Ten practical assignments that train disciplined, evidence-backed texture work on manufactured equipment. The brief starts with a masked base-color edit and ends with a full capstone review.
20 native AI referencesEvery assignment has an overall material goal and an informative close-up. They are concept references only.
LOCAL FILE · /Users/forgebot/forgeapps/artifacts/texturing-intern-practice/index.html
COMMON SUBMISSION STANDARD
Scope First. Prove It. Reopen It.
These briefs follow the Blender texturing workflow procedure. The source FBX stays read-only. Every edit has an explicit allowlist and protected set. A saved file, native pixel diff, byte/hash report, and live comparison are required before the work can pass.
Every assignment requires
Original FBX and source-map hashes
Native dimensions, bit depth, color spaces, materials, UV set, and mask hash
Isolation proof at native resolution plus filtering/mipmap risk
Saved-output reopen and before/after controls QA
What the images mean
AI concept references only
Not engineering evidence or actual textures
Do not infer shader packing or channel semantics
Do not treat a reference as geometry-change permission
ASSIGNMENT 01
Powder-Coat Base Color
1/10 — very easy
FIXTURE · TXT_BRACKET_POWDERCOAT_AEFFORT · 45–60 minutes, planning estimate only
Starting setup
Clean mounting-bracket FBX, single base-color map, clean UV mask, inspection scene orientation, and confirmed literal object name. No model or UV edits are authorized.
Task
Make one restrained base-color adjustment to the powder-coated steel bracket so the finish reads as a consistent approved dark graphite coating.
Allowlist
One base-color map
Pixels inside the TXT_BRACKET_POWDERCOAT_A mask
Base-color values only
Protected
Source FBX bytes
UVs and geometry
Material slots
Roughness, metallic, normal, and AO
Every non-target pixel
Labels and fasteners
Atlas dimensions
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7
Action steps
Hash and inspect the original
Confirm the supplied mask against the literal object name
Verify zero overlap with protected regions
Confirm base color is sRGB
Apply the approved correction only inside the mask
Reopen the saved map and run native pixel/hash checks
Compare original and revision under identical lighting
Deliverables
Revised base-color map
Mask and mask hash
Unchanged source-map hashes
Native diff report
Before/after screenshot
Verified-results note
Measurable pass / fail gate
Pass only when the file reopens at native dimensions, the FBX hash is unchanged, all changed base-color pixels are inside the mask, zero pixels change outside it, all non-target maps hash identically, and the bracket is the only visible change.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
Molded control grip, roughness-only mask, source packed-map contract, viewer preset, and clean base color.
Task
Improve roughness response so the black molded grip reads as manufactured elastomer. Change roughness only; do not fake the result with color shading, added highlights, embossed patterns, or wear.
Allowlist
Roughness channel inside the TXT_CONTROL_GRIP_RUBBER_B mask
Protected
All color channels
Normal, metallic, AO, and packed neighbors
Grip silhouette and seams
Handle geometry
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7
Action steps
Record the channel contract and linear data status
Prove the mask does not touch the handle or seams
Make one roughness-only change
Reopen the output
Prove every other channel is byte-identical
Compare under identical exposure and verify color did not change
Deliverables
Revised roughness or documented packed-map revision
Contract note
Mask evidence
Native diff report
Viewer screenshot
Saved-output reopen evidence
Measurable pass / fail gate
Pass only when roughness is the only changed channel, the black base color is unchanged, no protected pixels move, and the material difference is visible without color shading.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
New yellow equipment cover, approved paint swatch, protected fastener mask, protected blank label-plate mask, clean map, and named viewer setup.
Task
Match the approved yellow paint across the cover without chips, rust, grime, scratches, or edge wear. Keep every fastener and the label plate unchanged.
Reopen and run native diff/protected-region checks
Compare under identical exposure
Deliverables
Revised base color
Swatch record
Three masks
Hashes and diff counts
Comparison screenshot
No-wear scope note
Measurable pass / fail gate
Pass only if the paint matches the approved swatch, protected pixels are unchanged, no wear appears, and fasteners and label plate remain intact.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
Hydraulic valve assembly whose target shares an atlas with hoses, fittings, and neighboring hardware; object list and inspection scene orientation supplied, mask not assumed safe.
Task
Correct the target valve-body coating on a shared atlas while preserving hose, fitting, bolt, and neighboring-object coverage. Prove coverage before changing color.
Allowlist
Base-color pixels covered only by TXT_HYDRAULIC_VALVE_ATLAS_D
No edits to shared neighbors
Protected
Hoses
Fittings
Bolts
Neighboring objects
Packed channels
Normals
UVs
Atlas size
FBX bytes
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7
Action steps
Enumerate every mesh using the atlas
Rasterize target and non-target coverage at native resolution
Inspect geometric UV intersections and filtering footprints
Stop if isolation is not proven
Make the smallest correction only when safe
Reopen and verify target-only changes
Deliverables
Object and UV coverage table
Mask
Revised map
Overlap evidence
Changed-pixel report
Isolated-object screenshot
Stop/report note when needed
Measurable pass / fail gate
Pass only with zero non-target changes, proven isolation at native and filtering footprints, unchanged protected channels, and a viewer screenshot showing neighboring objects are unchanged.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
Mentor supplies a painted metal workshop-tool housing, written authorization for one local edge-wear zone, and a confirmed metalness-workflow contract distinguishing dielectric paint from exposed metal. Adjacent polymer parts are protected.
Task
Add only the authorized local wear. The exposed area must read as bare metal while the surrounding coating remains intact. Do not distress the whole housing or use roughness to fake bare metal in base color.
Allowlist
Authorized base-color, roughness, and metallic pixels inside the local wear mask for TXT_TOOL_HOUSING_WEAR_E
Only the named channels
Protected
All other housing pixels
Adjacent polymer parts
Labels
Fasteners
Normal
Metallic outside the local wear mask
AO
UVs
Geometry
Unrelated maps
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7
Action steps
Write the wear boundary and material contract
Prove the boundary does not reach a seam or label
Build a reversible baseline
Add the smallest authorized wear transition
Reopen and check coated/bare-metal channel separation
Verify no broad grime was added
Deliverables
Scope note
Recoverable baseline
Revised maps
Local mask
Channel diff report
Before/after screenshot
Preservation note
Measurable pass / fail gate
Pass only if wear stays inside the authorized zone, the material transition is channel-correct, no unrelated pixels change, and the baseline remains recoverable.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
Manufactured engine casing, documented nonstandard packed-map contract, smoothness inversion toggle for diagnosis, and consuming-shader notes.
Task
Correct smoothness interpretation in the viewer and document exact conversion to roughness without modifying unrelated packed channels. Keep the original packed file unchanged unless a written allowlist names one channel.
Allowlist
Documented packed smoothness channel only
Viewer routing toggle when deliverable is diagnostic
Protected
Unknown packed channels
Base color
Normal
Metallic
AO
Source FBX
Every unnamed channel
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7
Action steps
Record the documented mapping and shader expectation
Prove inversion with a numeric sample
Label viewer-only work diagnostic
Change only the named channel when authorized
Reopen and compare original versus corrected routing
Deliverables
Channel contract
Numeric inversion note
Original and revision hashes
Viewer toggle screenshot
Optional one-channel revision
Diagnostic-versus-faithful label
Measurable pass / fail gate
Pass only when channel meaning is documented, inversion is numerically correct, unknown channels are unchanged, and the viewer does not claim faithful evaluation when it is diagnostic.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
Hose/manifold assembly, normal map, base-color map, two viewer presets, and unresolved normal-Y/color-space status.
Task
Diagnose whether the issue is normal-map Y orientation, color-space interpretation, or material assignment. Produce diagnostic and faithful renders only when supported by evidence. Do not repaint the assembly to hide the issue.
Allowlist
Viewer normal routing
Normal data color-space setting
One explicitly authorized normal-map revision
Protected
Geometry
UVs
Base color
Packed maps
Source FBX
Unrelated materials
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7
Action steps
Establish a control render
Test Y flip, data color space, and material input one at a time
Record every result
Separate diagnostic from faithful target-engine interpretation
Save only the authorized change
Reopen and compare
Deliverables
Control render
Diagnostic matrix
Selected normal settings
Optional revised normal
Before/after comparison
Evidence-based conclusion
Measurable pass / fail gate
Pass only if the diagnosis identifies the changed variable, no unapproved color repaint is used, the source is unchanged, and diagnostic and faithful renders are labeled separately.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
Service panel with tight UV islands, deliberately unsafe fixture variant, native atlas, and target-engine capture baseline.
Task
Evaluate subpixel overlap, filtering footprint, and mipmap bleed risk. Make a reversible edit only if isolation is proven. Otherwise stop and report a safer scope. Never force the unsafe fixture into compliance.
Allowlist
Native atlas pixels inside a proven safe mask
No pixels when the stop gate fails
Protected
Unsafe fixture
Neighboring islands
Latches
Seams
Labels
All non-target channels
UVs
Geometry
Source FBX
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7
Action steps
Rasterize native and representative mip-level coverage
Measure island distances and filtering footprints
Record the unsafe case
Edit only when the safe mask survives
Otherwise preserve originals and submit a narrow-scope recommendation
Deliverables
Native and mip-level coverage evidence
Unsafe-fixture report
Safe-mask decision
Revised map only if justified
Comparison or stop screenshot
Measurable pass / fail gate
A correct stop with evidence passes. A forced edit on a risky mask fails even if it looks good in one screenshot.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
Multi-material pump skid, multiple maps, locked camera/light setup, previous accepted revision, original FBX, and browser viewer that loads original FBX bytes directly.
Task
Deliver a controlled multi-map revision without changing camera or lighting. Show original versus previous versus current revision. Compare browser viewer with target-engine baseline and run full controls QA.
Allowlist
Written list of pump, hose, fitting, and guard map channels
Pixels inside proven masks
Protected
Locked camera/light/exposure
Source FBX
UVs
Non-target maps
Labels
Neighboring objects
Unnamed channels
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7
Action steps
Hash original and previous revision
Load original FBX bytes directly
Verify locked setup
Prove each mask and channel
Make one bounded revision set
Reopen every map
Exercise all controls
Record browser-versus-engine differences
Deliverables
Original, previous, and current maps
Revision manifest
Locked camera/light JSON
Original-FBX browser evidence
Full controls QA
Native diffs
Recoverable package
Measurable pass / fail gate
Pass only if source FBX and locked setup are unchanged, every map reopens, every changed pixel is allowlisted, the three-way comparison is readable, and all viewer controls work without console errors.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
ASSIGNMENT 10
Diesel Power Unit Capstone
10/10 — extremely difficult
FIXTURE · TXT_DIESEL_POWER_UNIT_CAPSTONE_JEFFORT · 8–12 hours across two review rounds
Starting setup
Complex industrial diesel power unit, multiple atlas maps, protected lettering mask, ambiguous packed-channel gate, target-engine capture, browser comparison fixture, and two-round stakeholder review schedule.
Task
Inspect the full unit, resolve confirmed channel contracts, stop at any unresolved packed-channel gate, and complete only authorized texture-map changes across two review rounds. Report the dangerous UV-overlap variant instead of mutating it.
Allowlist
Only channel/object/map cells confirmed in the capstone scope matrix
Each revision round has its own allowlist and mask hashes
Protected
All lettering
Source FBX bytes
UVs
Geometry
Material slots
Ambiguous packed channels
Unsafe-overlap variant
Locked target-engine camera/light
Every non-authorized map/channel/pixel
Procedure sections
§1 · §2 · §3 · §4 · §5 · §6 · §7 · §8
Action steps
Build a scope matrix and hash the full source
Inventory atlas users, UV sets, materials, channels, and lettering
Resolve the contract or stop at the ambiguous gate
Prove every mask and dangerous-overlap variant
Create a recoverable baseline
Run round one and verify
Apply only authorized round-two changes
Run regression checks and browser/engine comparison
Record separate artist signoff
Deliverables
Full scope matrix
Original hashes
Two revision manifests
Two sets of masks and diffs
Protected-lettering proof
Ambiguous-channel decision
Dangerous-overlap report
Recoverable package
Browser and engine screenshots
Regression report
Separate artist signoff
Measurable pass / fail gate
Pass only if every changed cell is authorized and proven, protected items are byte/pixel unchanged, both rounds reopen cleanly, regression checks pass, dangerous overlap is reported rather than mutated, browser/engine differences are named, and the artist signs off separately.
REFERENCE PAIR
Overall material goalOpen zoomAI concept reference · not engineering evidenceInformative close-up material studyOpen zoomAI concept reference · not engineering evidence
