Jakkanna SCAIL-2 Validate Sampler

The worst kind of bug in a SCAIL-2 Production run is the one that finishes. You queue it, wait the minutes, save the video, write the manifest - and only then notice you ran it with the wrong CFG or a LoRA strength that wasn't what the manifest claims. Jakkanna SCAIL-2 Validate Sampler exists to make that failure impossible: it reads your actual executed graph, cross-checks the sampler, scheduler, CFG, mode, and LoRA strengths against what the production manifest declares, and refuses to proceed if anything mismatches.

The author's own description is the cleanest summary: wire it between the scheduler and the sampler so a mismatch stops the run before sampling rather than after the manifest is saved. That's the whole design.

How it works

One required input:

• sigmas (SIGMAS) - the schedule coming out of your scheduler node (in the production workflow, from Jakkanna SCAIL-2 Flow-UniPC).

It then inspects the executed prompt graph. Because it needs the full graph, ComfyUI only provides it when the workflow is actually queued - run it in preview/modal mode and you'll get a clear error telling you the prompt graph isn't available yet.

Outputs:

• sigmas - the schedule passed through unchanged. If it validates, nothing about the sampling changes; this node is a tripwire, not a transformer.
• sampler_report (STRING) - the JSON verdict: which sampler/scheduler/CFG it found, and confirmation that it validated before sampling began.

The gotcha that matters

Because the validation needs the executed graph, this node can't do its job if you've only connected sigmas in the editor. That's not a bug - the check literally has to see the KSampler/SamplerCustomAdvanced settings and the LoRA nodes connected elsewhere in the graph. If you get the "needs the executed prompt graph" error, you're testing it the wrong way; queue the workflow and it resolves.

Install

One node in the teenu/ComfyUI-Jakkanna pack. ComfyUI Manager: search Jakkanna → Install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/teenu/ComfyUI-Jakkanna.git
cd ComfyUI-Jakkanna
pip install -r requirements.txt

Restart, and don't install the upstream vnccs-utils pack in the same instance - same node IDs, guaranteed conflict.

Should you use it?

If you're iterating on a personal workflow, honestly, skip it - it's ceremony when nothing is at stake. It earns its place when a workflow will be shared or re-run where the settings can silently drift, which is exactly the "deterministic output when sharing graphs with other users" goal the pack is built around. When a shared Production workflow loads and runs to the sampler only to stop with a mismatch, that's the system working: someone changed a setting and the manifest caught it before the compute was spent.