Rendered Image Off by One Pixel

I’m using the holdout option in Embergen to render the simulation so I can composite it in Blender. The problem is when I bring the rendered simulation into Blender, it’s off by about one pixel. This is just a render of a box in front of a fire, but you can see at the top how it doesn’t match up.

Hey thanks for letting us know about this. Can you send us the blender and embergen file so that we can troubleshoot it? You may need to offset the start frame in Blender but there are also like 3 or 4 other things it could be

I tried it again with a stationary camera and still got the same result so I don’t think it’s the offset.

Okay thanks for this. I think the real cause here is the anti-aliasing in Blender. You can see in my screenshots that with the contrast of the fire in the background it looks like there is a gap between the two. But if I remove the EmberGen layer, you see a faint edge of the box still there.

I believe this is due to the Anti-Aliasing (or lack of AA) in Blender. I tried fixing it by increasing the min samples but i wasn’t very successful. I’ll continue to try other things in the meantime. Let us know if you find a solution or workaround!

This was my thinking as well. The problem is that in the file that I actually need this for, the gap is much more dramatic and seems to get worse as the scene progresses.

This isn’t the anti aliasing. This is off by a pretty good amount. Does anyone have any thoughts on what would cause this?
Car Picture

Hey sorry for the delay but thank you for the reminder! I am seeing more issues with the project you provided, but not as much as the screenshot you sent. I am not sure what is going on with Blender but something tells me that the frame rate is messing with the capture here. It doesn’t look as bad when i import the assets into After Effects so whatever the problem is, Blender is making it worse somehow…

One thing I’m noticing is that if I make the car larger to fill more of the bounding box, I seem to get less of this effect

One thing that may work is to render the Render All channel, and instead of using ‘holdout’ for your shapes, you can set that to ‘ignore’ and add that alpha shapes capture to your render output. That way you have more ability to manipulate that mask to make sure that the mask appropriately matches the Blender output.

This certainly isn’t a long term solution but this will help bridge that gap until we come up with a solution.

So this is just a bug? Ok, I’m glad it wasn’t something I did.