Optimized Image batching for Starling.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


An optimized Image batch class for Starling.

Note: Beta code and API

The ImageBatch can efficiently (in terms of CPU and memory) render a large number of images sharing the same texture. This class is designed as an object pool: items are recycled and vectors are never modified.

Try the demo: 8k particles with position/alpha/rotation animation and different textures from a shared spritesheet.

If you're unimpressed, that's ok - just notice that Starling rendering model adds an important overhead compared to heavily optimized demos or pure GPU particles (like ND2D offers). But here you can animate each element independently, and this is a big difference compared to animating a few high-polys models.


  • low memory footprint, items pooling
  • animate: x, y, scale, alpha, color, rotation, texture
  • (shared) spritesheet support
  • store custom data in items' "tag" property


Starling TextureAtlas always returns new Texture instances - you must cache them if you manage animations yourself.

Important implementation detail:

The batch has a vector of items (the pool) and a 'count' index indicating how many items from the pool to render. When removing an item, it isn't actually removed:

  • the item to remove, say at index 'i', is swapped with the last item, at index 'count-1',
  • count is decreased because now the last item should not be rendered,
  • if you were iterating the items vector, the item at index 'i' should be processed again (if 'i' is still < count).


This class is inspired by Starling's Particle System extension but with geometry building inlined.