~$1.90 compute · 1.2h across 12d · 3 iterations · you pay 1000% of that

Faces stop drifting when the reference is clean, not when the prompt is longer. This is the turnaround that produces a usable reference plate, plus the censor bar step that gets it past a platform filter without changing the face.
Tested the boring way: A/B across the same thirteen-cut episode, with and without. Prompt text made no measurable difference. The reference did.
The real production cost behind this recipe, so the price has something to sit against.
Compute spent
~$2
$0.63 avg / iteration
Hours worked
1.2h
across 12 days
Iterations run
3
18 failed (600%)
You pay
$19
1000% of compute cost
Totals are summed from FC Studio's recorded cost ledger across every attempt on this workflow, successful or not. Estimates, not invoices: provider pricing shifts over time.
The full executable workflow — every step runs inside FC Studio, which hosts all the models it calls. Structure is public; each step's prompt and parameters are the paid layer.
FC Studio phase map — steps mirror the project structure (script, world, board, generation, edit), so the export imports aligned.
Templates and references bundled with the workflow, plus the workflow.json every connect target imports.
3 recorded. The creator picked 2 to show; buyers see the full log with every prompt.
Adjusted props
#1 · changed: props · 80d ago
Adjusted light
#2 · changed: light · 76d ago