· realm
Local art for unreleased IP
Keeping a concept-art workflow on a studio workstation can reduce some data transfers, but “local” by itself does not establish who can access the files or where copies travel. Treat unreleased art as a project with a defined boundary from intake through review and delivery.
Map every asset and dependency
List the source sketches, character sheets, prompts, model checkpoints, adapters, generated options, review exports and final files. Then name the applications and services that touch each item: image software, plugins, model download sites, sync clients, crash reporting, backups and collaboration tools. A model that runs on a studio GPU can still depend on a cloud account for download or updates, so record those steps separately from generation.
Choose a workspace with access for the people who need to review the work. Keep source art and generated drafts in separate folders, and label outputs so a later editor can tell which model, prompt version and source set produced them. Before the first session, check the model card and the terms for the model, adapter and plugins. A private compute path does not settle rights, licensing or attribution questions.
Make review safer and faster
Writers and art directors often need to compare a few directions, not receive an unrestricted dump of every output. Prepare contact sheets or watermarked review copies with the project name and version. Record what was approved, who approved it and which source files were used. If a reviewer needs to share feedback outside the studio, use the project’s approved review channel and remove access when that round closes.
Before the first unreleased-IP run
- Identify approved source material and any “do not generate” characters or plot details.
- Check that the workstation, model files, caches, logs and backups fit the project’s handling rules.
- Test a non-sensitive sample and inspect network, sync and export behavior.
- Agree on how to report an accidental disclosure and how to stop the workflow.
NIST’s Generative AI Profile offers voluntary risk-management actions for generative systems, including risks around generated content and testing. It is a useful checklist source, not a promise that local inference is secure. For broader rights and community boundaries, see AI and fandom ethics; for character memory, see memory limits. Explore services or contact for a prototype discussion.