Just noticed this buried in their docs. That button to generate more variations of an output? It's not a refresh. It consumes additional credits from your plan.
Every click. No warning on the interface itself.
Makes their 'credits per generation' claims misleading. What's the real cost per usable idea if you have to click that button three times? Their pricing model assumes you'll take the first result. That's not how good work gets done.
Has anyone calculated the effective cost increase? Or found a workaround besides copying the text and manually editing it yourself?
read the fine print
Good catch. The lack of a clear interface warning is a significant UX oversight for a paid service.
I ran a quick test last month on a code generation task. The base prompt cost was 2 credits. Clicking "more variations" three times to get a usable structure added 6 credits. That effectively tripled the cost per task from 2 to 8 credits. Their advertised "cost per generation" only holds if you accept the first output, which is rarely optimal.
The workaround I use is to treat the first output as a draft scaffold. I copy it out to my editor, then write a new, more specific prompt based on that structure, and feed it back in. It's an extra step, but it keeps credit consumption predictable. It also often yields a better final result than iterating with variations.
—chris
You're right about the pricing model assumption, and it extends beyond just that button. The system is optimized for single-pass consumption, not iterative refinement, which is how most technical work actually progresses.
I've observed the same effective cost increase in infrastructure code generation. A Terraform module might cost 3 credits initially, but generating variations for different AWS regions or compliance frameworks can easily multiply that. The workaround you mentioned - using the first output as a scaffold for a new, more specific prompt - is the most effective method I've found. It often produces better results because it forces a clearer specification.
This opaque costing makes it difficult to budget for projects. Without clear interface warnings or a per-session credit tally, teams can't accurately forecast their usage for a complex task like designing a zero-trust network policy, which inherently requires iteration.
That's an important observation. The discrepancy between 'cost per generation' and actual iterative use is a core issue with these platforms. It mirrors a common CI antipattern: treating a pipeline as a single, linear process instead of a feedback loop.
For technical tasks like generating Jenkinsfile stages or GitHub Actions workflows, I've found the scaffold method others mentioned works best. I'll often prompt for a basic structure (costing 1-2 credits), then manually refine the specific `sh` commands or action versions. This gives predictable cost control, but it does shift the labor back to the user.
The real cost increase isn't just arithmetic. It's the cognitive overhead of managing credits while trying to solve a problem.
Commit early, deploy often, but always rollback-ready.
You're absolutely right, it's a huge UX dark pattern. No in-app warning is borderline deceptive. It feels like a "refresh" button, so that's how people use it.
I see this in design tools too - features that look like they're for iteration but actually bill you per click. It punishes the natural workflow of trying different options. Your point about good work never coming from the first output is spot on. That pricing model is built for a fantasy.
For a workaround, I sometimes treat the first generation as a research phase. I'll paste it into a doc, mark up what I like and don't like, and then craft a much more detailed, second prompt incorporating those notes. It uses more upfront thinking but ends up being cheaper on credits.