From a403b513e66403fb0d57f5e46480e8408b2da960 Mon Sep 17 00:00:00 2001
From: kongo_jun <160564552+lovemachine100@users.noreply.github.com>
Date: Thu, 9 Jul 2026 18:19:45 +0900
Subject: [PATCH] README: clarify silent-skip behavior without
 --disable-dynamic-vram

Verified on-device (2026-07-09): with the flag the node swaps 20/40 blocks
(13.4 GB pinned) and completes; without it the workflow still runs to
completion with only a console warning, which is easy to miss from the UI.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 README.md | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index eb03b6d..264da7f 100644
--- a/README.md
+++ b/README.md
@@ -34,8 +34,14 @@ block to the GPU only while its forward runs.
 ## Notes / limitations
 
 - Chain: `UNETLoader → LoraLoaderModelOnly → BlockSwap → ModelSamplingSD3 → KSampler`.
-- Requires `--disable-dynamic-vram` (legacy ModelPatcher). With dynamic VRAM
-  enabled the node logs a warning and does nothing.
+- **Requires launching ComfyUI with `--disable-dynamic-vram`** (legacy
+  ModelPatcher). Without the flag the node does NOT error or stop the
+  workflow — it passes the model through unchanged and prints
+  `[BlockSwap] dynamic VRAM patcher detected, skipping.` to the console
+  only (nothing is shown in the browser UI). The run then relies on
+  ComfyUI's standard memory management: small jobs may still succeed,
+  large models will OOM or fall back to slow partial loading. If block
+  swap "doesn't seem to work", check the console for this line first.
 - Works on any model whose diffusion model exposes `.blocks` (Wan, LTX-style
   DiTs, Flux double blocks are NOT covered — Wan-family tested).
 - Do not chain two BlockSwap nodes on the same model.