# Recipe mapping gaps

Lane F18 closes the creator-template and advertised-domain gaps that can be
expressed as a finite flow graph. The shipped registry now contains every
linear/looped Asset UI pipeline shape and at least one template for every
domain advertised by `libs/ai/hub/registry.json`.

## Filled in F18

- Structure-guided image generation is represented by
  `image-control.splash` (DA3 depth → FLUX depth control) and
  `image-control-canny.splash` (a supplied image → FLUX Canny control).
- Video post-processing has separate intent-level templates:
  `video-enhance.splash` uses the creator defaults (`upscale=2`,
  `interpolate=2`, `flow_map=true`), while `video-tween.splash` asks only for
  interpolation (`upscale=1`, `interpolate=2`, `flow_map=false`). Both route
  through the real `enhance` domain; no fictitious tween domain was added.
- The creator's direct, expanded, keyframed, looped, edit, sprite, mesh, PBR,
  splat, world, and playable-character pipeline variants now have templates.
  Image stages use 512×512 and model-default steps, video stages use
  640×352/39 frames/30 steps, music uses 180 seconds, and the quality-model
  pins match the preset definitions.
- The previously uncovered advertised domains now have typed prototypes and
  templates: `body` (`body-pose`), `segment` (`segment` and
  `prompt-to-segment`), `stt` (`speech-to-text`), `beats` (`audio-beats`),
  `stems` (`audio-stems`), and `notes` (`audio-notes`).
- Completed values use `Publish` whenever the asset-library type supports
  them. Companion opaque values such as MIDI and a stem bundle remain typed
  `Output` values because `Publish` deliberately does not accept `bytes`.

## Remaining language gap

The two Asset UI fan-out presets—`fleet images → choose → video` and
`expand → fleet images → choose → video`—remain the sole creator pipeline
shape without an exact template. They require a language-level `Map`/fan-out
operator plus a run-time choice gate over the generated candidates. A linear
template would hide the selection step and would not be parity, so the
closest reusable pieces remain `prompt-to-video-keyframe.splash` and
`expanded-prompt-to-video.splash` after a candidate has been selected.

## Intentional semantic adaptations

- There is no `image.enhance` kind. The creator's sprite-enhance pipeline is
  an `edit` request pinned to `flux2-dev`, so `sprite-enhance.splash` uses
  `ImageEdit` rather than inventing a routing domain.
- OCR is an advertised hub domain but not an Asset UI generation kind.
  `ocr.splash` therefore keeps the inline `Gen{domain: "ocr"}` idiom and
  carries the backend's HTML result as typed `text`.
- `text.expand` is a queued hub domain, while authored recipes intentionally
  use `Llm` so the system instruction is visible and editable. This does not
  expose the creator job's `target_domain`, `identity_anchor`, `style`, and
  `variants` fields as a dedicated node.
- `annotate.asset` mutates an existing asset annotation record rather than
  publishing a normal artifact. `annotate.splash` performs the value-producing
  portion (vision plus strict JSON parsing); the store mutation remains an
  application operation, not a hidden graph side effect.

## Parameter and post-processing boundaries

- The base `Image` prototype remains 1024×1024/8 steps for language
  compatibility. Creator-parity templates explicitly select 512×512 and omit
  `steps`, preserving the creator's model-default setting.
- `Upscale.factor=4` is translated to the hub request's `upscale` field, and
  `Image.negative` is translated to `negative_prompt` by the executor.
- Paired image dimensions and video `(frames, steps)` choices are coupled UI
  presets. A generic inspector must not turn them into a Cartesian product.
- Resize, thumbnail derivation, GLB inspection, media probing, catalog
  dressing, provenance, and rights processing remain importer/store work.
  Templates make generation and publication explicit without pretending those
  application steps are hub domains.