Tutorials, comparisons and model selection for AI APIs.
Five identical prompts on both models, unretouched. Both spelled every word right. GPT Image 2.5 Sunburst rendered richer scenes and readable background signs; Nano Banana 2.1 kept edit inputs untouched, offers 8:1 and 1:4 formats, and costs less at 1K and 4K.
Six months, 293x, and I cannot write a line of the code that runs it. What actually mattered: the two things customers pay for, why I refused every open-source gateway, and the metric that caught a supplier substituting quality when every dashboard said 200 OK.
Named subjects, per-asset roles, stages with verifiable end states, and the inheritance clause that makes object swaps hold. Plus a status update: 2.5 is now callable through our API, from $0.120/s — measured, not assumed.
Sol, Terra and Luna are one model line at three price points (about 8 : 3.4 : 1 on input, 7 : 3.4 : 1 on output). Here is when each is worth it — and three upstream behaviours to budget for.
Native audio vs flat pricing vs real-person reference — pick by your priority.
Switch between Claude, GPT and Gemini by changing one "model" string — same key, same endpoint.
gpt-image-2 from $0.025/image, Gemini image from $0.025, and how the channels actually compare on cost.
