The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Add new attention and compiler stuff to AGENTS.md (#16762)
comfyanonymousf1072eb03506
diff --git a/AGENTS.md b/AGENTS.mdindex 51dd6dd65..abee3157f 100644--- a/AGENTS.md+++ b/AGENTS.md@@ -186,6 +186,10 @@handwritten implementations of the same operation. Remove duplicate localkernels and adapt inputs to the shared operation's documented layout whilepreserving the model's original math and output contract.+- Model implementations must use `AttentionTensorContainer` and per-module+ `self.comfy_attention = ComfyAttention()`, passed to attention as `preferred_attention`.+- Integrate model block loops with the memory compiler and prefetch helpers in+ `comfy.model_prefetch`, following existing model patterns.- All models should use the optimized attention function selected by ComfyUI.Treat optimized backend functions, dispatch helpers, and capability-selectedcallables as opaque. Higher-level code must not inspect function identity,