Jingzao Image Forge v1.11.0

Adds Lilac Material Reveal / 紫序·产品微观, an adopted, opt-in style package for product-led imagery: lilac-white studio depth, controlled shell reflections, distinct dry-pigment/fiber materials, purposeful product arrays and element effects with a visible origin or destination.

The guide distills documented product-research and targeted-repair methods without distributing client reference images, brands, private paths or full source prompts. Product geometry, actual material, color and identity remain target-owned; powder effects are conditional rather than a universal style accessory.

Five generation/refinement records cover artist pastels, an electronics array and fiber contact. The refined pastel and wide speaker support two different product mechanisms; the macro supports contact-phase appearance. The initial coarse-powder image remains partial and some coarse material persists after refinement. Initial 3:2 outputs and the later explicit 16:9 prompt suffixes are preserved in the evidence record rather than relabeled as exact-format successes. Four new unbranded example images are included in the README/style guide.

Validation: 255 tests, full local workflow, four-platform target/material preservation, capsule validation, reproducible sent-prompt hashes, published PNG hashes/dimensions, independent visual/package review, PR/main/tag CI, and fresh-process installed-compiler readback. All 176 tracked files and executable modes match the merged commit and ZIP.

The package does not establish physical particle sizes, animation continuity, source-project client approval or universal repeatability. The native-tool handoff note clarifies that standalone compiler parameters must actually be forwarded to prompt-only image tools.

Jingzao Image Forge v1.10.0

Reference distillation now retains selected hair, makeup, expression, gaze, pose, action, body, wardrobe and photography instead of reducing the reference to color and material. The optional --reference-profile path separates observed evidence, target choices and positive generation prose, with a facet-level audit kept outside the prompt.

• Incorporates DIR's evidence-to-controls/review method without requiring or modifying the upstream DIR executor.
• Includes four source-profile branches and text-only refinement examples. Unknown source regions cannot be projected as preserved or adapted facts; target-only original design stays in the base specification.
• Preserves existing specs and style-capsule behavior; profile and capsule flags are mutually exclusive.
• Rebuilds the English/Chinese README landing pages, guide, AI-readable llms.txt map and repository discovery metadata, with newly generated introduction artwork and a current public sample. Older graphics remain clearly historical.

Validation: 251 tests, the full local workflow, PR/main/tag CI, independent code and visual review, live README checks, and a fresh-process installed-compiler smoke test. All 164 tracked files and executable modes match the merged release commit and ZIP.

Four female profile-generation calls returned images. Attached-reference trials retained their main traits; the text-only refinement improved finger placement and ribbons but still varied in fringe/material and cropped the hat top. A separate window-side body/hand trial was blocked at input moderation and returned no image. These are different inputs from the v1.9 failures: no moderation root-cause fix, universal acceptance, full-body validation or independent identity guarantee is claimed. The older generic capsule remains draft.

The ZIP contains the jingzao-image-forge/ Skill folder. User source images and source-attached portrait outputs are not distributed. One original text-only sample and the two new introduction layouts are included.

Jingzao Image Forge v1.9.0

Adds the optional Soft Editorial Character / 柔映·人物质感 family, with daylight lifestyle, cream-rose couture and crimson-black couture branches. Person-style distillation now records a body-profile form and carries target-adopted physique, garment fit, support and coverage into existing specification fields and visual review.

• Four supplied references were inspected; shared mechanisms are separated from branch lighting, palette and source-specific content. Source images and private paths are excluded.
• Includes three reproducible cross-subject prompts, a reusable body form, branch guidance and explicit unknown-region handling.
• The new capsule remains draft / partially verified. One mature male portrait returned with material/light transfer but skin, garment closure and aspect-ratio deviations. Two female portrait tests were blocked at the image tool's output stage and returned no image. The release does not claim a successful reference match or validated full-body rendering.
• Includes the previously local garment-input fidelity update: preserve tested text plus its actual reference bundle, and keep user-reported success distinct from local outcomes.

Validation: 233 deterministic regression tests; local CI workflow; four-platform target/body/coverage preservation; reproducible sent-prompt hashes; structural Skill validation; independent visual review and fresh-process instruction routing/compilation. Global install and extracted payload are checked against the release commit.

The ZIP contains the jingzao-image-forge/ Skill directory. Original reference pixels, generated test pixels and local work records are not distributed. This release updates Skill instructions and style resources, not model weights.

Jingzao Image Forge v1.8.0

Re-distills the Changsheng wardrobe style into a character-led design system rather than a fixed dark/gauze/gold recipe.

• Resurveyed 149 original-resolution film samples, with 31 native-frame close inspections; observations, design inferences and unknown manufacturing details remain separate.
• Outfit silhouette, palette, pattern coverage, material behavior and dressing now follow character, role and occasion. Plain and rich garments both retain tangible construction quality.
• Six locally reviewed image tests cover four contrasting characters, one changed occasion with a single visible garment, and one dialogue close-up retaining rich textile detail. All passed independent visual review with recorded limitations.
• The v1.7.0 capsule is frozen as a historical test fixture. Existing generation prompt hashes remain unchanged; new text-only tests bind to the revised capsule.

Validation: 227 deterministic tests, four-platform target-preservation checks, source/spec/capsule validation, linked local reviews and reproducible text-only prompt hashes. Source and newly generated pixels are not included in the release. Exact manufacturing, user acceptance, deterministic continuity and video quality are not claimed.

The attached ZIP contains the jingzao-image-forge/ Skill folder and is checked against the release tag after extraction.

Jingzao Image Forge v1.7.0

Adds two opt-in, reference-learned style packages to Jingzao Image Forge:

• 长生·华构人间 / Inhabited Chinese Fantasy — inhabited grand architecture, role-based costume/hair/makeup, varied narrative camera positions, and separately labeled original magic-combat studies.
• 晴海·角色写真 / Azure Summer Character Photography — airy daylight, canopy shade, sunset and close interaction branches, including portrait/widescreen/ultrawide restaging and explicit style ownership for architectural combinations.
• Adult body and garment fit — keep requested physique, garment structure, coverage and mood separate; use fitting and fabric controls without silently flattening or exaggerating bodies.

Validation: 225 deterministic regressions, four-platform compile contracts, linked local-review cases and reproducible current text-only prompt hashes. Original film/reference pixels and newly generated test pixels are not distributed. Local visual studies record partial results such as an incomplete stone bind and neckline drift during an edit; no model training, exact identity/geometry, video-quality, A/B causal or acceptance-rate guarantee is claimed.

Install by extracting the jingzao-image-forge/ folder from the attached ZIP into your Skills directory. The ZIP is built from the release tag and checked with scripts/verify_install.py after extraction.

Jingzao Image Forge v1.6.2

Fixed

• Replaced generic image-creation discovery wording with Jingzao-specific triggers: visual_generation_spec, reference-aware prompt compilation, artifact cleanup, style learning, and multi-frame continuity.
• Aligned the Codex UI short description with the structured-spec and clean-prompt role.
• Renamed the deterministic metadata test so it claims only the observed host length guard; real adoption remains a fresh-task test.

Fresh-task evidence

• v1.6.1 baseline: the unnamed structured cleanup request selected other image Skills and missed Jingzao.
• v1.6.2 candidate at 38d721e: the identical unnamed request selected jingzao-image-forge first, then optional cleanup/material helpers.
• The candidate produced a valid no-reference clean_reset specification; validate_spec, compile_prompt, and prompt_lint passed with prompt_review ready and no generated image.
• Independent cold review marked the prior discovery P1 RESOLVED.

Quality boundary

• No prompt compiler, cleanup, oil-spot/noise, scene-redraw, reference, template, example, or production behavior changed.
• One successful discovery smoke does not prove every natural-language variant.

Verification

• 219 deterministic tests pass.
• Ruff 0.16.3 and Skill structural validation pass.
• Python 3.10 GitHub Actions pass on merged main commit 8e036f0.
• Release ZIP round-trip matches all 96 tracked files and modes.

Artifact

• jingzao-image-forge-v1.6.2.zip
• SHA-256: 8eef296e680ed9c7c681788046034b7629302607f4b9736a2cabd87ffa40c81a

Jingzao Image Forge v1.6.1

Fixed

• Replaced the overlong workflow-style Skill description with a 143-character trigger-only description that survives the current host catalog normalization boundary without an ellipsis.
• Added commit-pinned verification for installed tracked files, symlinks, content, executable modes, missing files, unexpected files, and unsafe symlink traversal.
• Distinguished release parity from working-tree validation and from fresh-task discovery/adoption.

Quality boundary

• No prompt compiler, clean_base, clean_reset, oil-spot/noise cleanup, scene-redraw, template, reference, example, or agent behavior surface changed.
• Empty directories carry no Git payload and are outside install comparison.

Verification

• 219 deterministic tests pass.
• Ruff 0.16.3 and Skill structural validation pass.
• Python 3.10 GitHub Actions pass on merged main commit c773077.
• Release ZIP round-trip matches all 96 tracked files and modes.
• Independent native cold review: READY.

Artifact

• jingzao-image-forge-v1.6.1.zip
• SHA-256: e9f721a7244dcabd3ead3810c184bfe6dfa817dbf0a6da7baffbf3c9ae1960ee

v1.6.0 — Production Coverage and Quality-Preserving Cleanup

Highlights

• Adds an opt-in production manifest: every planned shot names its necessary image anchors or a specific video-only reason. Each complete single-image spec compiles independently, without implicit cross-frame text, subject or reference inheritance.
• Preserves native canvas metadata in full and selected frame envelopes; validates coverage links and real reference requirements. Portable output no longer carries the Codex executable call plan or its stale execution warning, while genuine diagnostics remain intact.
• Improves medium-relative, wet/gloss and bokeh cleanup suggestions without silently rewriting user prompts. Explicit clean-reset wording protects declared medium/surface traits, identity-critical edges and exact text. The ordinary clean base and semantic budgets remain unchanged.
• Completes the evidence-first quality workflow: source/region diagnosis, suitable clean-master selection, bounded retries, and separate surface versus whole-image acceptance. Texture-free layout guides remain a conditional fallback, never a default, pixel lock or guaranteed cure.
• Corrects the legacy documentation claim that an independent_frames styleboard was automatically split into executable image calls.

Verification

• 205 regression tests pass; original 169 tests retained.
• Ruff, Python compilation, Skill validation, specification/capsule/evidence checks, prompt lint, reference preflight and four-platform compilation pass.
• Independent native review and two DeepSeek review rounds completed. Findings were addressed and independently rechecked; mutation controls detect cleanup-boilerplate growth and upstream warning-wording drift.
• Feature branch, main and v1.6.0 tag GitHub Actions all pass on the release commit.
• Fresh archive and exact-commit GitHub installation tests pass. All 94 tracked payload files match source, installed copy and release ZIP; executable flags were also checked.

Boundaries

Private scene experiments showed surface improvement without global blur, but some contact/action constraints still failed. No private images, prompts or project details enter this release, and those experiments are not added to the public gallery or forward-test evidence manifest. They are not a controlled word-only A/B or a universal denoising guarantee. A ready prompt package is not generated media, visual approval, user acceptance or proof of future host adoption.

v1.5.0 — Layered UI Motion

Highlights

• Adds explicit UI Motion hierarchy profiles: minimal_state, layered_editorial, spatial_system, custom, and auto.
• Adds per-frame L0 primary focus, L1 proof, L2 sequence continuity, L3 ambient scaffold, calm-zone, accent-owner, and suppressed-competitor fields.
• Compiles the full hierarchy contract for OpenAI, FLUX, Midjourney, and generic targets.
• Adds duplicate-resistant hierarchy_layer_count to dynamic semantic prompt review so layered sequences receive review space from actual structure rather than frame count alone.
• Defaults production UI Motion with frame-specific typography and density to independent native-ratio frames; direct sheets remain useful for calibration and comparison.
• Upgrades the four-frame example, UI Motion workflow, visual-spec reference, bilingual README, llms.txt, and discovery metadata.

Verification

• 169/169 regression tests passed.
• Ruff 0.16.3, Python compilation, schema validation, prompt lint, reference preflight, four-platform hierarchy compilation, Skill validation, and staged-diff checks passed.
• Independent cold review initially found two release blockers: unhashable hierarchy-profile handling and missing accent-owner enforcement. Both were fixed, covered by regression tests, and the reviewer returned final PASS.
• Feature branch, main, and v1.5.0 tag GitHub Actions passed.
• A private-reference, session-only 16:9 forward test passed the visual hierarchy gate at 92/100. The private reference, prompt, and output are not included in this public release or claimed as public execution evidence.

v1.4.0 — UI Motion Storyboard

Highlights

• Adds a first-class UI Motion storyboard workflow for four-frame interface-state storytelling.
• Preserves exact in-frame Chinese copy and separates viewer-facing UI from production metadata.
• Restricts #FF6A2A to active semantic states, keeps inactive frames charcoal/gray, and distinguishes pointer interaction from hand/finger interaction.
• Excludes visible frame IDs, shot numbers, duration labels, review arrows, and production annotations after a single-variable visual retry exposed unwanted 01–04 labels.
• Adds a validated UI Motion example, dedicated workflow reference, CI coverage, README/README.zh-CN documentation, and 162 regression tests.
• Corrects evidence wording for the existing intro infographic example so local/manual review is not overstated as receipt-bound public execution evidence.

Verification

• Local release gate: 162/162 tests passed; Ruff 0.16.3, schema validation, prompt lint, reference preflight, compilation, and diff checks passed.
• Independent cold review: PASS with no high-confidence findings.
• GitHub Actions: feature branch, main, and tag workflows passed.
• A session-only visual forward test passed after the metadata-exclusion retry; it is not committed or claimed as public execution evidence.