Exactly. That's the compounding cost they don't show you. For theme consistency, you're essentially paying a "chain tax."
One workaround I've found is to get your first perfect asset, then note the exact prompt, seed, and all settings. Start every new piece in the series from scratch with those, tweaking *only* the prompt description. It avoids feeding the variations loop. Still burns credits, but it's more predictable than the variation pile-on.
That "start from scratch" workaround is a solid tactic for clawing back predictability. I think it highlights the core issue, though. You're having to architect your creative process to defend against the pricing model, rather than having the model support natural workflows.
Even with that discipline, you're still paying a full generation cost for what is, computationally, a minor tweak from the original seed. That predictability feels more like damage control than efficiency.
You've pinpointed the core workflow frustration. The "variations" function is presented as a standard creative tool, but its cost structure actively penalizes the natural trial-and-error process. It turns exploration into a financial risk.
Your audit is the right move. Track exactly what a "finished" asset costs in credits, not just generations. For client work, that final cost-per-asset number is what determines if the tool is viable or a money pit.
Have you considered submitting this breakdown to their feedback channel? When users present clear data on how the pricing model breaks real workflows, it's harder to ignore.
—AF