Feature request: Production-grade image editing workflow for Spine / Live2D / 2D game assets

Summary

I would like Codex's image editing workflow to support production-grade 2D character asset preparation for Spine, Live2D, sprite animation, and similar pipelines.

Current image editing is already useful for creative work, but it is still difficult to use reliably for character asset decomposition and occlusion-aware repair. This request is about a specific professional workflow with stricter requirements:

• hard-mask behavior
• deterministic preservation of untouched regions
• occlusion completion
• reference-preserving structural edits
• layered RGBA output for downstream rigging/animation tools

Problem

In 2D character production workflows, users often need to:

• remove an occluding object or garment part
• reconstruct the hidden body or clothing behind it
• preserve all unedited regions exactly
• keep transparent background and alpha correct
• keep image dimensions and coordinates unchanged
• optionally use another visible body part or accessory as a structural reference
• export the result as separate RGBA parts or layered output for rigging

Current image editing behavior can be visually impressive, but it is still unreliable for this use case because:

• edits may affect pixels outside the selected region
• transparent edges may shift
• unrelated structures may change
• reconstructed parts may look plausible but are not stable enough for animation assets
• there is no structured layered output for downstream tools like Spine or Live2D

Requested capabilities

1. Hard-mask editing mode

Please provide an editing mode where:

• pixels outside the mask are guaranteed unchanged
• alpha outside the mask is guaranteed unchanged
• image width and height remain unchanged
• output alignment remains unchanged

For unmasked regions, the original image should remain the source of truth.

2. Deterministic compositing

Even if the model internally regenerates more than the masked area, the final output should support deterministic compositing so that:

• masked region = edited content
• unmasked region = exact original pixels

3. Occlusion completion

Please improve or expose a workflow for reconstructing content hidden behind removed foreground elements, for example:

• removing a cape and reconstructing the torso behind it
• removing an arm overlap and reconstructing the body underneath
• removing layered costume elements and restoring the hidden shape

4. Reference-preserving structural edits

Please support stronger reference-based editing where one visible part can guide reconstruction of another part, for example:

• matching the direction or structure of one leg to the other
• matching shoe orientation
• matching armor or clothing style from another visible side
• preserving texture direction, stripes, material identity, and lighting consistency

5. Layered RGBA output

Please support output formats useful for rigging and asset workflows:

• separate RGBA layers or parts
• PSD export or equivalent layered output
• optional semantic part decomposition for characters

Examples of useful part outputs include hair front/back, torso, upper arm, forearm, hand, skirt pieces, thigh, calf, shoe, cape, accessories, and armor pieces.

6. Character-asset / rigging-oriented workflow mode

Please consider a dedicated workflow optimized for:

• Spine
• Live2D
• 2D game assets
• sprite pipelines

This mode should prioritize structure preservation, part completeness, overlap-friendly reconstruction, clean transparency, stable coordinates, and production reliability over freeform creativity.

Suggested acceptance criteria

1. Unmasked regions remain bit-identical to the input image.
2. Alpha outside the mask remains unchanged.
3. Canvas size remains unchanged.
4. Reconstructed masked content blends naturally at the boundary.
5. Reference-guided edits preserve structure and material identity.
6. Optional layered output is available for character parts.
7. Output is suitable for downstream 2D rigging and animation workflows.

Why this is different from generic image editing

This is not just a request for "more fidelity" or "better inpainting". It is a request for a structured production workflow where untouched pixels must remain exact, hidden geometry must be reconstructed, parts may need to be separated into independent RGBA assets, and the result must be useful in animation pipelines rather than only visually plausible as a flat image.

Possible implementation directions could include a true hard-mask mode, deterministic post-compositing, stronger occlusion-aware reconstruction, semantic character part decomposition, layered export, or a dedicated character asset editing mode.

No private images or personal information are included in this request.

No implementation files, tests, or entry points are identified in the issue, so begin by locating Codex's existing image-editing workflow and its mask and export handling. Compare the current behavior with the acceptance criteria, especially bit-identical unmasked pixels, unchanged alpha and canvas dimensions, occlusion completion, and layered RGBA output. Done requires a defined implementation scope and validation for the requested rigging workflows.