Why track each step separately instead of one flat cost?

Real pipelines rarely produce a usable image in one shot. You might generate four candidates, upscale the winner, inpaint a flaw, then remove the background. Each stage has its own price, so a per-step model gives a far more honest cost than a single generation figure.

How do I model generating four images to keep one?

Set the generation step's "runs per image" to 4. The tool multiplies the per-run cost by 4 for every finished image, capturing the real cost of selection and rejection.

Does this include compute or only API fees?

It tracks whatever per-run cost you enter. For hosted APIs use the published price; for self-hosted GPUs, divide your hourly GPU rate by how many runs you complete per hour and enter that.

Is my data saved anywhere?

No. All calculation happens in your browser and nothing is uploaded or stored. Refreshing the page clears everything.

AI Image Workflow Total Cost Tracker

AI image workflow cost tracker

A single “cost per image” number almost always understates reality. A finished, client-ready image usually passes through several billed stages — you generate a batch, pick a winner, upscale it, fix a detail with inpainting, then strip the background. This tool models that whole pipeline so you can see what one finished image actually costs and what your monthly bill will be.

How it works

Each row is one workflow step with two numbers: cost per run and runs per image. The runs-per-image field is what makes the estimate honest — if you generate four candidates to keep one, generation runs four times per finished image. The tool multiplies cost by runs for every step, sums them into a per-image cost, then multiplies by your monthly volume for the total. You can add or remove steps freely to match your exact pipeline.

Tips and examples

Count rejections. The biggest hidden cost in image work is the images you throw away. If your hit rate is one in five, set generation to 5 runs per image.
Upscaling is often the priciest step. High-resolution upscalers can cost more per run than the original generation — keep it as its own line so it stays visible.
Compare pipelines. Duplicate the workflow with cheaper steps (fewer candidates, a lighter upscaler) to see the monthly saving before you change anything.
Self-hosted? Convert GPU time to a per-run cost: hourly rate ÷ runs per hour, then enter that as the step cost.