Sigmas CNFInverse

The name is a real reference: CNF stands for continuous normalizing flow, part of the same mathematical family - flow-based generative modeling, transport between a noise distribution and a data distribution described as an ODE - that gives rectified-flow and flow-matching models their name. RES4LYF's whole reason for existing is that it's built specifically for rectified-flow and probability-flow architectures (Flux, Wan, Qwen-Image, Z-Image, Anima, and the rest of the current generation), so a node named after flow theory fits the pack's DNA even without documentation to back it up.

What it does

It remaps the timing of your sigma schedule through one of four curve shapes - flow_type: linear, quadratic, sigmoid, or exponential - computed over time_steps points, with reverse flipping the direction of the mapping.

Why this matters for flow-matching models specifically

Flow-matching architectures expose a shift parameter that controls how sampling effort splits between composition and fine detail across an otherwise near-straight noise-to-image trajectory - it's roughly what replaced scheduler-curve tuning once models moved off the old DDPM-style Karras/exponential shortlist, and it's genuinely model-specific (a Z-Image Turbo workflow wants a shift well above ComfyUI's default). Sigmas CNFInverse reads like a more general, hand-tunable version of that same idea: instead of one shift number, you get a whole reparameterization curve to apply to the schedule's timing.

The inputs and outputs that matter

• sigmas (SIGMAS, required) - the schedule to remap.
• time_steps (default 20, range 5–100) - resolution of the remapping.
• flow_type (enum: linear, quadratic, sigmoid, exponential; default sigmoid) - the shape of the time curve.
• reverse (default true) - flips the mapping direction.

Output is a single SIGMAS list.

How to install it

• ComfyUI Manager - search "RES4LYF", install, restart.
• Manual - activate your venv, cd ComfyUI/custom_nodes && git clone https://github.com/ClownsharkBatwing/RES4LYF, cd RES4LYF, pip install -r requirements.txt (portable builds: use the embedded pip.exe). Restart.

Common issues & troubleshooting

No README coverage, no community discussion - this is genuinely deep-cut. I couldn't find a single thread anywhere discussing this node, which puts it well past even most of RES4LYF's other sigma-manipulation tools in terms of obscurity.

If you're chasing a flow-matching quality improvement, shift is the better-tested lever to reach for first. It's a single, well-understood number with real community-tested values per model family (the anima/Z-Image/Klein/Flux ecosystem guides all give concrete numbers), whereas this node's four curve shapes have no equivalent track record. Use CNFInverse if you specifically want to experiment with schedule timing beyond what a single shift value can express - not as your first move.

reverse defaults to true, which is easy to miss. If your remapped schedule looks like it's running backward compared to what you expected, that's the setting to check first before assuming something else broke.

Sanity-check before committing to a full render, as with everything else in this part of the pack - Sigmas Count for step count, and a preview of the raw values if your node setup supports one.