After I rendered out all my shots, I comped the background in Nuke. I want to have depth of field for first and second shots. I found these two videos about how to add DOP in Nuke.
https://www.youtube.com/watch?v=oTBOXgdKqzo
I used depth channel which is Z pass in Maya AOVs to add depth of field. I found that my denoised sequence don’t have depth channel anymore. So I shuffled the original sequence to depth channel, and used Copy node to copy the depth channel to the denoised sequence.
And then I exported the camera in Maya, and imported into Nuke to do a 3D project for the sky.
Lighting in this project is really challenging, because I have so many lights that are coming from different directions, and especially it’s a night scene.
I did a test render and it took me more than one hour to render one image, and it’s very grainy. So I looked up on google to find a solution to reduce my render time. I found this video and I tried almost every tip to see if my render time gets faster.
Firstly, I tried to find where my noise come from. I cropped region to render just one small area and tested all the sample values in the render setting, such as diffuse, specular, SSS. If I changed one of the samles and there is no difference, then I would keep a very low sample value. So what I found is that the camera AA and lighting samples are the main reason why I have the noise. But the problem is that if I increase those samples, my render time is still too slow.
In the video, he mentioned that mesh lights produce lots of noise. Almost every lights I used in the scene was mesh lights, so I swithed all my mesh lights to area lights. And it didn’t really reduce my render time.
And then I saw someone on Reddit mentioned Low Light Threshold in Arnold. I go to my render settings and increased the value of low light threshold to the maximum, then my render time incredibly reduced to 9 minutes.
There is definitely some loss of quality or color in the image with high samples of low light threshold. But there is no big difference between these two images, at the same time my render is much more faster now.
Now it’s still grainy but much better than before. In the youtube video, he talked about arnold denoiser is really helpful for removing the noise. So I tried it for the first time and it worked really great.
A comparison of the undenoised and denoised image with arnold denoiser.
I masked out the texts from reference images and then exported as PNG files. I then imported into Maya, and applied aiMixShader to combine both text and the wood texture.
Building 1
Building 2
Building 3
Building 6
Final Scene
I added lights inside the buildings. I deplicated a few more buildings at the end and make a turn to create an end on the street.
Shot01: 8 seconds, wide shot.
Shot 02: 4 seconds, close-up, camera not moving. In this shot, I will animate the lanterns, they will be slightly swinging in the camera.
I imported every building I built as reference into a new scene to get an idea what it will look like. I need to add texts on the boards, and put more items on the street to fill it up.
I started to model my fourth building this week, but I don’t know how to make this hanging cloth.
So I found this tutorial below.
https://www.youtube.com/watch?v=cZbXefPap6U
I followed step by step from the video, it’s actually very simple. I just created a cube and add more subdivisions. And then used ncloth to constrain it to the hanger. But it looks not soft enough.
I also tried to use cloth to animate on this hanging cloth I modelled, but it didn’t work really well. So I just moved the verteices by hand to get the waves and wrinkles.
I modified from last building. I switched the position of the door and window. I added lanterns and one more board. I also changed the color of the textures.
Dom told me that I can use substance painter to achieve the texture I want. Last time, I tried apply the different texture for different parts of the building, but it just apply the same texture to the whole building and I didnt know how to apply different textures separately. He said that I should export each part separately from maya to substance painter. So this time I exported the same kind of the woods all together, and I exported other parts separately, and it worked.
I also found that the color of the textures showing in Maya is different than it’s in substance. I don’t know why, but I just added a aicolorcorrect node to correct the color.
I also did a rough draft of the layout of the scene.
I found two videos of how to make the wrinkles on the lanterns.
So I applied a basic fabric texture, and then I added a bitmap to project stipes on it. And I changed UV projection to Planar projection so the edges will be closing smoothly. In the youtube video, it should be cylinder projection, but there isn’t this option in my substance, so I just chose planar projection.
Now I got the wrinkles on the lantern.
I also added mesh lights inside the lanterns to make them glow.
I talked to Dom this week, I asked him about the texture I want to achieve. He told me to do it in Substance Painter, and what I should do for now is to apply basic textures. So I go back to my first building and do the texures. I also figured out why my texture looked wrong, it’s because my displacement scale was too large.
For my first house, I tried to do texture in Substance Painter but it didn’t work really well. I don’t know how to paint different textures on each part in Substance Painter. Then I tried to texture it in Maya, but it came out like this, which I have no idea why it looks so wrong in render view. It looks not bad in viewport but it just didn’t work in render view.
I’m thinking why we cannot distinguish the fake visual effects even it’s very realistic. Sometimes even the VFX is realistic enough, human eye still cannot realize the fake spot. I would be confused by the VFX in some films.
I found a video about a test of inattentional blindness. I tried to spot the change in the video, but I didnt notice at all. I also know that lots of VFX artists used this inattentional blindness to save their time, so the audience won’t be noticed even if there is a mistake on the VFX.
This Youtube channel talks about the fakeness of the VFX in films.
