WWI Trench Locomotive

Overview and initial decisions

For my second project I decided that I wanted to make a diorama. The brief called for a diorama that should contain a vehicle, hero object or character.  Having enjoyed working on the helicopter previously, I was motivated to try another vehicle; something more complicated and ambitious that would then fit into a scene. I had a looked at what my peers had created from the previous year as a source of inspiration and thought about my own interests.  I have always had an interest in the railways and so, after a while I settled on a steam locomotive.

When thinking about what kind of steam locomotive to make, I was mindful about the budget for the project (50,000 tris). This limit would make larger locomotives unfeasible as the detail would be compromised, I opted to make something small but detailed and settled on a first world war steam locomotive, a narrow-gauge locomotive. I had recently been on holiday to Devon where I had visited a preserved narrow-gauge railway. Ever since, I have had a strong interest in doing something around this theme. History is another of my interests and this project was a way of tying these together. At the time I took the opportunity to take hundreds of reference images thinking I might end up modelling something like this in the future. In the second year I had drawn a scene from this place as part of my traditional art practice.


The next stage of the project was to source adequate reference material. I had previously used Pureref as a way of organising and displaying all of my reference images and decided that I would continue for this project what I had already found that worked.  I compiled a large mood board of all the reference images I could find. Below are some of the main images I used to develop my ideas for this project.

Unfortunately, there seemed to be a distinct lack of high resolution images which hindered my modelling later. The biggest issue with my reference gathering was the complete and total lack of any kind of plans or drawings for the selected locomotive. This stalled my work for a good time, I even joined the narrow-gauge garden railway society in desperation looking for information beyond google searches. All this effort resulted in nothing unfortunately and I ended up having to make my own orthographic drawings, working from the grainy images sourced from google This taught me a valuable lesson about finding good reference images and the importance of building my own library, something I have been doing for a while already.

I did consider going to a Welsh narrow-gauge preserved railway in the hopes of finding a surviving example of the train I was making, however after much research I discovered that the only preserved example of the train that I was making was in France. Unfortunately, it was not feasible for me to make a trip to France to photograph this train as much as I would have liked and needed to at some points in the modelling process when I could not understand how certain elements went together as there was no adequate reference material on the internet. Instead I made use of the images I could find as well as using some images for other models of similar trains.


With freshly minted orthographic images in hand, I quickly set about working on the modelling and blocking out the shapes. This process showed that there were some problems with my orthographic images as certain elements where out of scale. I discovered that the issue here was working from frontal images of a model, and side images of the real locomotive. I did not count on the model and the real thing being so different, to a such an extent to cause a major issue with my orthographic images.  I made new orthographic images which this time worked flawlessly, and I was able to block out the rough shape of the vehicle and was satisfied that it looked correct.

I modelled the vehicle mostly in 3dsmax to make the low poly. Once all elements had been created, I then used the low poly as a basis to make my high poly by using the open subdivided modifier. My workflow became taking the low poly, using Open Subdivide and then exporting the object from max to Zbrush where I would add an extra level of detail, such as dents, dings and so on before decimating and re importing into max. Once all the high poly items had been created I took all the assets into substance painter where I baked by mesh name, I was careful to set my 3dsmax project up with all the parts of my locomotive separated to avoid baking errors.

In this project I broke the modelling down into the major elements, such as the body, wheels, details ect. I found this approach was easier than trying to make everything all at once. I organised my project along similar lines inside 3dsmax. I arranged all my objects in layers named after the elements they related to. This made a massive difference and made my life easier as it was much easier to find the elements I was trying to work on. Gone were the days of scrolling down a seemingly endless list of objects looking for one single part. This was an error I had made and learnt from with earlier projects.

At the very start of this project I had taken texture space into consideration with the way I modelled the train, I wanted as much of the train that could to be mirrored so that I could achieve the highest resolution in the available space. not thinking about how the final uv map would look with the helicopter project had been a massive mistake as when I got to the end I had to take great pains to retrospectively work on the model to make the best use of the space. The first texturing attempt of the helicopter looked terrible even with a rather large texture map.

With the train however, I decided to make heavy use of the symmetry modifier and unwrapping as I went on certain elements, this allowed me to perfectly stack the uv shells. The only down side to this approach was that certain elements, such as the water tanks had mirrored text. To overcome this issue I made some separate models and used them instead of painting directly onto the mirrored parts with anything that might look odd when mirrored, such as text.

Another big change that I made in this project was to, from the very start of the project, use two bright contrasting colours inside 3dsmax on the high and low poly models. I did this so that I could quickly see at a glance which elements where high and which were low. This process helped to avoid the issue of having certain elements missing from the high poly. As I had adopted the baking by mesh name it was possible to miss a few items as I had so many and this process certainly helped me avoid this from happening.

Smaller details

Next came the smaller details on the train, for example the cast iron elements, such as the manufactures plates on the sides of the cab and water tank as well as on the front of the smokebox door. I firstly made the correct shape of the item in 3dsmax as a base before exporting into zbrush and subdividing the object to provide enough polygons for the next step. I then created a basic heightmap of the design in photoshop which I then took back into zbrush and applied to the brush as an alpha mask.

Adding this process to my workflow has been one of the most successful things I’ve tried in this project, as stated previously it allowed me to model in a way that previously was not possible using exclusively open subdivision. I can see this technique lending itself to other applications in different settings. I decided to expand upon my workflow by adding an extra step by taking other elements of my model into Zbrush. This allowed me to impart more detail into the models and do things that previously were not possible with tools like open subdivision, one such example of this was the water tanks on the locomotive.

Although only subtle I feel the denting makes a big difference on an object that would otherwise be rather boring and flat. We talked in the first and second year about making objects have an interesting silhouette and I felt that this denting adds just that little bit of extra believability to the work. Unfortunately, though, I think that in the end the subtle nature of the effect has not shown through on the final product, perhaps if I used larger texture sheets in the future this might correct this problem.


After all the modelling had been completed the next task for me was to unwrap the entire thing. I was dreading this step as it had felt extremely monotonous on all the projects I had worked on up until now. I was determined to find a better way around the process, one such way was to look at tools that did the work for me, such as Unrwrella.

I experimented with these tools but found them rather lacking as the automatic packing functions where not much better than what I could achieve myself with little effort. I also found that a manual pack to be a much better way of using the space as automatic systems tended to make stupid mistakes like not packing circles inside round objects ect. As I had planned the unwrapping into the modelling workflow this time around it wasn’t as slow as it had been on my previous project. After a few hours of work, I was very pleased with the way that I had packed my uvs, compared the abomination that was the auto unpacker. Below on the right can be seen my manually packed sheet and on the left is the automatically unpacked.

The size of the elements really suffers, as the automatic unpacking function cannot take into consideration the stacked uv islands unless using the grouping of elements button.


Before I conducted the final baking process, I did a number of small tests to make sure that I would not be in the same situation as I had with the helicopter which had been a disaster. I had ended up having a nightmare with the helicopter as almost every object that I tried to bake on it ended up in failure with things baking on the wrong objects and all kinds of issues.

I tended to get carried away with the baking tests on the smaller self-contained sections, I feel that this was actually useful however as it did not only show me that the bake had worked as intended but showed me that I could hide minor baking errors under layers of rust and other such effects. This was something that had been passed on by tutor Mike Kelly. He has advised that pouring over the minute details was not always needed and that fixing tiny errors could be a waste of time. This hard to come to terms with as I am a perfectionist. This is definitely something that I have improved and made progress with over the years.

During this project I learnt to test things along the way rather than baking everything at the end and fixing problems then. This made sure I could fix any issues along the way. As my knowledge of baking improved, I could also foresee errors before they happened which led to making changes before issues arose.

Due to planning my project carefully from the start and learning from my previous mistakes I was able to have an almost flawless bake the first time saving me time for the next stages. This was a great weight off my mind as I had previously got rather stressed with the helicopter as the baking was still not coming out right even right up until the deadline. In the end I managed to pull it together, but it was tight. This reinforces the importance of proper planning and preparation in my mind and something I will take forward into other projects.


Texturing the model is the most enjoyable stage for me, seeing everything come together is great, as before this point the model didn’t really look much like anything as it relied heavily on symmetry to save texture space and make the most of the budget available to me.

With this project, like the previous, one I did all my baking and texturing inside substance painter. I started by firstly masking of all the key elements of my project into separate layers, such as I had done in a similar way with my modelling workflow. This allowed me to avoid getting confused with millions of layers all stacked on top of each other not knowing what each one did and having effect spill over onto areas that were not desirable. For example, when I started my train before I had set my masks up correctly, I was getting the soot effect from the chimney layering itself on the interior elements of the cab, which just didn’t look right at all.

To apply the details like the lettering on the side of the wooden box and the gauges in the cab, among other small details. I created a series of simple alpha masked textures to use as stamps inside substance painter. I had tried this on the helicopter to apply the decals like the soviet star and numbering with success and wanted to build upon it here. I created the decals inside Photoshop and used levels and contrast to control how the height map would come out, much in the same that I had created the height maps for the stamping process inside Zbrush.

I found that I had a need to use substance designer in this project, to create elements of the train as seen from my reference images. one such example was a metal pattern found on the front plate of the locomotive. although only a simple material, it produced the desired result.

Ultimately I only required the normal map channel of the material I had created, using some masking I cut the new piece into the location and the result was perfect. Making this as a mesh inside 3DSmax, which is something I would have probably done in the past, is a much slower and less efficient way to work. Little techniques like this end up saving me time and processing power. My old method ended up generating meshes with very high polygon counts in the high poly, which had the unfortunate effect of causing my computer to slow down quite a bit, further hindering effective working.

With the texturing in the train project I mostly just went onto auto pilot and quickly found myself with a fully textured locomotive. I decided to texture the train using the approach of making a clean pristine train first, and then applying all the weathering at the end as a final coat.

I felt the weathering effect was rather too strong in the above example, so toned it down. I feel that in some cases less is more. If I have more time I would like to totally redo the textures as I feel that I have learnt so much in the first pass, another fresh attempt would yield much better results.

One thing I am very aware of with my project is that the subject matter is perhaps not the most interesting to most people, and although I feel I recreated the object in question quite well. The original colours of the locomotive are perhaps not the most interesting or eye catching.


The final stage for me was to polish the model and improve the textures to make the engine feel more weathed and worn, as if it had been in service in a battlefield for many years. I consulted my peers and some industry contacts on the best way to go about this final stage, as it had not been my greatest strength on the helicopter project. I was advised to look at how plastic model kits are weathered. I found this advice really useful and the final result improved no end.

Presentation and critique

I finally uploaded my completed project to Sketchfab and presented it with all the post processing and lighting effects I could muster and was quite pleased with the finished result. I sought critique from peers and past tutors. I also have a wide range of contacts on LinkedIn who I have found useful for honest and critical critique. One of these connections was a contact from Bohemian Interactive who had said I could send work over whenever I need honest feedback. He said that he could see lots of progress with my modelling compared to previous work I had shown him. He liked how my geometry was tidy and neat. He did however advise that my normal map was perhaps too strong on the metal wear, something I had considered myself. We also discussed how to add more interest to the large grey panels on the body. I thought about how my Dad was creating a model railway and was using a weathering service to add realism to his trains. A website called Grimy Times sells weathered locomotives. I used these images as inspiration and kicked myself that I had not thought of this sooner. I made this the last few changes that I made to my piece before submitting it. Below is the locomotive as presented inside Sketchfab.

Looking at the renders inside sketchfab left me with the feeling that I could do better. I had wanted to learn Marmoset for awhile as I had heard from my peers that it gave more control over lighting and other functions to the artist and was generally a beter alternative. After following some tutorials and getting to grips with the package, I imported my locomotive. After a few hours of tinkering with various settings I was quite pleased with the results.

Final Thoughts

Although this was the second project I started, it was my last to finish as this was the one I decided to polish to the highest level. I have learnt a lot from these projects including tightening my workflow. I have further demonstrated the need for high quality references and, where possible, images I have myself taken. I have also created ways of making unwrapping and baking quicker and much more reliable. I am pleased with how this project has ended and feel ready to tackle my FMP. Below is the original Reference compared to the final version of the model.

The most successful element of this project however, i feel, has to be that the train itself matches very closely the original source reference. I set to make a replica of the locomotive from little to no plans, working mostly by eyeball from grainy reference images. This project has give me the confidence the try bigger and more.

Leave a Reply

Your email address will not be published.