Matte Painting

General Approach to the Photo-Bash

As part of a subscription-based learning material for FXPHD, Mike James (2010) outlined the application of a total production workflow and pipeline for a matte painting created for the Red Dwarf series. This process was different to any I had seen to this point in my research insofar as he started by building the primary geometric elements for the G-Deck environment in 3D from his concept sketch but then used a selection of arbitrary output variables (AOV passes) from the 3D software as the basis for starting work on the matte painting. This contrasts with the more traditional method in which the sketch would be developed into the final matte painting prior to the commencement of any 3D work. I found this to be an approach worthy of exploration in this project for three primary reasons. Firstly, my work on establishing the projection cameras resulted in the reframing of the primary shot so, had I simply built the matte painting from my concept sketch, it would have been incorrect and requiring of significant modification to the ground and all three buildings for them to work as sources of projection. Conversely, by using this alternative approach, the wireframe renders from each projection camera provided clear and accurate boundaries from which the artwork could be constructed.

In testing this workflow, I worked on a section of one of the primary buildings, which I acknowledge conflicts with one of the key points raised in my work on the first project in this case study, during which I advocated creating the least possible amount of topological detail in the projection geometry. In this case a significant amount of extrusion, bevelling and chamfering work has taken place on the surfaces to add architectural detail.

My approach was to render diffuse, wireframe and Ambient Occlusion passes from each projection camera and at the resolution determined by the focal length calculation that was established by my projection strategy. To ensure the final matte painting has the requisite image quality (resolution) it is critical that the Photoshop canvas has at least the same height and width ratio as these output renders.

The image (below) shows these renders outputted for the central projection for the ground surface, stacked ready to begin matte painting. Note that the AO pass has a Multiple blend operation applied in order to leverage the self-shadowing details. This works by assigning a degree of transparency to greyscale pixels commensurately becoming more transparent as they get brighter. Therefore white pixels are fully transparent and black pixels are fully opaque. An additive blend operation is assigned to the wireframe pass which has the opposite effect, resulting in the lines being retained but the black pixels becoming transparent. The diffuse pass simply sits at the bottom as a point of reference and all digital paint work takes place above this in the stack but below the wirefame and AO pass.

The wireframe pass (below), used in conjunction with the base diffuse render, provided extremely accurate guides for placing and aligning textures in perspective.

The image (below) provides an illustration of how the textures, in this case a stone material for the door surround, could be manipulated and trimmed to align to the topological reference provided by the wireframe pass with a high degree of accuracy.

The Ambient Occlusion pass (below) contains self-shadowing information derived from the topology of the geometry prior to the addition of additional light sources.

The image (below) illustrates the benefit of this pass and justification for the extra level of topological detail in geometry. By layering the pass over the textures and with a multiply blend mode assigned, the pass adds physically accurate self-shadowing information without the need for painting in this detail manually using traditional digital painting techniques and with a greater degree of accuracy.

Utilising these passes, across each projection and in conjunction with texture sources and visual reference from my library, I was able to complete the matte painting process.

Building 3

This issue should have been picked up during development of the projection strategy but only really became clear and obvious when the checkerboards on the patch projection was swapped out for the DMP texture.

From the view of the camera in the latter frames of the shot, the contours of the stone surface along the edge of the building would be visible but the hard edge at the far side of this façade was clipping this off as a clean edge.

The building 3 geometry was duplicated so I could perform some digital sculpting along the offending edge to extend the projection surface.

The wireframe was then re-rendered from this perspective and used to modify the DMP work on the patch projection by adding in the additional stone material. The modified texture was then reloaded and projected onto the modified geometry.

Patches (i.e. B1 doorway, road) to follow…