Texture Packing, Compression, and Virtual Texture Guide

Packing is a runtime decision

Channel packing reduces texture samples and file count by placing grayscale signals such as occlusion, roughness, and metallic into separate channels of one image. It works best when those signals use the same UV set, dimensions, mip policy, and material lifetime. Packing unrelated data can waste memory because every channel inherits the largest resolution and must stream together. Document the channel order because thumbnails cannot reveal semantics.

Production workflow from source to engine

Author source textures at justified resolution, preserve uncompressed masters, and export through a named preset. Build the packed texture, inspect every channel independently, then import with color conversion disabled when it contains data. Select the engine compression profile, generate mips, and test at representative camera distances. Compare GPU memory, sample count, streaming behavior, and visible artifacts. Keep source channels so the package can be rebuilt when specifications change.

Compression and resolution settings

Base color usually tolerates color compression; normal maps need a profile designed for directional data; masks may need high precision if thresholds or smooth roughness expose banding. Alpha can change the chosen block format and memory cost. Use power-of-two dimensions where the platform requires them. Set maximum size per platform rather than shipping every source at authoring resolution. Inspect mips, not only mip zero, because distant artifacts often dominate play.

Tiling, unique detail, and virtual pages

A seamless tiling surface is efficient for repeated material detail, while unique maps preserve authored wear and identity. Trim sheets and decals can bridge the two. Virtual texturing divides large images into pages and streams visible regions, which helps landscapes, UDIM-like surfaces, and expansive unique materials. It adds page tables, feedback, cache pressure, and latency. Small props, rapidly changing textures, or platforms with tight support may be better served by conventional mips.

Decision example: an open-world cliff

A cliff set uses repeated rock detail, unique macro variation, and sparse decals. Keep the high-frequency rock as a seamless tiling texture, pack compatible masks, and reserve a lower-resolution unique map for macro breakup. Consider virtual texturing only if many large unique surfaces exceed conventional residency and profiling shows stable page reuse. Choose compressed formats on target hardware, then test camera flight for cache misses rather than approving a still frame.

Failure diagnosis by symptom

Colored masks usually indicate accidental sRGB conversion or channel swizzle. Roughness banding points to compression precision, bit depth, or an overly sensitive material range. Dark seams at distance suggest insufficient padding or bad mip generation. Sparkling normals indicate the wrong compression profile. Blurry virtual textures can result from low residency, feedback delay, or cache pressure. Sudden memory growth may come from alpha, oversized dimensions, or packing channels with mismatched needs.

Evidence checklist for platform approval

For each texture, record source dimensions, delivered maximum size, channel semantics, color-data flag, alpha use, compression profile, mip policy, and estimated GPU memory. Capture mip-zero detail and typical gameplay distance on each target tier. For virtual textures, profile page faults, cache occupancy, feedback cost, and visible recovery during fast motion. Confirm that the material uses the documented channel swizzle and that source masters can regenerate every packed output.

Related references and next step