name: photo-abstract-editorial description: Use when one or more photographs must become adaptive photo-plus-abstraction editorial compositions while source facts, spatial relationships, and strict photo preservation remain controlled.

Photo Abstract Editorial

Create a lightweight editorial composition from a supplied photograph. The photograph remains the sole factual source; generated content is an abstract motif, never a replacement for the final photo region or exact title.

Capability routing

Choose the highest-fidelity mode the host can actually execute:

• Strict Fidelity: host visual-understanding capability, host image-generation capability, local filesystem access, local Python/script execution, Pillow-compatible image processing, and a usable serif font. This mode deterministically assembles photo, panel, motif, and typography, then runs machine validation.
• Native Image Edit: host visual-understanding capability plus host image-editing/composition capability, without the complete local Strict toolchain. Use the source as the edit/reference image and disclose that pixel-exact, deterministic, machine-verified photo preservation is unavailable.
• Reference Generation: host visual-understanding capability, host image-generation capability, and reference-image support. Use the source as the sole factual reference, perform best-effort fidelity, and disclose the same Strict guarantees are unavailable.

If visual understanding or every image-generation/editing capability is absent, stop rather than fabricate a result.

Workflow

1. Validate the required image input and inspect the source.
2. Classify the scene and resolve the four controls in control-system.md using scene-profiles.md.
3. Select a layout from layout-profiles.md using scene facts, orientation, dominant axis, subject location, center of gravity, and negative space. Omitted layout falls back to Lower Editorial.
4. Read art-direction.md. Generate or edit only what the selected mode permits; keep the motif sparse, traceable, text-free, and source-derived. When the source supports a decisive vertical structure, use one primary and one subordinate irregular vertical stroke, unequal in height and width.
5. In Strict Fidelity, run remove_chroma_key.py, compose_editorial.py, and validate_editorial.py. Never overwrite source, motif, output, or manifest.
6. Run the structured eight-dimension QA in quality-check.md. Strict machine validation is necessary but not sufficient; visual/editorial QA is also required.
7. Apply at most one targeted correction. Re-run the relevant generation/composition, machine validation, and QA; stop after the second attempt if a critical HARD FAIL remains.
8. For an explicitly requested multi-image series only, load series-style.md and lock the visual family while re-analyzing each new source.

The validated example pair demonstrates package structure and regression expectations; do not copy its subject, palette, title, or coordinates.

Return the final artifact, saved path, selected mode, and concise QA result. Do not claim Strict guarantees for Native Image Edit or Reference Generation.

🎨 Photo Abstract Editorial

Turn one photograph into an editorial composition built from the original photo, an abstract memory panel, and a poetic English title.

English · 简体中文 · 繁體中文

This project keeps the source photograph factual, uses a restrained abstraction panel to carry visual memory, and renders the editorial title locally when Strict Fidelity is available.

One project, two editions. You choose.

🧭 Choose Your Edition

Original Edition

Choose Original when you want the historical fixed workflow, already use Codex, and prefer the simpler original art-direction contract.

V3 Adaptive Edition

Choose V3 when you need scene adaptation, four creative controls, automatic layout selection, structured QA, bounded correction, or a consistent multi-image visual family.

🧭 Which Edition Should I Choose?

• Want the original fixed visual behavior? Choose Original Edition.
• Want scene adaptation, controls, layout profiles, or series consistency? Choose V3 Adaptive Edition.
• Using a non-Codex Agent? Original is not supported. V3 may be suitable when the host provides the required image and local-processing capabilities. The tested DeepSeek Harness capability path is independently verified; this does not imply that every Agent exposes the same native tools.
• Want the smallest and most historically faithful package? Choose Original.
• Want a more explicit contract around capabilities and verification? Choose V3.

🔍 Original vs V3 Comparison

🎯 Scope, Limitations & Guarantees

Original Edition

Best suited for one photograph, one text-free motif, and one fixed editorial composition in Codex. It provides deterministic local assembly and the Original validator's source/panel checks. It does not provide the V3 control system, scene profiles, canonical auto-layout, Series Style Lock, or cross-Agent contract. It is not supported outside Codex.

V3 Adaptive Edition

Best suited for photographs that benefit from explicit scene reasoning, identity and spatial controls, adaptive layouts, or a series-level visual family. V3 does not invent unsupported people, buildings, objects, geography, text, logos, watermarks, swatches, or decorative facts. Image generation can vary across hosts and runs; current public evidence includes Codex Strict Fidelity and the independently tested DeepSeek Harness capability path.

Guarantee Matrix

Native Image Edit and Reference Generation must never be described as machine-verified Strict output.

💡 What Problem Does This Project Solve?

Generic image prompts may redraw the source photograph, lose subject identity, invent unsupported facts, produce inconsistent abstraction levels, or render unreliable typography. A fixed layout may also fit one photograph but not another.

Photo Abstract Editorial separates the factual photograph from the generated abstract motif, makes layout and creative decisions explicit, renders typography locally when Strict Fidelity is available, and adds a quality gate before delivery. Original keeps the first stable workflow available; V3 makes the adaptive decisions more explicit.

✨ Key Features

Original Edition

• One source photograph and one sparse text-free motif.
• Source-derived bands, gaps, offsets, and unequal vertical anchors when supported by the source.
• Warm-ivory panel, muted palette, exact English title, and optical spacing.
• Deterministic local composition and JSON manifest.
• Original source/output collision guards and pixel-level photo-region validation.
• Machine validation plus visual QA.

V3 Adaptive Edition

• Capability-based routing across Strict Fidelity, Native Image Edit, and Reference Generation.
• Four Creative Controls: Abstraction, Creative Freedom, Identity Preservation, Spatial Fidelity.
• Seven Scene Profiles and portrait-aware abstraction.
• Five deterministic Layout Profiles.
• Structured eight-dimension QA and bounded targeted retry.
• Optional Series Style Lock process.
• Portable chroma cleanup, exact local typography, manifest, machine validator, and reproducible package.

🧩 V3 Execution Modes

Strict Fidelity

When the host has visual understanding, image generation, local file access, Python, Pillow-compatible processing, and a usable serif font, V3 can provide deterministic composition, exact local typography, a manifest, a machine validator, and pixel-exact photographic-region verification when no resizing is requested.

Native Image Edit

Uses the source as the edit/reference image and follows the same art-direction and QA contract where the host supports it. It is best effort: no pixel-exact, deterministic, exact-local-typography, or machine-verified Strict guarantee is made.

Reference Generation

Uses the source as the sole factual reference for a model-dependent generation path. It is best effort and does not provide Strict Fidelity guarantees.

📋 Requirements

Original Edition

Codex only. The Original workflow requires Codex visual inspection, the built-in image-generation path, the Codex works