I'm not particularly experienced with working to such extremes so there may be a better accepted solution to this but I would probably opt for using the compositor to combine the different elements (very large scale and much smaller scale).
The reasons for the distortion are - as you say - due to floating point innaccuracies. Floating Point representations ...
It appears you are using blender 2.8. In blender 2.92, the set alpha node is updated to allow apply mask mode. To fix this, drop an alpha convert node after the set alpha node. Make sure it is set to convert straight to premul. That should fix your alpha problem.
Png does not work with cryptomatte, as it requires a 32bit float precision which PNG can not provide.
You must use EXR in float32bit !
Blender will save out either a OpenEXR or OpenEXR multilayer. If you want all layers in one EXR file per frame, choose the multipayer option.
If you wish to save out multiple EXR per pass (Diffuse, Specular, Crypto ..) you ...
It looks to me like you’ve got the glare affect set to low. Turn it up. If that doesn’t work, separate the glare effect from the main image, brighten the glare, and then mix them back together, instead of using the glare effect’s automatic mixing.
The problem is that your node setup is ignoring the existing Alpha channel. Instead of overwriting it, you need to multiply it with your mask.
I used the Separate RGBA node to extract the current Alpha value, and a Math node to multiply the two values:
The main problem is that the alpha channel generated by the keying node has to be associated with the picture. Use an Alpha Convert node in straight to premultiply mode.
The artifacts at the top of the image are caused by the compression on the original image. Some compression algorithms for video work by considering similar colors to be one solid block of ...
You can try to connect the image (or image sequence) to the denoise node without using the Normal and Albedo sockets but the result won't be as good as it could be.
In general, Intel's OpenImageDenoise is the best option for now, see how to set it up properly: How to properly connect the Denoise node in Compositor?. Alternativly, you can try using other ...
When you press "setup tracking scene" after solving the camera movement, blender creates two collections: a foreground (with a cube as reference) and a background (a plane to catch shadows).
You can safely delete the background one if you don't need a shadow catcher.
You can't write directly to a "C:", "D:", "E:" drive base root directory in Blender. You need to create a folder. Blender is assuming 3.png is going to be your "root base" folder. This is something implicit Blender does. You can "D:\myFolder\3.png" and it will work. Also, consider changing the number from: &...
I found a simple solution to this problem using Python:
def render(rex_x, res_y):
oldx = bpy.context.scene.render.resolution_x
oldy = bpy.context.scene.render.resolution_y
outpath = bpy.context.scene.render.filepath
bpy.context.scene.render.resolution_x = res_x
bpy.context.scene.render.resolution_y = res_y
I'm not sure how well it would work for your specific image, but you could try flipping the image, adding the glare, then flipping it back. The glare should remain "flipped" (because it was added normally to an upside-down image) while the image gets put back "upright":
Hmm .... blender uses your PNG's alpha channel in a very strange way on output. Even PNG has black&white alpha it is displayed as partially transparent ... almost without transparency. So I would expect your question should be - Why is ignored PNG's alpha channel in my render?
What you can see as "white noise" around alpha edge looks ...
The error is this you are setting the offset to 116, but the start frame is set to 1, so the 14 frames you want to render would be rendered in frames 1 to 14. Otherwise you are getting a repeat of the last frame.
On a side note, it is quite unfortunate that you chose to use png as image format, by doing so you've lost a lot of the information that would ...