From beautiful sculpt to a native weapon

Pinned study: TaCZ 1.1.8-hotfix2 / Forge 1.20.1. Actual released classes inspected; native rendering is still untested.

The Soul Storm Blender studies are valuable editable source, but a .blend or glTF export is not automatically a TaCZ gun model. This page records the actual boundary and the routes worth testing before more sculpting is mistaken for finished implementation.

Existing community route comes first

The direct TacZ Mesh Loader study now pins the actual Forge 1.20.1 TML 0.1.8 artifact and its source. It already supplies polygon-mesh integration, LOD handling and optional accelerated rendering. Prioritize evaluating and improving that existing project before building a parallel renderer. Keep the original bridge prototype below as a diagnostic and possible targeted contribution.

The base-parser findings below concern TaCZ without the additional loader. They do not mean TML's extended polygon path is unavailable. Native compatibility, hand rig, shaders and Hari/Vulkan remain unverified. The direct study includes source-derived validation gates and a license-metadata discrepancy that must not be silently ignored.

What the released files show

• GltfManager loads AnimationStructure from animations/*.gltf. The associated raw structure contains accessors, animations, buffers, buffer views and nodes. This inspected path is for animation, not proof of arbitrary glTF mesh/material import.
• BonesItem stores cuboids. BedrockModel constructs BedrockCubeBox or BedrockCubePerFace from those records.
• An executed test using the unmodified released POJO classes confirms that a poly_mesh payload is silently discarded. A mesh-only bone parses with no cuboids. A mixed bone retains its one real cuboid and drops the polygon payload. Java 17 and 21 both passed this boundary test.
• BedrockModelPOJO.getGeometryModelNew() selects the first geometry entry. Do not export several entries and assume all are drawn.
• The inspected BeforeRenderHandEvent exposes a pose stack. Its existence alone is not a complete custom renderer API.

A successful JSON parse can therefore conceal missing geometry. Our export gate must check recognized geometry and actual rendered output, not just “no exception.”

Reproducible released-parser tool v2 · Java 17/21 results · Exact class hashes and reference counts

Route A: native cuboid model with authored texture detail

Use the released format deliberately: purposeful rotated/beveled-looking cuboid assemblies, per-face UVs, authored textures, sensible bone hierarchy and separately validated visual effects. Compare the in-game silhouette and material response against the concept. Do not silently replace the sculpt with crude boxes and call that equivalent.

The improved presentation audit checked actual Ars Armorer, Create Armorer and Helldivers packs: 8 + 13 + 57 display records. None of their main models in this bounded scan declared poly_mesh or multiple geometry entries. That is evidence for these particular files, not a statement that every addon uses the same route or that their visual quality is fully reproduced.

Static presentation audit v2 reports polygon payloads and multiple geometry entries alongside UV dimensions, cuboids and exact-path LOD references. Its reader supports comments/trailing commas while preserving quoted URLs; it is not a complete Gson/JSON5 implementation.

Route B: scoped custom geometry bridge, pending native prototype

The released BedrockCube is an interface with a compile(PoseStack.Pose, VertexConsumer, …) method. BedrockPart holds a public list of those objects, and BedrockModel.getNode(name) exposes a named part. This is a candidate extension seam, not proof of a supported stable mesh-import API.

A prototype could attach immutable custom geometry only to our own model parts and preserve their native bone transforms. Before adopting it, demonstrate all of the following:

1. Correct vertex format, face winding, normal transforms, UVs, alpha, light and overlay in the real renderer.
2. Valid topology for the render pipeline. Do not assume triangle data or degenerate quads work with every collector.
3. Resource-reload lifecycle and installation after model availability, before any relevant geometry cache is populated.
4. First/third-person, dropped, inventory, fixed-display, attachment and LOD behavior.
5. Shader, accelerated renderer and Hari/Vulkan compatibility with identical fallback behavior and measured frame times.
6. No mutation of another pack's shared model instance and no cross-item cosmetic state leaks.

Executed bridge prototype: immutable quad geometry

The candidate now has original Java source implementing BedrockCube. A harness calls it through the actual unmodified released BedrockPart.compile method, rather than merely checking whether the interface exists. Java 17 and Java 21 pass the dispatch test, and the package also passes from a clean extraction.

The checked path covers model-unit scaling, position/normal transforms, UV/tint/light/overlay argument order, multiple entries, defensive input copying and malformed geometry/transform rejection. Geometry stays immutable after construction. The implementation avoids allocating a Vector per vertex in its own loop; this is source-level behavior, not a measured game-speed improvement.

It accepts bounded planar convex quads with normalized atlas UVs. Nonfinite, degenerate, nonconvex, nonplanar, oversized and overflow-prone input is rejected before emission. The 4,096-quad cap is a guardrail, not a proven performance target.

Still outside this test: the full native render method, real PoseStack/buffer behavior, Forge/mixins, custom model installation, imports, resource reload, materials, hands, accelerated rendering and GPU performance. Narrow Minecraft types are harness stand-ins; real JOML math and fastutil are used. The missing unused Forge annotation warning is preserved in the result rather than hidden.

Original bridge source and reproducible runner · Exact JVM results and source hashes. This package is not an installable mod or a completed Soul Storm export.

Accelerated rendering cache warning

The shipped Accelerated Rendering compatibility mixin calls the BedrockCube.compile interface while building a collected mesh and reuses its cached mesh on a later cache hit. This makes the seam interesting, but also makes dynamic vertex/UV mutation inside a cached part unsafe to assume. A changing flame atlas or per-frame morph must use a verified dynamic route, immutable prebuilt frame parts, or explicit cache lifecycle support. Do not rebuild meshes every frame or globally disable acceleration to hide the problem.

Static source inspection does not prove cache behavior under every runtime configuration. Verify reloads, attachment swaps, alternate emitter counts and multi-player/multi-weapon displays in game before promotion.

Acceptance rule

The Blender source stays preserved. Visual quality, native geometry, animations, materials and performance are separate gates. A parser smoke test, a render image, a candidate Java interface or a generated concept cannot substitute for an actual accepted weapon.

Editable Soul Storm studies · Released parser scope · Reusable presentation core