FlippedNormals published "What You Get Wrong About Smart Materials in Substance Painter" on 14 February 2019. It is thirteen minutes long, it has been watched more than two hundred thousand times, and their own description says beginners "rely far too much on smart materials and smart masks", disregarding reference and the story of the model.
I make the things they are describing. There are twenty-nine volumes of Substance Painter smart materials in my shop — copper, clay, leather, plaster, plastic, wood, rusty metal — and I have looked at several thousand student files that used them badly. So I am the last person who should agree with that video. I agree with about half of it, and I think the half most people took away from it is the wrong half.
What they get right
The symptom is real and I see it every week. A student drags a smart material onto a mesh, the preview looks convincing at 400 per cent zoom, and the asset is dead on arrival. Everything on it is the same age. The edge wear runs along edges no hand ever touched. Dirt sits in crevices that would have been wiped clean by the same hand that supposedly wore those edges. Nothing on the model has a history, because nobody decided what its history was.
Their chapters run Smart Materials, then Brushes, then Individual Variation, and uniformity is the correct diagnosis. If two bolts on the same plate have identical wear, the viewer knows something is wrong before they can say what.
Notice what they actually claim, though. The complaint is about reliance, not about the tool. That distinction got lost somewhere between the video and its comment section, and the version that spread was "smart materials are for beginners". That version has cost people more time than the habit it was meant to cure.
What a smart material actually is
Here is where I part company, and it is a definitional disagreement rather than a matter of taste.
A smart material is not a texture. It is a saved layer stack whose masks are bound to your mesh maps — curvature, ambient occlusion, thickness, position. When you drop one on an asset you have not applied a surface. You have applied an opinion about how this kind of surface ages, written as generators, and it arrives ready to be argued with.
Treating that opinion as a finished texture is the whole error. But "stop using them" is not the correction, any more than the correction for a badly traced reference photo is to stop looking at reference.
The default that ruins it is scale
If I could change one setting on every student file that arrives looking like painted plastic, it would not be roughness, and it would not be generator opacity. It would be scale.
I author at 4096 by 4096. The scratches, the grain in a plaster, the brush direction in a satin metal are all sized for the object I built them on. Put that stack on an asset with a different texel density and every frequency in it is wrong. Too fine and the metal reads as woven cloth. Too coarse and the whole prop reads as a toy. The material is not fake. The material is the wrong size.
So fix the projection scale in the fill layer before you touch anything else, then look at the asset from the distance the camera will actually sit at, rather than from 400 per cent zoom where you have been judging it. A good share of the "smart materials look fake" complaint dies right there, and it costs about eleven seconds.
The bake underneath it matters at least as much, but I have already written about mesh maps and will not repeat that here.
This is my one narrow disagreement with the video itself, rather than with its audience. The order of the fixes is wrong. Brushes and hand-made variation are the expensive corrections, and they come after two cheap ones: a bake that is not muddy, and a stack that is the right size for the object.
Where hand-painting everything stops working
This is the real objection.
If the remedy is to build every material by hand, from brushes, per asset, then it works beautifully for one hero prop in a portfolio and it does not survive a schedule. My studio is four people. A prop set is not one steel pot. It is thirty objects that have to agree with each other about which factory they came from and how many winters they have stood outside. Hand-authoring thirty surfaces from scratch does not produce more consistency than a shared stack does. It produces less of it, and it takes a fortnight.
The professional answer is neither extreme. It is a base you did not have to invent, followed by a deliberate spend of hand-work on the ten per cent of the asset that carries the story — a custom mask painted in Photoshop for the one area the generators will not reach, after the stack is down rather than instead of it. The longest free thing on my channel, a 59 minute walkthrough called "How to increase the speed of texturing In Substance Painter", is an hour on the same question: where the time actually goes.
Hand-painting is not the opposite of a smart material. It is the second half of one.
Judge a pack by what it lets you change
All of which puts the burden back on people like me, because a smart material you cannot argue with is a bad product, and the criticism in that video lands hardest on exactly those.
The test I would apply to anything you download, mine included:
- Are the generators still exposed, or have they been flattened into a layer you cannot get behind?
- Is the aging a parameter? In 30 Copper Smart Materials & PBR Textures — VOL 16, oxidation is exposed as a parameter, because the difference between a two-year roof and a forty-year roof is one number and it should be yours to set.
- Does it come apart? If you cannot delete the top three layers and still have a usable base, it is a screenshot of a material, not a stack.
- Does the documentation say what scale it was authored at?
Number four is the one almost nobody answers, and it is the cheapest thing on the list to fix. The information a buyer needs in order to use a stack correctly is a line of documentation, not a feature of the material.
Where they are still right
None of this rescues an artist who has not decided what the object is.
No stack will tell you that this rifle belonged to someone who cleaned it obsessively and dropped it once. It will not tell you which edge the strap wore smooth and which edge never touched anything for forty years. That decision sits upstream of every tool in the program, and if you have not made it, no generator, no brush and no amount of individual variation will cover for it.
That is the part of the video that has not aged in seven years, and it is the part I would keep. The part I would drop is the one they never actually said, and that their audience heard anyway — that the tool is the problem. Thirty objects still have to agree with each other about the same factory and the same winters. That is not something you brush on by hand, one asset at a time.
