AI QC Supervisor v2.1 - Akki

The AI QC Supervisor v2.1 is the least glamorous node in the AkkiNodes pack and one of the more honest ones. It takes the Cinematographer's raw shot breakdown and cleans it: removing junk asset entries like PROPS (None), fixing malformed lines, and rebuilding the report so the downstream parser doesn't choke. In other words, it exists to make the LLM look better than it is.

That's not a dig. Every LLM-based pipeline needs this, and the author's own bug log says so - it explicitly tracks that the QC Supervisor exists to clean up garbage keys the Cinematographer produces, with a plan to eventually harden the Cinematographer so this node becomes obsolete. For now, it's the glue that makes the pipeline reliable.

How it works

A "single-pass sanitizer" architecture: one LLM pass with a carefully worded prompt (refined across versions to avoid over-aggressive cleaning), followed by a deterministic Python rebuilder that reassembles the output. So the model identifies what's junk, and code guarantees the report is structurally valid again. Classic AkkiNodes: AI for judgment, Python for correctness.

Inputs and outputs

It's a thin node. Two required inputs:

• llm_model - same LLM_MODEL object as the rest of the pack.
• shot_breakdown_report - wire this from the Cinematographer's output of the same name.

Two outputs:

• clean_shot_breakdown_report - the sanitized report. This is what feeds the Pro Shot List Parser that produces your production CSV.
• full_llm_process_log - the raw pass, for when a clean result still looks wrong and you want to see what got stripped.

Where it sits in the pipeline

In the full flow: AI Cinematographer → AI QC Supervisor → Pro Shot List Parser → production CSV. If you're building the pipeline by hand and the parser is erroring on "None" props or malformed keys, adding this node in between usually fixes it without you touching the CSV manually. That's the whole point - the pack wants you to never hand-edit data between stages.

Install

Same as every node in this pack - it doesn't have its own installer, the pack does:

cd ComfyUI/custom_nodes
git clone https://github.com/routhakash/AkkiNodes-LLM-Suite-for-ComfyUI

Or install "AkkiNodes LLM Suite" via ComfyUI Manager, run install.bat as admin on Windows, and have a GGUF in ComfyUI/models/llms/. Restart and you're set.

One honest note: because this node is essentially a band-aid, you could skip it if you're confident in your GGUF and willing to clean the odd bad key yourself. But the whole pitch of this suite is "no manual data editing between stages" - and this node is how that stays true. It's a small price for a pipeline that doesn't fall over at step two.