TorchCompileModelFluxAdvanced

If you found this node in an old workflow and it's not doing anything you can inspect, that's expected: it says so itself, "This node has been replaced with TorchCompileModelAdvanced node, please use that instead." This is the middle rung of a three-step lineage inside KJNodes - plain TorchCompileModelFlux came first, then this Advanced variant, then TorchCompileModelFluxAdvancedV2 with a proper set of exposed settings, and now all of that has been folded into one unified TorchCompileModelAdvanced that isn't Flux-specific at all. Each step was kijai consolidating what had been separate per-architecture nodes into fewer, more general ones - a pattern you'll see across the pack. If you want the actual dial-in options (compile mode, backend, which transformer blocks get touched), that's the V2 node's article, not this one; this node's current schema doesn't expose any of that anymore.

What it was for

Flux's diffusion transformer is a heavy model to run eagerly, and torch.compile is PyTorch's built-in way to claw back speed: it traces the model's computation graph once and fuses the operations into optimized CUDA kernels instead of executing them one at a time in Python. The catch is that the first pass through the compiled model is slower than normal, because that's when tracing and kernel generation actually happen. Every subsequent generation at the same resolution reuses those compiled kernels and comes out ahead - sometimes meaningfully so on a busy generation session, not at all on a single one-off render.

Inputs and outputs

By the time this node was marked deprecated, its schema had been trimmed to a bare pass-through: one required input, model, typed as a wildcard rather than something Flux-specific, and a single wildcard output that mirrors it. Whatever settings this node exposed while it was still actively maintained - and Flux compile nodes in this pack did expose real settings, see the V2 article - aren't present in the current registration. You get model in, model out, nothing to configure.

How to install it

It lives inside the same KJNodes pack as everything else here - no separate install for this one node. Through ComfyUI Manager: search "KJNodes for ComfyUI," install, restart. Manually: cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-KJNodes, then pip install -r ComfyUI-KJNodes/requirements.txt (or the python_embeded\python.exe -m pip install -r ... form on the portable Windows build), then restart.

Common issues & troubleshooting

It's in an old workflow and doing nothing useful. That's the point of the deprecation notice - replace it with TorchCompileModelAdvanced rather than debugging it. A deprecated node with a stripped-down schema isn't going to get fixes.

You actually want the tunable version. Look for TorchCompileModelFluxAdvancedV2 in your node search, or better, go straight to TorchCompileModelAdvanced, which is the one still being maintained.

General compile prerequisites, if you're testing any of these Flux-compile nodes side by side. torch.compile needs Triton installed and, on Linux, typically a working C compiler (build-essential covers it). Windows setups hit this wall far more often - community-built Triton wheels are usually the only way in, and it's the same friction people run into trying to get SageAttention working.