GITSScheduler Func Provider

Impact Pack's own KSampler nodes support a few schedulers that vanilla ComfyUI doesn't expose directly - AYS, OSS, and GITS among them. This node packages the GITS scheduler as a SCHEDULER_FUNC object you can plug into those samplers. If you've seen "GITS" in a settings screenshot and wondered how to actually feed it into a sampler in ComfyUI, this is the adapter.

GITS is a step scheduler - it decides the sequence of noise levels (sigmas) the sampler walks through as it denoises. Like Karras or Exponential, it's not about which sampler you use but about how the steps are spaced, and people reach for these alternative schedulers chasing cleaner results at low step counts. It's a real, used option in the ecosystem, just a niche one.

How it works

Rather than being picked from a dropdown, GITS here is provided as a function that a compatible sampler calls to build its sigma schedule. You set the shape parameters on this node, it emits a SCHEDULER_FUNC, and the Impact KSampler consumes that to lay out its steps. The coeff controls the schedule's curve - how the sigmas are distributed across the steps - and denoise is the usual denoise strength (1.0 for a full generation, lower for a partial pass like img2img or a detail step). The narrow coeff range (0.8–1.5, stepping by 0.05) tells you this is a fine-tuning dial around a sensible default, not a wild knob; 1.2 is where it starts.

The inputs and output

• coeff (FLOAT, default 1.2, range 0.8–1.5) - the GITS coefficient shaping the sigma schedule. Small adjustments; the default is a reasonable starting point.
• denoise (FLOAT, default 1, range 0–1) - denoise strength, same meaning as on any sampler.

Output is a SCHEDULER_FUNC. Wire it into the scheduler-function input of an Impact KSampler (the pipe KSampler nodes that accept AYS/OSS/GITS). It won't plug into a stock ComfyUI KSampler - this is for the Impact samplers that expect a scheduler function.

How to install it

Ships with Impact Pack. ComfyUI Manager: search ComfyUI Impact Pack, Install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Impact-Pack

then pip install -r requirements.txt in ComfyUI's Python env (portable: python_embeded\python.exe -m pip ...), restart. Pack's by ltdrdata, the ComfyUI-Manager author. No extra models - schedulers are pure math.

Common issues & troubleshooting

The mismatch that catches people: this outputs a scheduler function, which only the Impact KSampler nodes accept - you can't route it into ComfyUI's built-in KSampler, and the socket simply won't connect. If you're converting a scheduler widget to an input and hitting compatibility errors, the pack also ships an Impact Scheduler Adapter specifically for that situation; that's the fix, not a bug in this node. As for results: don't expect GITS to be dramatically better than Karras - schedulers are a subtle lever, most visible at low step counts, and the honest advice is to A/B it against your usual scheduler rather than assume the exotic one wins. Start at the default coeff of 1.2 and nudge from there.