Jimeng Image Editing Coach

What problem it solves

Jimeng image editing can easily turn a local fix into a regeneration: facial features, product structure, logos, text, lighting, or perspective change unintentionally, and failures are hard to trace. This skill treats image editing as a verifiable workflow rather than prompting vague terms like “more premium” or “more natural.”

How it works

It builds an editing task around preserve + modify + blend + exclude:
- Lock invariants: person identity, face features, product shape, logo, text, camera angle, pose, and brand colors.
- Iterate on one variable: each round targets one goal, such as removing background people, changing the sky, extending the frame, replacing clothing color, or fixing hands.
- Choose the right mode: inpainting, outpainting, reference-image editing, or global style adjustment, depending on the risk surface.
- Constrain masks and prompts: keep masks slightly larger than the target for blending, but shrink them near faces, hands, logos, and text; use references only for specified attributes like pose, clothing, color, material, or scene.
- Version and verify: back up the original, preserve uncompressed dimensions, record prompts and parameters, then check subject, blending, and output at the pixel level.

Boundaries

It fits iterative image edits that require rollback and acceptance checks, not one-shot creative generation. Multi-variable changes, large outpainting, hair, glass, or reflections still need separate passes and manual review of text, privacy, and watermarks.