From a8b4fa64104a4d834c51396eef5c62518b7691c9 Mon Sep 17 00:00:00 2001
From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com>
Date: Mon, 31 Aug 2026 16:09:43 -0700
Subject: [PATCH] Add how to cache model stuff to AGENTS.md (#16003)

---
 AGENTS.md | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/AGENTS.md b/AGENTS.md
index b63b7a1d4..51dd6dd65 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -297,6 +297,12 @@
 - Avoid caches that persist across different executions as much as possible.
   Persistent caches are acceptable only when they use a very minimal amount of
   memory and have a clear ownership and invalidation story.
+- When condition-dependent model work would otherwise repeat on every denoising
+  step and preprocessing it once materially improves performance, expose a
+  model preprocessing method and call it from `BaseModel.extra_conds`, following
+  patterns such as LTXAV and Anima. Pass the result through normal conditioning;
+  do not add model-owned caches, sampler-option caches, or cache-management
+  wrappers for this work.
 - When optimizing, favor small measurable changes: fewer allocations, fewer
   device transfers, less peak memory, better batching, or use of a faster
   existing backend op.