Impatient Procrastinators Visual Effects.



  • @spydurhank Windows 10 version 1709. Nvidia driver 388.13

    The only thing I did was raise the diffuse color and try cook torrance.

    I did import the OBJ version of the red+white model. It is about 200K less surfaces than the FBX composite shot version. I'm not sure what you are seeing but I can try the FBX composite shot.

    I've always seen jagged shadows up close. Other models as well. Even with my previous video card(s). A couple of cards ago was an AMD 7950 but that is too long ago to remember specifics. That project was using the particle sim with 3D models and the particle sim really craps out shadows. That's all I remember from that time. It did crap out on the AMD. The project on Nvidia does crap out now. Support reproduced as well on their hardware.

    No particle sim enabled.

    With particle sim. Model shown is not a particle sim model. It is a hero ship.


  • @NormanPCN

    Ooh, ooh, try raising the shadow bias to 26 in the models material settings in the control panel other than that, it looks like your shadow map is being rendered at 512 or 1024, ya know? Something really low.

    That shadow stepping is weird though, your shadows should fly colors around my shadows. You've got a 1080 dude! :)

  • Thanks for reminding me about the self shadow bias setting.  Of course it reduces the excess self shadowing. I looked at an old project and I was using a value of 25 for self shadow.

    It doesn't do anything with Hitfilm sawtooth shadows when up very close and personal. They are what they are. Who knows why they are so jagged. Anyway, we normally don't get that close to the model, so we don't see them.

    Don't confuse the extra jagged particle sim example I showed. That is a different issue. Actually a couple of things but I digress.

    Hitfilm shadows do have some issue(s). They are a quick (performance) and dirty (limitations) approximation. Point lights have the biggest limitations. They are suspect with distance. e.g. A point light should converge to a directional light as distance increases. Does not happen in Hitfilm. The point shadows absolutely falls apart and become very jagged and large as distance increases. Directional lights are okay. Spot lights can be okay depending on the spot angle and distance of course. Anyway, I put a project together to demo this for support. I briefly installed Blender to test with my X-wing model to compare against Hitfilm. Figuring out enough of Blender was a pain. 


  • Ooh... Blender, it is a different beast. That's why I build my models specifically for HitFilm use although they'll import into any other dedicated 3D app via the .obj and .fbx files. 

    You think maybe that shadow map stepping is a bug? 

  • Shadow map stepping isn't a byg, it's a limitation of the tech. The shadow map is rasterized to the dimensions set in the project properties, but the shadow map is calculated from the position of the light. This is why increasing distance increases shadow jags, and why point lights have the most issues--because it's creating a spherical panoramic map.

    Three basic ways to improve shadows. Increase map size, increase shadow diffusion, or create a master point, parent everything to it and scale up the whole scene and try sneaking lights closer. 

    Blender, of course, does full raytracing (or radiosity, depending on the renderer). We can hope Hitfilm adds raytracing later. There are those who will take the performance hit for quality.

    All the above applies to refractions as well. 

  • edited December 2017

    Stumbling in late - couldn't agree more "We can hope Hitfilm adds raytracing later" +1

    You can get some pretty decent results with the shadows up close if you keep the lights tight and pump up the shadow map to 4096.

    I'm definitely liking your models @Sypdurhank. Although maybe a bit taxing on the load at first nothing I can really complain about. You did a really great job on the geometry and textures can get up pretty darn tight and still maintain quality. I did have to mess with the normals a bit to swash the fireflies during close and personal movement but nothing that can't be dealt with :-) 

  • @GrayMotion

    Glad you like them. For sure, you will need to adjust the smoothing in the .fbx model. The .obj should be fine but apparently results vary. :)

  • Thank you so much for the freebie- really appreciate it! I won't have time this week to play around with it but I will definitely be all over this when I get some free time. Finally finishing up the project that's kept me busy for the past year or so.


  • Coming in really late on this, and piggybacking on the comments from @Triem23 re: general shadow map behavior, and pulling from somewhat basic shadow map experience in Maya, and admitting that I've not done a lick of 3D model importing in HitFilm yet, so this is pure speculation... :)

    Looking at the most recent stills posted by @NormanPCN, I wonder if the increased shadow jags when the particle sim is active are caused by HitFilm applying the same shadow map across several models instead of just one.  In other words, a 4K map used to shadow the single X-wing cause the shadow steps to be much smaller than the same map spread across several models generated by a simulation.  In other other words, perhaps HitFilm's shadows are calculated based on the full scene content, so that an individual object in a multi-object scene will have worse shadows than the same object by itself.

    That said, I ran a quick test of this theory using planes, and said theory doesn't appear to hold a single drop of water, so maybe I'm high on something.

  • @jsbarrett I didn't know that about HitFilm shadows.

  •  @spydurhank Just to reiterate, I was simply making a wild guess, and only tested my guess (and failed) using planes.  Maybe actual 3D models are treated differently and the theory actually works, but I haven't gone that far yet.

  • Gotcha, I want to test that out too. :)

  • Please do, because it's going to be a LONG time before I get to it. I don't currently have any 3D models, and have numerous other things on my priority list higher than testing this theory.

Sign in to comment