Mixing Smaller Renders

For a still image, a possible soulution is to render less samples multiple times with the seed changed each time, and mix the images according to the samples. For example:

You have a scene which you would like to render at 3000 samples.

• First night you render 1000 samples with seed 0.
• Second night you render 1000 samples with seed 1.
• Third night you render 500 samples with seed 2.
• Fourth night you render 500 samples with seed 3.

The you take an image editor and put these images as layers:

• First render: alpha 0.33
• Second render: alpha 0.5
• Third render: alpha 0.5
• Fourth render: alpha 1.0

Each layers' alpha is: current samples / (all layers samples below + current samples)

5 samples:

15 samples:

Mixed image:

Added Note

As mentioned in the comments: The resulted image may have a slight lighting-loss compared to the full sampled image, however will also have less noise. You get (almost, very near, way below significance) the same result as using Clamp Direct and Clamp Indirect with the number of samples of the individual renders. So this lighting-loss will be the same as if you were to set Clamping to 100 for example, which is usually very small. When the Clamp values are already set and lower than the samples, then no additional lighting-loss should occur.

We discussed this topic in the comment section of some older blog posts, it might be worth reading them:

https://sinmantyx.wordpress.com/2015/02/21/grave-of-the-blender-fireflies/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-1/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-2/

Mixing Smaller Renders

For a still image, a possible solution is to render less samples multiple times with the seed changed each time, and mix the images according to the samples. For example:

You have a scene which you would like to render at 3000 samples.

• First night you render 1000 samples with seed 0.
• Second night you render 1000 samples with seed 1.
• Third night you render 500 samples with seed 2.
• Fourth night you render 500 samples with seed 3.

The you take an image editor and put these images as layers:

• First render: alpha 0.33
• Second render: alpha 0.5
• Third render: alpha 0.5
• Fourth render: alpha 1.0

Each layer's alpha is: current samples / (all layers samples below + current samples)

5 samples:

15 samples:

Mixed image:

Added Note

As mentioned in the comments: The resulted image may have a slight light-loss compared to the full sampled image, however will also have less noise. You get (almost, very near, way below significance) the same result as using Clamp Direct and Clamp Indirect with the number of samples of the individual renders. So this light-loss will be the same as if you were to set Clamping to 100 for example, which is usually very small. When the Clamp values are already set and lower than the samples, then no additional light-loss should occur.

We discussed this topic in the comment section of some old blog posts, it might be worth reading them:

https://sinmantyx.wordpress.com/2015/02/21/grave-of-the-blender-fireflies/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-1/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-2/

Mixing Smaller Renders

For a still image, a possible soulution is, to render less samples multiple times with seed changed, and mix the images according to the samples. For example:

You have a scene, you would like to render 3000 samples.

• First night you render 1000 samples with seed 0.
• Second night you render 1000 saples with seed 1.
• Third night you render 500 samples with seed 2.
• Fourth night you render 500 samples with seed 3.

The you take an image editor and put these images as layers:

• First render: alpha 0.33
• Second render: alpha 0.5
• Third render: alpha 0.5
• Fourth render: alpha 1.0

Each layers alpha is: current samples / (all layers samples below + current samples)

5 samples:

15 samples:

Mixed image:

Added Note

As mentioned in the comments: The resulted image may have a slight light loss compared to the full sampled image, but will also have less noise. You get (almost, very near, way below significance) the same result as using clamp direct and clamp idirect with the number of samples of the individual renders. So this light loss will be the same as if you would set clamping to 100 for example, which is usually very small. When the clamp values are already set and lower than the samples, then no additional ligh loss should occure.

We discussed this topic in the comment section of some old blog posts, it might be worth reading them:

https://sinmantyx.wordpress.com/2015/02/21/grave-of-the-blender-fireflies/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-1/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-2/

Mixing Smaller Renders

For a still image, a possible soulution is to render less samples multiple times with the seed changed each time, and mix the images according to the samples. For example:

You have a scene which you would like to render at 3000 samples.

• First night you render 1000 samples with seed 0.
• Second night you render 1000 samples with seed 1.
• Third night you render 500 samples with seed 2.
• Fourth night you render 500 samples with seed 3.

The you take an image editor and put these images as layers:

• First render: alpha 0.33
• Second render: alpha 0.5
• Third render: alpha 0.5
• Fourth render: alpha 1.0

Each layers' alpha is: current samples / (all layers samples below + current samples)

5 samples:

15 samples:

Mixed image:

Added Note

As mentioned in the comments: The resulted image may have a slight lighting-loss compared to the full sampled image, however will also have less noise. You get (almost, very near, way below significance) the same result as using Clamp Direct and Clamp Indirect with the number of samples of the individual renders. So this lighting-loss will be the same as if you were to set Clamping to 100 for example, which is usually very small. When the Clamp values are already set and lower than the samples, then no additional lighting-loss should occur.

We discussed this topic in the comment section of some older blog posts, it might be worth reading them:

https://sinmantyx.wordpress.com/2015/02/21/grave-of-the-blender-fireflies/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-1/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-2/

Mixing Smaller Renders

For a still image, a possible soulution is, to render less samples multiple times with seed changed, and mix the images according to the samples. For example:

You have a scene, you would like to render 3000 samples.

• First night you render 1000 samples with seed 0.
• Second night you render 1000 saples with seed 1.
• Third night you render 500 samples with seed 2.
• Fourth night you render 500 samples with seed 3.

The you take an image editor and put these images as layers:

• First render: alpha 0.33
• Second render: alpha 0.5
• Third render: alpha 0.5
• Fourth render: alpha 1.0

Each layers alpha is: current samples / (all layers samples below + current samples)

5 samples:

15 samples:

Mixed image:

Mixing Smaller Renders

For a still image, a possible soulution is, to render less samples multiple times with seed changed, and mix the images according to the samples. For example:

You have a scene, you would like to render 3000 samples.

• First night you render 1000 samples with seed 0.
• Second night you render 1000 saples with seed 1.
• Third night you render 500 samples with seed 2.
• Fourth night you render 500 samples with seed 3.

The you take an image editor and put these images as layers:

• First render: alpha 0.33
• Second render: alpha 0.5
• Third render: alpha 0.5
• Fourth render: alpha 1.0

Each layers alpha is: current samples / (all layers samples below + current samples)

5 samples:

15 samples:

Mixed image:

Added Note

As mentioned in the comments: The resulted image may have a slight light loss compared to the full sampled image, but will also have less noise. You get (almost, very near, way below significance) the same result as using clamp direct and clamp idirect with the number of samples of the individual renders. So this light loss will be the same as if you would set clamping to 100 for example, which is usually very small. When the clamp values are already set and lower than the samples, then no additional ligh loss should occure.

We discussed this topic in the comment section of some old blog posts, it might be worth reading them:

https://sinmantyx.wordpress.com/2015/02/21/grave-of-the-blender-fireflies/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-1/

https://sinmantyx.wordpress.com/2015/03/18/perfect-clamp-2/