GPT Image 2

What it is good at

• The draft loop, which is the whole point

  A low-quality draft costs about $0.006 against roughly $0.21 for a final, so the workflow generates a draft, shows it to you, and only spends the real money once you have said yes. The approved draft's seed carries into the final, so upgrading the quality keeps the composition you actually picked instead of rolling a new image at ten times the price. Across a ten-slide carousel that is $0.06 to find the direction rather than $2.10.

• Telling you the cost before it spends it

  Every run estimates first. Below $0.50 it proceeds; at or above, it stops and asks. One flag prices a whole batch without generating anything, and another prints the fully assembled prompt without making a request at all. A tool that spends your money should never surprise you about how much.

• One dependency, and no vendor SDK

  PyYAML, for the catalogues. The API calls go out over the standard library. That is a deliberate trade: this is a skill that holds an API key, and the less third-party code sits between the key and the wire, the less there is for you to audit before you trust it.

• Presets that carry the whole prompt, not a style word

  21 of them, each pairing a short description you choose from with a full prompt fragment that does the work - editorial, blueprint, ink, risograph, wireframe, brutalist and more for visual work, and a separate set for infographics, slides, diagrams and posters where the text inside the image has to be legible. Platform sizing covers the eight places images actually go.