Skip to main content

Performance & Build Tips

Strategies to reduce generation time and bundle size.

Use Minimal Mode for Fast Iteration​

mode: "minimal" prunes deep nested inputs and disables select/include.

Targeted Model Generation​

Models are enabled by default — adding an entry under models does not restrict generation to just that entry. To skip a model you have to disable it explicitly:

zod-generator.config.json
{
  "models": {
    "AuditLog": { "enabled": false },
    "LegacyImport": { "enabled": false }
  }
}

Use per-model operations to trim the CRUD surface of the models you do keep.

Disable Unused Categories​

Turn off emit.crud, emit.results, or emit.variants when not needed.

Single File for Deployment​

useMultipleFiles: false produces one file—ideal for serverless bundling. Since v2.1.5 the shared literalSchema / jsonSchema helpers are hoisted and deduplicated instead of being repeated per schema, so the bundled file is smaller than on earlier releases.

Lean Pure Models​

pureModelsLean is already true by default, so pure models are emitted without the JSDoc/stat comment blocks. Only revisit this if you have set it to false.

Relations are likewise excluded by default (pureModelsIncludeRelations: false). If you opted in, trim expensive relations with globalExclusions.pure or pureModelsExcludeCircularRelations: true.

Avoid Enum Explosion​

Exclude enums or limit variants if you have large enum sets.

CI Pipelines​

Generation is a single prisma generate step, so the wins in CI come from not repeating it: cache node_modules and the generated output directory, and run prisma generate once in a prebuild step rather than per test shard.

Since v2.1.5 the generator reuses the DMMF that Prisma hands it (options.dmmf) instead of re-parsing schema.prisma with @prisma/internals, which removes a redundant full-schema parse from every run.