SuperAPI AI Directory

jianying-draft-filters

Design & creation

View repository

Updated 10-27 10:06

About this skill

JianYing Filters

Used for the CapCut "Filters" capability. A filter expresses an overall color style; individual parameters such as exposure and color temperature belong to adjustment skills and cannot impersonate one another.

When to use

  • Search and filter style filters, previewing different intensities.
  • Apply, adjust, replace, or delete filters on clips or explicit time ranges.
  • Verify filter consistency across shots and its impact on skin tones.

Capability boundaries

  • ✅ Can do: filter catalog, candidate preview, explicit range/intensity application, replacement, inspection, and deletion.
  • ⚠️ Requires conditions: official filters require entitlements; brand colors/skin tones require references; completion requires playback comparison.
  • ❌ Should not use: not a substitute for fine exposure/white balance adjustment, no default whole-video coverage, and do not judge final color from thumbnails.

Quick start

You can say directly: "Find a clear filter for the seaside shots", "Lower the filter intensity to 35", "Only remove the retro filter from 5–8 seconds". On first use, load capability workflow。

Run contract

  • Trigger conditions: the user requests filter search, preview, application, intensity adjustment, replacement, or deletion.
  • Required capability:media.catalogue、media.enums、timeline.filter、media.official_resource_apply、timeline.remove。
  • Minimum CLI version:1.6.19; fixed-catalog filters go through Rust commands, and online resources stop at the Adapter gate when unsupported.
  • Default risk:reversible_write。
  • Input facts: filter slug/resource ID, target clip or range, intensity, visual style, skin tone/brand color protection, and entitlements.
  • Confirmation points: confirm separately for downloading restricted resources, whole-video application, overriding already color-graded clips, and native playback.
  • Success evidence:playback; structural existence cannot prove color perception.
  • Prohibited actions: must not use filters to simulate unsupported fine exposure/white balance adjustment, must not default to whole-video coverage, and must not infer final color from thumbnails.

Workflow

Step 1 — Filter candidates

Filter candidates based on the creative brief, shot lighting, and entitlements.

Step 2 — Preview

Preview candidates and intensity; obtain a receipt for official resources first.

Step 3 — Apply

Follow the capability workflow to execute on an isolated copy within an explicit range.

Step 4 — Verify

Read back the filter material, intensity, and range, and compare skin tones, brand colors, and shot consistency via playback.

Step 5 — Restore

If it fails, delete the target filter segment or restore the snapshot.

Rules

  • Filters and adjustments are modeled separately and do not substitute for each other.
  • Each application binds slug, intensity, and explicit range.
  • Choose to satisfy entitlements first, then compare styles.
  • Skin tone and brand color protection take priority over "more atmospheric".

Gotchas

  1. Filter replacing white balance: switch to the adjustment skill when color temperature/tint is needed.
  2. Default whole-video coverage: if the range is missing, request it first or generate a review plan.
  3. Confusing intensity scale: use the CLI/catalog agreed scale and do not normalize on your own.
  4. Different shot lighting: compare per shot and do not blindly unify intensity.
  5. Looking only at thumbnails: the final judgment is based on playback and reference frames.

Safety and data

The filter pipeline processes locally by default and does not upload frames; online resources only store resource identity and minimal rights evidence.

See the minimal Job at examples/minimal-job.json。