the language
vision
image inputs for multimodal models. One of 59 words the served contract declares, projected from workflow.schema.json, the same file your editor validates against.
- 1surfaceinfer
- optionaleverywherea choice, never a miss
- arraytypethe declared shape
- schemathe sourcenever prose
the contract
1 declarationDescriptions are the schema's own; the deeper invariants (value languages and regexes) ride the same projection. A miss is a nika check finding before anything runs.
- infer
arrayimage inputs for multimodal modelsinside infer:
in a real file
a crafted check-green fileimage_fx.nika.yaml
review: with: poster: ${{ tasks.stylize.output }} infer: model: ollama/llama3.2-vision prompt: "Poster at ${{ with.poster }} — is the subject still legible after the treatment? One line." vision: - { source: file, path: "./shots/hero-poster.png" } thinking: { enabled: true, budget_tokens: 2000 }a verbatim slice, real line numbers from image_fx.nika.yaml, nika check green; the drift gate re-proves this copy on every test run.
cross-references
no skeleton carries it and no registered code names it; the schema row is the whole story today.
accepted by
defined by
the block it lives in
infer: inside infer:
The whole grammar in one place: the four verbs · the standard library · the spec. Try it in the playground. Read the reference →