMs

Horza is a visual-identity infrastructure project, designated GSV-X2 in the General Systems Ventures portfolio, that generates project cover images by fine-tuning LoRA adapters on curated visual references rather than relying on generic base-model image outputs. Its purpose is to close the gap between generic AI-generated imagery and a project-specific visual identity: by training on a curated reference set per style, it produces cover images that read as stylistically consistent with a project's identity and with each other across the portfolio, rather than defaulting to the recognizable, unbranded look typical of off-the-shelf diffusion output. The concrete mechanism is a FLUX-based image generation pipeline: a working four-stage cover pipeline runs via a `pnpm cover:flux` command that executes a build-flux-cover-photo.ts script and scores candidate outputs, indicating an automated generate-then-select workflow rather than manual curation of single outputs. Architecturally, Horza functions as shared infrastructure rather than a standalone product -- it is explicitly described as gold-canon visual-identity infrastructure for GSV as a whole, meaning its output feeds branding across other portfolio projects rather than serving its own end users. Within the portfolio it is grouped with underground-beets, vibecology, and auxology, and tagged across the intercognition-jazz, media-ecology, life-systems-heart, and infrastructure designations, reflecting its role as a cross-cutting utility rather than a single-category product. Current status as of late May 2026 is in-flight: the four-stage FLUX cover pipeline with scoring is built and working, while the LoRA fine-tuning and broader adapter-training workflow are represented in the project's stated method without further documented detail on training data volume, adapter management, or rollout across the full portfolio.