Johnny Mnemonic¶

3D art, troubleshooting, best practices, scripting, and software setup across a comprehensive VLM toolkit build. VP9-to-VPX conversion with full toolkit pipeline to v1.2.2 release, featuring extensive optimization work and VR implementation.

Build Notes¶

VP9-to-VPX Conversion¶

The existing Johnny Mnemonic VPX table was based on VP9 code. The playfield dimensions were incorrect -- the table had been horizontally correct but vertically stretched to fit an incorrect playfield image. After fixing dimensions, all objects were slightly out vertically but not horizontally, requiring realignment of all walls, triggers, and elements.

Insert Creation Workflow¶

Insert creation workflow for VPW tables:

Use Photoshop with the playfield image at full resolution
Use polygonal lasso tool to trace insert shapes
Create layer masks for light transparency and text transparency
Two separate mask files needed: lightmask (which parts let light through) and textmask (which parts have text/graphics)
White areas in the plastic scan indicate clear/transparent portions
Insert blocker meshes placed under playfield prevent light bleeding between adjacent inserts

VLM Insert Tray System¶

With the updated VLM toolkit, the insert workflow was simplified:

No need to cut out playfield and do inserts separately
Inserts are now inside 3D trays in the toolkit library
Toolkit uses a mask to show transparent parts of the playfield
Another mask handles insert letters/text that let some light through
Two mask files required: jm_pf_insertmask and jm_pf_textmask

Plastics Transparency Workflow¶

Process for handling plastic transparency:

Source PSD file (660MB for JM) contains all plastics on layers with color correction
White areas in plastic scans indicate clear/transparent sections
Cutouts must be made in Photoshop -- extract clear sections for alpha masks
Ramp decals can have a white backside visible from below -- for VR, keep them visible from both sides
Wire mesh flooring decals require alpha PNG with transparent background
The dolphin decal is physically a bent piece of card/plastic jammed inside a ramp

3D Hand Scanning¶

The Johnny Mnemonic data glove was 3D scanned using an Einscan EP scanner. The scan was "fairly difficult to get due to the blackness" of the glove material. Actual glove dimensions: 7in x 4in, scan is 1:1 scale.

After decimating the 3D-scanned glove from raw scan to 25k triangles, UV maps broke. Instead of trying to preserve the scanned texture:

Separated the black and white parts of the mesh
Applied two different plain materials (no texture)
Result looked great without any texture mapping needed
The thumb was split as a separate mesh piece and rotated independently
Left hand was derived from the right hand scan by modification
Magnet cylinder placed under hand to close bottom gap

Blender Plastics Workflow¶

Recommended approach for plastics in VPW tables:

Put all plastics source images in one PNG
Drag and drop that image into Blender
Draw the plastic shapes in Blender using the image as reference
Work entirely in Blender rather than building plastics in VPX first
For transparent/clear sections, use alpha masks
Decals on ramp walls should be visible from both sides for VR compatibility

Ramp Modeling¶

Key ramp modeling principles:

"The ball goes to plastics level" -- ramp base height equals plastic height
Ramp drops should lower the front, not raise the back
Ramp end should be approximately 25-30 units higher than sling plastic
Inlane plastic height is definitely higher than rest of plastics -- visible above flipper level
Metal flap under ramp is diagonal (45 degrees), not a prism with separate top and side
Figure under ramp stickers are at 45 degree angle
Post heights between VPX and Blender library may not match -- metal washers on real machines add height

AI Image Upscaling¶

AI upscaling tool produces "very nice results" for pinball textures. Used for improving resolution of ramp decals and backboard textures. The technique was applied to existing scanned assets to improve detail for the high-resolution VLM renders.

VLM Toolkit¶

GI Collection Structure¶

Proper GI collection setup for dynamic shadows:

Need one "grouped" collection per GI string + one "split" collection for shadow-casting lights
Grouped collections: all bulbs turn on/off together, share timerInterval in VPX
Split collection: individual lights that cast raytraced ball shadows
For performance: only put a subset of GI lights in the split collection (e.g., one of two adjacent bulbs)
JM has 3 GI strings on playfield, so needed 3 grouped + 1 split = 4 collections
Lights in split collection driven by respective VPX lights independently

Dynamic Shadows Setup¶

Dynamic shadow setup process:

Enable raytraced shadows checkbox on selected GI lights
GI lights that cast shadows go in the "split" collection
GI light Z position affects shadow calculation -- street-level GI should be at Z=25
Setting Z to 1 hides oversized shadow artifacts (workaround for JM)
Ambient shadows separate from dynamic shadows -- can combine old ambient with new dynamic
The blend file must be set up for new dynamic shadows with GI split before batch rendering

4K Batch Rendering Pipeline¶

Full VLM batch rendering pipeline:

Set render height to 4096, render ratio to 100% for 4K (50% = 2K test)
Texture size should be at least render size (4096 or 8192)
Max lights setting affects RAM usage: 16 lights with 16GB RAM, 32 with 32GB, 64 with 64GB
Steps: Renders (Groups) → Meshes → Nestmaps → Export
Can skip completed steps by unticking "perform groups" in batch all
Renders can be force-quit and resumed -- toolkit checks renders folder and picks up where it left off
Delete specific renders from folder to force re-render of those items
Don't force-quit during nesting step -- must start over
CUDA renders significantly faster than CPU; check CUDA stays selected after crashes
Denoising runs on CPU regardless, uses peak RAM during this phase

Bake times: ~10 hours on 3070ti at 4K, ~5 hours at 2K

Refraction Probes¶

Refraction probe management for transparent ramps:

Ramp decals must be split into smaller pieces to avoid VPX confusion about rendering order
VPX can't determine if a large decal primitive is above or below a refracting ramp
Overlapping ramp layers need different probe numbers
Wall decals should be UV-unwrapped and baked rather than projected
Refractions need to be reapplied after batch import
The refraction "thickness" parameter causes decals to appear to "sink" into ramp plastic

Refraction thickness: On the central plastic ramp should be set to 1 or 2 to avoid duplicated decals.

Post-Batch Import Checklist¶

Critical post-batch-import fixes:

Set all Parts primitives to "static" for performance
Check primitive name lengths -- VPX has a max name length; too-long names break script references
Lightmaps need VPX material set to VLM.Bake.Lightmap (or VLM.Lightmap)
BM_room layers need correct refraction probe assignments
playfield_mesh must exist but be invisible and non-collidable (set to "toy")
playfield_physics in VLM.Visuals layer must be deleted (keep only the one in Physics layer)
Movable primitives need VLM.Bake.Active material
Pincab rails need bakemap color modulation added for room brightness to affect them
GI light Z height affects dynamic shadow size

UV Corruption at 4K¶

The glove and rail primitives rendered correctly at 2K but corrupted at 4K due to mesh optimization. Objects were decimated to extreme levels (from 19555 to 208 faces for glove).

Root cause: A UV map named "BakeMap" on the objects was confusing the toolkit's optimization.

Fixes:

Quick fix: Tick "no optimization" on affected parts
Proper fix: Delete the "BakeMap" UV from objects -- "all the uvs are useless on projected objects"

Insert Light Positioning¶

Insert overhead lights were positioned beneath the playfield (negative Z), causing inserts to appear dark or only partially lit. Moving lights above PF resolved the issue, but revealed they were all set to white instead of insert colors. Solution: Use Niwak's newer playfield material that allows some subsurface transmission, plus colored overhead lights.

Polygon Count Optimization¶

Screw primitives missing the noExp modifier contribute unnecessary polygon count. The noExp custom property on objects tells the VLM toolkit not to apply subdivision/expansion during mesh generation.

Select Occluded: The "select occluded" function in the VLM toolkit removes mesh faces not visible from the player's POV. This helps reduce file size and improve performance. However, it can cause problems in VR where the player can look at the table from angles not anticipated by the occlusion calculation. Compromise: set backface to 45 degrees.

Major Optimization via Re-bake¶

Re-baking the table through the updated toolkit cut polygon count roughly in half:

Old: 2,358,056 faces, 219 MiB GPU memory
New: 1,063,488 faces, 96 MiB GPU memory

The table looked better despite fewer polygons.

Scripting¶

Glove Magnet Mechanism¶

The original JM code used destroy/create ball sequences for the glove magnet. Rewrote it to track the ball with the glove without destroying it:

Original approach: TriggerGloveMag.DestroyBall then GloveMag.CreateBall to reposition
New approach: Start a timer and move the active ball position to follow glove movement
Key insight: "don't destroy a ball. Just start a timer and start moving that ball with the glove" (like Judge Dredd implementation)
The original CheckPickup function that used proximity-based destroy/create was found to be unnecessary

Matrix Mechanism¶

The JM cyber matrix (9-slot ball lock) was reimplemented:

9 kickers with different kick angles and slight scatter for ejection
Droppable walls added around kickers to create proper ball pockets
Space is very limited -- magnet position had to be raised to prevent balls going past walls
Original MatrixMetal kept collidable but should be replaced with a wall or ramp that becomes collidable when DoClearMatrix is called
The matrix plates have springs underneath -- circles are offset because the spring sits on the metal plate, not centered with the lamp
This offset causes the characteristic half-lit appearance of matrix insert lights

GI String Research¶

The team reverse-engineered JM's GI wiring through multiple methods:

Wire color tracing from under-playfield photos
Visited a real arcade to run GI string test from ROM menu
Frame-by-frame video analysis using , and . keys in YouTube to identify which lights share strings
3 GI strings on playfield (not 5 as manual suggested), string 4 is only the spotlight on the left glove, string 5 is backbox/coin door only
Key finding: GI strings are NOT arranged in geographic "islands" -- they alternate across the table creating a "sweep" effect when cycling 3-2-1-3-2-1

VPX 10.8 Incandescent Fader¶

VPX 10.8 introduced built-in incandescent light fading, replacing the external Lampz system:

Select "incandescent" fading type to get correct fading curve
Physics-based fade up/down: ~40ms for insert lamps
Suggested starting values: 40ms up, 120ms down
Same value for fade up and down is physically correct -- the fading curve handles the asymmetry
Insert lamps use 6.3V bulbs powered at 12-16V with short overcurrent peaks
All lights should be set to hidden with ball reflections enabled

Room Brightness Control¶

Room brightness control implementation:

Bakemaps rendered with environment/room lighting baked in
Can only make table darker than base render, not brighter
SetRoomBrightness function modulates bakemap colors at runtime
Key learning: "Always render with bright room as default, then darken in VPX as needed"

Ball Brightness Balancing¶

Ball brightness code (UpdateBallBrightness) adjusts ball appearance to match the baked lighting environment. Key finding: "It's not really worth balancing the ball brightness until the final render is imported."

The ball brightness equation can be tuned via script:

Dim v: v=(lvl * 240 + 15)/255

This formula controls the range of brightness adjustment.

GI Light Naming¶

GI light naming and hookup process:

GI lights named as l[number][letter] -- e.g., l142k, l142l, l141k for pop bumper GI
The spotlight (only light on string 4) named l143a
Pop bumper GI lights had to be invented since they didn't exist in VPX
GI naming must match between VPX light objects and Blender bake names
VLM Script Helper file must be updated when lights are renamed

UseVPMModSol=2 Updates¶

v1.1.3 updates for UseVPMModSol=2: Removed InitPWM sub and ModSol option, updated UpdateGI and FlashXXX subs for new physics output, set all light fader models to None. The InitPWM initialization is now automatic. Also fixed timers causing stutters, added FlipperCradleCollision, updated CheckLiveCatch, enabled more reflections.

Scriptable Refraction Materials¶

VPX added the ability to change refraction materials for ramps at runtime. Fish Tales implements this as an F12 option menu. This allows players to toggle refractions on/off based on their hardware.

Game Knowledge¶

Broken Dataglove Compensation Mode¶

When the JM dataglove is broken, the ROM enters compensation mode:

Locks become virtual; balls auto-plunged for multiball
Crazy Bob's gives awards immediately (no player choice)
Drop target registers as center lane shot (warning: tends to drain SDTM)
Spinner/right loop shots that would lock go around left orbit instead
When ball reaches VUK, game "repeatedly and quickly fires the kicker, so that the ball will eventually accumulate enough spin to bounce off the dataglove sideways" -- typically 5-10 kicks
This compensation makes the game harder by removing the best scoring opportunity

Standard Flipper Length¶

Standard flipper length in VPX is 147 vpunits.

Troubleshooting¶

VPX Playfield Physics Duplication¶

The VLM toolkit consistently exported a playfield_physics primitive into VLM.Visuals layer that was collidable but had no holes. This overlapped with the proper physics mesh in the Physics layer.

Fix: Delete the playfield_physics in VLM.Visuals, keep the one in Physics layer.

Table VSync Override¶

Johnny Mnemonic shipped with table vsync override set to 1 in User Properties tab. Changing to -1 can help performance but users reported intermittent crashes and inability to save the setting. Loading through Popper (front-end) bypassed the crash issue.

Negative Depth Bias Breaking Ball Trails¶

The table's playfield had a negative depth bias causing the playfield to render after the ball trail, making ball trails non-functional. Ball trails were on the list of features to be removed in favor of proper motion blur.

GPU Load and Refractions¶

Removing ramp refractions dropped GPU load from 100% to 90% on a GTX 1060 at 4K 60Hz, making the table playable. To disable refractions without deleting them: in Render Probe Manager, set the ramp's Plane Reflection probe Max Quality to "Disabled."

Pure Black Breaks Rendering¶

Using RGB 0,0,0 (pure black) in VPX images breaks renderers -- causes divide by zero errors or gets hard-coded to 1,1,1 minimum. A pure black playfield makes lights not work properly. Use at minimum 10,10,10 for blacks. Same applies to pure white.

VR Room VRAM Impact¶

Adding DaRdog's 1K mega VR room to JM increased video RAM from 2.6 GB to 4.3 GB (+1.7 GB). The room needed optimization before inclusion.

Post Passing Adjustments¶

Issues identified during JM testing:

Post passes were not working at all
Inlanes needed speed limits applied
Flippers felt too powerful
Ramp entrance plates needed sounds

Software Setup¶

Required PinMAME Build¶

v1.1.3 requires an updated PinMAME build. Bumpers and slingshots stopped working without the update.

VPX Build Compatibility¶

VPX 10.8.1 BGFX should not be used for JM. Solenoid callbacks are now auto-optimized in the newest version of VPX.

Tools & Resources¶

VPX Live Editor¶

VPX live editor usage tips:

Rotate view by clicking and dragging the box on the RIGHT side (not left -- the "stop" button)
Pan view: hold Shift + middle mouse drag (same as Blender)
Can edit table during actual gameplay with a ball in play
Changes show instantly (e.g., changing light color immediately affects ball reflections)
Must press "save" icon next to a property to persist changes -- otherwise lost on exit
"Halo" light mode is the traditional light display; "Classic" shows through playfield

Texture Swapping in Toolkitted Tables¶

In toolkitted tables, textures are not easy to swap directly -- the toolkit generates a base image and overlays lighting scenarios on top. New base textures need to be plugged into Blender as the basis to generate new images. In non-toolkitted tables, texture swapping is straightforward.

Ramp Material Reuse¶

Ramp materials can be reused across table Blender projects. The Earthshaker ramp material was suggested for JM's ramp.

Best Practices¶

Playfield Source and Color Correction¶

The JM playfield image was not a scan but a high-quality stitch of photographs with decent halftones. Despite looking dull/gray in areas, manual color correction could address this without a full re-render.

Boosting Lightmap Brightness¶

To boost brightness of a dark object without re-rendering: change the opacity of all lightmaps associated with that movable in the VPX editor to a value greater than 100. For example, the holographic spinner lightmap opacity was set to 300 to make it appear properly lit.

NoExp Script Optimization¶

Before re-running a Blender batch render, run the NoExp script on the project first to optimize results. The toolkit has progressed significantly, allowing more optimized results with a re-run.

Contributors: sixtoe, iaakki, tomate80, apophis79, flux5009, baldgeek, its.me.dazz, niwak, benji084, DJRobX, leojreimroc, DGrimmReaper