⚓ Capitan Linear
The zero-shift sigma schedule for when you want no heroics
- SIGMAS
Capitan Linear is the plainest node in the CapitanFlowMatch pack, and it's meant to be. Where Capitan Presets picks a model-specific schedule for you and Capitan Shift lets you dial in a shift value, this one just outputs a straight 1.0 → 0.0 linear schedule and calls it a day. It's the "euler + linear" of the flow-matching world - boring on purpose.
One input, one output
steps- INT, default 8, range 1–10000.- Output:
sigmas(SIGMAS), feeding SamplerCustom.
The mechanism is a single line: steps + 1 evenly spaced values from 1.0 down to 0.0. No shift formula, no model table, no CFG output. If you've already internalized the shifted schedule from this pack's other nodes, think of this as shift = 1.0 - the special case where every step removes an equal amount of noise.
Why you'd reach for it
Three uses, roughly in order of how often you'll actually need them.
First, the README's stated purpose: models that expect unshifted schedules. Not every flow model wants its sigma curve pushed around, and for those this is the correct baseline.
Second, as a control in an experiment. If you're trying to figure out whether shift is helping your model at all, run CapitanLinear at a given step count and compare against CapitanShift at the same step count. Identical output? Your model doesn't want shift - keep the simple one. That's a genuinely useful debugging move, and this node makes it a two-minute change.
Third, distilled models at low step counts. The default of 8 steps is tuned for that world - Z-Image-Turbo and similar fast checkpoints - where a plain linear schedule over few steps is already close to right.
Wiring and gotchas
sigmas goes into SamplerCustom's sigmas port. You still need to pick a sampler - capitan_flowmatch or stock euler both work; this node is only the schedule, not the whole sampling stack. And remember the standard trap: KSampler has no sigmas port, so if you connect this to one, nothing happens.
If the output looks flat or washed out with this node, that's usually the model telling you it does want shift. Swap to CapitanShift or the presets rather than fighting the linear schedule. If it looks overcooked, check CFG - that's a guidance problem, not a schedule problem.
Install
Same as the rest of the pack. ComfyUI Manager → search ComfyUI-CapitanFlowMatch → install, restart. Or:
cd ComfyUI/custom_nodes
git clone https://github.com/capitan01R/ComfyUI-CapitanFlowMatch
Restart ComfyUI. One install covers all three nodes plus the samplers, and there are no models or pip dependencies to chase down. It's a single pure-Python file.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| steps | INT | 81–10000 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| SIGMAS | SIGMAS | — |