Profile Inspector
Which Krea 2 checkpoint did you actually load?
- model
- profile_report
Krea 2 ships as RAW and Turbo, they're different models with different schedules, and the filenames in your models/checkpoints folder are frequently not a reliable tell. The Profile Inspector is Sigmax's way of asking the loaded model directly: it inspects a supported Krea 2 MODEL and reports which profile it resolves to - without serializing or mutating it. You wire in a model you already have loaded and get a profile_report string back telling you what Sigmax's validation logic thinks you're running.
It sits next to the CheckpointEvidenceInspector in the pack's inspection family, but they answer different questions. The evidence inspector reads the header of a .safetensors file on disk. This one inspects an already-loaded MODEL object - the actual in-memory sampling configuration - and checks it against the pack's supported Krea 2 profiles. The evidence inspector is the cheap pre-flight; this is the "okay, it's loaded, what is it?" moment.
Inputs and output
- model - a
MODELinput from your Krea 2 loader. - variant -
TurboorRAW, your declared intent. The inspector validates the loaded model against it rather than trusting it blindly. - steps, width, height, strict_official - these mirror the scheduler's parameters because the profile check is resolution- and recipe-aware. If the loaded model's profile doesn't match what the schedule you're about to build expects, the report is where that mismatch surfaces.
The output is a single string, profile_report. It's a report node - you read it, you don't wire it downstream.
Why you'd actually bother
Two workflows benefit. First, the diagnostic one: you loaded a checkpoint, you're not sure whether it's RAW or Turbo (or someone renamed a file), and you want the model's own evidence before committing to a schedule. Second, the reproducibility one: this pack is obsessive about fingerprints and traceable schedules, and a profile report is part of that paper trail - proof of what a workflow was actually running when it produced a given output.
Installing
One pack, all nodes:
cd ComfyUI/custom_nodes
git clone https://github.com/rookiestar28/ComfyUI-Sigmax comfyui-sigmax
Restart ComfyUI, search Sigmax. ComfyUI Manager lists "ComfyUI-Sigmax". Python 3.10+, ComfyUI 0.29.0+, no extra dependencies.
Where it bites
Scope, mostly. The description says "a supported Krea 2 MODEL" - that's the promise. It's not a general-purpose model inspector; feed it a Flux checkpoint and don't expect enlightenment. And remember it's advisory in the same way the whole pack is advisory: the report tells you what the model's profile resolves to, which is a strong signal, not an oracle. If the report and your workflow disagree, trust the report over the filename.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| variant | COMBO | 2 options: Turbo, RAW | |
| steps | INT | 81–10000 | — |
| width | INT | 102416–65536 | — |
| height | INT | 102416–65536 | — |
| strict_official | BOOLEAN | true | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| profile_report | STRING | — |