---
name: sumi-svg
description: Create original layered SVG scenes for Sumi when vector art is requested or image generation is unavailable; inspect a complete proof before a full batch.
---

# Beyondsumi · SVG studio

SVG is the explicit vector choice or disclosed fallback, not a fourth style. Images remain preferred for detailed painted artwork when available and chosen. Explain the simpler painted detail/complex-pose fidelity without claiming SVG is inherently poor or identical to generated paintings.

Read [shared SVG workflow](references/workflow.md) and only the selected guide: [Original](references/styles/sumi-original.md), [Light](references/styles/sumi-light.md) or [Dark](references/styles/sumi-dark.md). Use the current story/direction/runtime and prove one complete contextual scene before a full batch. A requested three-style teaching sample is one composition, not three films.

Use original self-contained paths/gradients/masks and semantic layers; no traced video frames, raster embeds, remote image/font links or scripts/event handlers. Complete subjects, props and support share the planned perspective/light. Filters add surface, not acting drawings. Keep a deterministic output clock and whole pose/action preparation; avoid icons or rigid translations standing in for performance.

Run `python3 scripts/check_svg.py artwork.svg`, then inspect actual render and adjacent movement for complete framing, hands, soles/support, grip and expression. Structure checks do not certify anatomy. `assets/lantern-at-the-river.svg` is a static example; reuse only when its setting suits. Teach labels belong to a comparison/review interface, not the exported film. Retain approved titles and compact story captions, with readable backing and no sample signature/technical text.