ComfyUI launch/performance flags

Overview

ComfyUI's runtime behavior is controlled by CLI flags passed to main.py (e.g. python main.py --reserve-vram 2 --use-sage-attention). The three that matter most for making a graph run — rather than OOM or crawl — are the VRAM strategy, the attention backend, and the cache mode. This skill is the decision matrix for choosing them.

> ⚠️ Verification note (June 2026). Every flag below was checked against > upstream comfy/cli_args.py. > ComfyUI adds/renames flags often — when in doubt run python main.py --help > in the target install and prefer that over this list. One common non-upstream > flag: --enable-triton-backend is a SwarmUI backend flag, NOT a ComfyUI > main.py flag — don't pass it to ComfyUI directly.

> ℹ️ How to apply today. The MCP's restart_comfyui (with action: "start") > currently replays the exact argv of the previous run — it does not compose > fresh flags. So set these when you launch ComfyUI yourself (the > python main.py … line, a run.bat/shell alias, or the SwarmUI backend args > box), and the tool will preserve them on restart. (Injecting flags through the > tool is a tracked follow-up.)

Decide first: which flag do you need?

Symptom                                             ▶ Flag(s) to try
─────────────────────────────────────────────────────────────────────────────
CUDA out of memory, long video (LTX 2 / WAN)        ▶ --novram  (+ --cache-none)
OOM, still want models resident when they fit       ▶ --reserve-vram N  then --disable-smart-memory
GPU slows to a crawl, spills into "shared GPU        ▶ --reserve-vram 2..4
  memory" (Windows WDDM) mid-run
RAM blows up switching between models, or a huge     ▶ --cache-none
  text encoder (FLUX 2 / Mistral) won't unload
Plenty of VRAM (48GB+), want max throughput         ▶ --gpu-only  or  --highvram
Want faster sampling on NVIDIA                       ▶ --use-sage-attention   (see caveats)
Z-Image produces BLACK / wrong output               ▶ --use-pytorch-cross-attention (NOT sage)
Sage gives black output on some models              ▶ --use-pytorch-cross-attention (or fix dtype)

VRAM strategy and attention backend are each mutually exclusive groups — pass at most one from each. You can combine one VRAM flag + one attention flag + one cache flag (e.g. --novram --use-sage-attention --cache-none).

VRAM strategy (mutually exclusive)

| Flag | What it does | Use when | |------|--------------|----------| | --gpu-only | Keep everything (incl. text encoders) on GPU | 48GB+ card, single model, max speed | | --highvram | Keep models resident in VRAM after use | High-VRAM card, repeated runs of one model | | (default) | ComfyUI's smart offload | Most setups — try this first | | --lowvram | Offload text encoders / parts to CPU | Mid card OOMing on load | | --novram | Extreme offload — minimal VRAM footprint | OOM on long video / huge models; pair with --cache-none | | --cpu | Everything on CPU (very slow) | No usable CUDA GPU only |

Modifiers (combine with the above):

• --reserve-vram N — reserve N GB for the OS / other apps. The fix for the

Windows failure mode where the GPU quietly starts using shared VRAM and throughput collapses. Typical 2–4; bump to 10 for heavy video decode.

• --disable-smart-memory — force aggressive offload to regular RAM instead

of keeping models cached in VRAM. Reach for this when a run gets stuck or OOMs intermittently. Slightly slower, much more robust.

• --async-offload — async weight offload streams (default on where

supported); --disable-async-offload to turn off if it misbehaves.

Attention backend (mutually exclusive)

| Flag | Notes | |------|-------| | --use-sage-attention | Quantized SageAttention kernel, ~20–40% faster sampling. Needs the sageattention package installed and version-matched — see triton-sageattention. | | --use-flash-attention | FlashAttention kernels. Needs flash-attn built for your torch/CUDA. | | --use-pytorch-cross-attention | PyTorch SDPA. Highest quality, always available, no extra deps. The safe default and the correct fallback. | | --use-split-cross-attention / --use-quad-cross-attention | Memory-optimized math attention for older/low-VRAM cards. |

Two gotchas worth memorizing:

1. Z-Image + Sage = broken. Z-Image (Turbo/Base) does not sample

correctly under --use-sage-attention — you get black or garbled output. Launch Z-Image with --use-pytorch-cross-attention instead. See z-image-txt2img.

1. Sage black output on other models. If a model outputs black only with

Sage, either switch to --use-pytorch-cross-attention, or (SwarmUI) set Advanced Sampling → Preferred DType = Default (16-bit). Sage-on vs Sage-off also produces slightly different images — expect non-identical seeds.

> When a graph hard-crashes with No module named 'sageattention' / > triton: unavailable, the fix is the sdpa / no-compile fallback in > triton-sageattention, not this flag.

Cache mode (mutually exclusive)

| Flag | Effect | |------|--------| | (default `--cache-ram`) | Cache results under RAM pressure | | --cache-classic | Aggressive result caching | | --cache-lru N | Keep at most N node results (LRU) | | --cache-none | Cache nothing — re-executes every node; lowest RAM/VRAM. Essential when switching between dual models or when a giant text encoder (FLUX 2's Mistral) must fully unload. |

Speed / precision

• --fast — enables experimental, potentially quality-degrading

optimizations. Accepts specific PerformanceFeature values: fp16_accumulation, fp8_matrix_mult, cublas_ops, autotune. Bare --fast turns them all on. Test output quality before committing to it.

• UNet/VAE/text-encoder dtype casts exist too

(--fp8_e4m3fn-unet, --fp16-unet, --bf16-unet, --fp32-unet, …) for forcing a compute precision; usually the model/loader picks the right one, so only reach for these to work around a specific dtype error.

Recommended combos (recipes)

Long video OOM (LTX 2 / WAN, 24GB):   --novram --cache-none
                                      (add --disable-smart-memory if it stalls)
Windows shared-VRAM creep:            --reserve-vram 3
FLUX 2 / huge text-encoder swaps:     --cache-none
High-VRAM throughput (48GB+):         --gpu-only        (or --highvram)
Fast NVIDIA sampling (most models):   --use-sage-attention
Z-Image (any):                        --use-pytorch-cross-attention

Cross-refs: video OOM specifics in ltxv2-video / wan-t2v-video; per-model VRAM math in troubleshooting and model-compatibility.

Acceleration stack & GPU coverage (context)

The attention/compile accelerators are version-locked to your exact torch + CUDA + Python. A mismatched wheel doesn't just fail to import — it can break the torch install. A known-good, mutually-compatible stack for late-2025 / 2026 NVIDIA (including Blackwell / RTX 5000, sm_120) looks like:

| Component | Role | Notes | |-----------|------|-------| | Torch + CUDA | base | e.g. Torch 2.9.x on CUDA 12.8/13; use the wheel index matching your driver | | Triton | torch.compile / inductor | Windows: triton-windows (woct0rdho) | | SageAttention | --use-sage-attention | wheel matched to torch/CUDA/python | | FlashAttention | --use-flash-attention | built per torch/CUDA/python | | xFormers | memory-efficient attention | optional | | InsightFace | FaceID / IP-Adapter / ReActor | onnxruntime-gpu alongside |

Operational facts worth carrying:

• No system-wide CUDA toolkit is required to run ComfyUI — an up-to-date

NVIDIA driver + prebuilt wheels are enough. A full CUDA/MSVC/cuDNN toolchain is only needed to compile kernels yourself.

• Broad arch coverage when building wheels:

TORCH_CUDA_ARCH_LIST=7.5;8.0;8.6;8.9;9.0;10.0;12.0+PTX spans RTX 20xx→50xx and datacenter (A100/H100/B200). +PTX lets newer archs JIT.

• DeepSpeed has no wheels for Python 3.13; several accel wheels lag the

newest Python — 3.10–3.12 is the safe range for the full stack.

• Clear the Triton cache (~/.triton / %USERPROFILE%\.triton and temp)

when you hit stale-kernel Triton errors after an upgrade.

• Prefer uv pip install over pip for the venv — dramatically faster

resolves/downloads. install_comfyui already supports this via preferUv.

• A single bad custom node can crash all of ComfyUI at startup. Install/test

acceleration and new node packs on a fresh/known-good install, not before a deadline. See troubleshooting.

Quantization quick take

• FP8-scaled (per-tensor scaled) is markedly higher quality than plain

base FP8, ~half the size of BF16, and usually faster.

• Prefer FP8-scaled over GGUF when you have enough system RAM — ComfyUI's

block-swap streams from RAM, so BF16/FP8 can run on 24GB GPUs given ample RAM. Fall back to GGUF (Q8→Q4) only when RAM is the constraint.

• NVFP4 / NVFP8 are markedly faster on Blackwell (RTX 5000) at near-BF16

quality for supported models; LoRA support on NVFP4 is still partial.

Sources

• ComfyUI CLI args (authoritative): <https://github.com/comfyanonymous/ComfyUI/blob/master/comfy/cli_args.py>
• ComfyUI startup flags docs: <https://docs.comfy.org/development/comfyui-server/startup-flags>
• Operational flag/stack guidance distilled from community ComfyUI auto-installer

changelogs (SECourses) — flags cross-checked against upstream above; no third-party scripts, presets, or model files are reproduced here.