Complete WPF UI Redesign + Animation frame editor + quest NPC Conversation preview + Spine-runtime 4.1 + major MapleLib Performance Overhaul

[lastbattle]

This release is unofficially sponsored by OpenAI for the ultra generous Codex rate limit reset the past 2 weeks I guess? 🤔 all of the AI tokens gonna go somewhere before its gone!
please please, its too much resets we cant take it anymore. Mr. Tibo President, its too much.
btw gpt-5.6-sol rocks! <3 Sam Bankman Altman

so is this the start of AI generated slop in MapleStory game assets?

---

Changes covered here are from 9.0.0 to master as of July 20, 2026.

TL;DR

• Full WPF Modern UI redesign for both HaCreator and HaRepacker, built on a shared design system:

  1. Native tree interactions, modern controls, and consistent desktop UX.
  2. Map explorer build within HaCreator's main editor page
  3. Live map preview during editing
  4. Full localisation for Japanese, Korean, Chinese (Simplified), and Chinese (Traditional)
     

• Animation Frame Editor & AI image asset generator in HaCreator

  1. Full frame-level editing with precise alignment controls, bitmap resizing, timeline image clipboard/drop support, compact preview toolbar, etc.
  2. Edit Mobs, NPCs, Equipments, Items, Reactors, Map Background, Map objects. (greatly recommend gpt-image-2 model using the default configuration)

• Quest Editor redesign with NPC Conversation Preview Studio:

  1. Live/editable conversation rendering (including stop-quest previews)
  2. full IMG selector support in the Quest Editor.

• Spine 4.1 support: raw skeleton loading, atlas handling, and rendering of Spine 4.1 map objects/backgrounds (GMS v230++). with spine-2.1.25 backwards compatibility for versions from 2015-2024 (Ursus, Kinesis)

• MapleLib performance & format work: major optimizations (.ms pack v4 encryption ChaCha20 support, WZ core parsing/lookup/save, PNG packed formats with SIMD/scalar paths, BC7 / format 4098 image support, concurrent loading, legacy IMG encryption fallback, etc.).

• AI map editing: OpenCode transport have been replaced with HaCreator native MCP, using openai's API format (supports openrouter)

• HaRepacker: WPF redesign, WZ key brute-force optimization (80,000x speedups, p.s. custom WZ keys dont work to secure your wzs anymore), unload tree synchronization fix.

• Cleanup: removed unused Squish dependency; MapleLib tests moved into the submodule; docs/AGENTS.md updates.

• Note: Some MapSimulator-related work remaining at staging was intentionally isolated via staging-no-mapsimulator branch, as it is not ready for production.

  

---

Highlights

HaCreator

• [NEW] Complete WPF UI redesign and localization
  1c0efcc c43b7b9

• [NEW] Animation Frame Editor (frame alignment, bitmap resizing, timeline clipboard/drop, compact preview toolbar, coverage)
  2c14372 e8f2f20 e7c0cda eabe48a 820b9eb 77692f6

• [NEW] NPC Conversation Preview Studio + editable rendered conversations + stop-quest preview
  6033817 53caba0 28239ef

• [NEW] Redesign Quest Editor and support IMG selectors
  7ac3107

• [NEW] Spine 4.1 raw skeleton / map object / background support
  39e133f 48a9de4 25a8ca3

• [NEW] Replace OpenCode AI transport with MCP
  050963f

• Map Explorer / loading UX improvements
  03d19b5 6500fc7 a203d31 600b8cc bc38b18 a85c3c4 acb2256 15f9d15 d7c93d5 4443e52 ce0decb b68cc1f 3b5310b 06f0545 b8d17aa ceeba5c 7435d33 ef1ea9c 7fd1290

HaRepacker

• [NEW] Complete WPF UI redesign and native tree interactions
  2cbdc75

• WZ key bruteforce performance 3a10875 Benchmark report

  • Reworked the 32-bit WZ IV key search used by WzKeyBruteforceForm with a preloaded encrypted-suffix probe, batched AES processing, reusable worker buffers, and parallel candidate scanning.

  • The implementation preserves the complete 0 through uint.MaxValue search space, four-byte key representation, cancellation, progress reporting, and final validation through the existing WZ parser.

  Single-worker benchmark

  Metric	Before	After	Improvement
  Time per candidate	239,282.059 ns	2.578 ns	92,817× faster
  Throughput	~4,179 candidates/s	~387.9 million candidates/s	92,817× higher
  Allocation per candidate	17,664 bytes	0 bytes	Zero steady-state allocation

  Parallel benchmark

  Workers	Aggregate time	Throughput
  1	2.7995 ns/candidate	~357 million candidates/s
  8	0.5513 ns/candidate	~1.81 billion candidates/s
  48	0.3415 ns/candidate	~2.93 billion candidates/s

  At the final 48-worker result, the signature-filter portion of the complete 32-bit key space takes approximately 1.47 seconds in the benchmark environment. Thread scheduling and the small number of candidates requiring full parser validation add some additional overhead.

  Each worker retains two reusable 256 KiB buffers at the selected 16K batch size. A 48-worker search therefore uses approximately 24 MiB of reusable buffers instead of allocating roughly 17.25 KiB for every candidate.

  • Additional improvements

  1. Check common zero, GMS, and MSEA keys before beginning the exhaustive search.
  2. Preload the encrypted .img suffix for fast candidate rejection.
  3. Process 16,384 IV candidates per AES batch.
  4. Run the full WZ parser only for signature-filter matches.
  5. Scale the search across Environment.ProcessorCount * 3 workers.
  6. Use the platform-accelerated implementation supplied by Aes.Create().

  Return a specific error for unsupported WZ files that cannot use the optimized search path.

• Fix WZ unload tree synchronization
  719cdb7

MapleLib / WzLib (submodule)

Major updates pulled via multiple submodule bumps (key MapleLib commits):

• Add MS pack version 4 support (ChaCha20) + parsing hot-path optimizations + benchmarks
  123a54 6510d3 1a6458 2a65e8

• Optimize WZ core parsing, lookup, and save paths
  0beb29

• PNG packed formats + WZ PNG/canvas hot paths 37d4e6da4db3
  Optimized packed 16-bit PNG formats with new SIMD and scalar paths, while reducing CPU and allocation overhead across WZ canvas lookup, PNG decompression, and image encode/decode operations.
  Performance benchmarks — median Release measurements

  Operation	Before	After	Improvement
  Format 1 encode — SIMD	9.466 ms	2.448 ms	74.1% faster
  Format 1 encode — scalar	9.288 ms	4.064 ms	56.2% faster
  Format 257 encode — SIMD	9.297 ms	2.463 ms	73.5% faster
  Format 257 encode — scalar	9.636 ms	2.636 ms	72.6% faster
  Format 513 encode — scalar	10.477 ms	2.621 ms	75.0% faster
  Format 517 decode — SIMD	0.075 ms	0.022 ms	70.7% faster
  DXT3 decode — SIMD	1.091 ms	0.656 ms	39.9% faster
  The Format 513 SIMD benchmark workload was corrected during development, so only its directly comparable scalar result is shown.

  Allocation benchmarks

  Workload	Before	After	Reduction
  Canvas lookup	~13.036 MB	~0.001 MB	>99.99%
  Live 40-image decode	~507.47 MB	~33.05 MB	93.5%
  Format 1 decode	3.767 MB	1.266 MB	66.4%
  Format 517 decode	1.272 MB	0.021 MB	98.3%
  DXT3 decode	~5.445 MB	~0.830 MB	84.8%
  BC7 decode	3.142 MB	0.641 MB	79.6%
  DXT3 encode	6.582 MB	0.650 MB	90.1%
  DXT5 encode	11.244 MB	0.655 MB	94.2%

  Additional changes include direct decoding into locked bitmap memory, reusable per-thread DXT buffers, exact integer color quantization, complete deflate reads, reduced compressed-stream copying, and corrected DXT3/DXT5 buffer sizing for non-block-aligned dimensions.

• Support BC7 WZ canvas format 4098 (FieldID: 993296000, Map/Back/2603nightmareForest.img/spine/nightmare2.png)
  08beb8

• Add Spine 4.1 raw skeleton support + atlas improvements (FieldID: 993296000), with backwards compatibility to spine-2.1.25 since Ursus and Kinesis patch (2015~2024)
  ea3dc5 0eb732 e5be64

• Stabilize concurrent WZ image loading
  c41dd9

• Support legacy IMG encryption fallback
  ada8a8

• Fix canvas shard path resolution, segmented canvas link resolution, IMG path traversal, IMG serialization for standalone images, RGB565 stride (thanks @sidyhe Fix RGB565 canvas decoding stride handling for Format513 images MapleLib#20) , empty paths, fixed-length string reads, keep reader position after parse, etc.
  115857 d86a65 f89dd9 50d67b
  7440b2 37ddb4 d0e6a1 a4b817

• Migrate JSON serialization to System.Text.Json (3x faster than Newtonsoft.Json in .NET Core 3.0 benchmarks .NET dev blog)
  75e0fc

• Support IMG version folder rename
  50332d

• Fix byte reading/writing to use the correct size in SaveImage() (thanks @genment Fix byte reading/writing to use the correct size in SaveImage() MapleLib#18)

• Fix v229.2 WZ file parsing for 64-bit format (thanks @ryantpayton Fix v229.2 WZ file parsing for 64-bit format MapleLib#17)

Other / Repository

• Document the unified desktop UI design system
  b29c4fc

• Remove unused Squish dependency
  1de4bc5

• Move MapleLib tests into MapleLib submodule + add WZ core / packed PNG regression tests
  81173f9 fb87aa8

• Multiple MapleLib submodule updates
  0567359 a16f29b cf4e3a1 356b4cc 8d4a53a f585c0a 6a5f150 9992bea dd342bc

• AGENTS.md / commit guidance updates
  f7e78e2 63de4c5 3a79995

• Update IMG_FILESYSTEM_MIGRATION_PLAN.md
  2a98eb5

• Prepare release branch excluding MapSimulator changes + related merges
  b9ea573 cd853ae 0a5e636 21593f4

---

Full Changelog: 9.0.0...10.0.0