Nano Banana Pro Image Generation and Editing

Problem

When wrapping the AI Hive image API, engineers often hit the same pain points: a wrong model ID, reference uploads that are not clearly tied to the prompt, task polling that can time out, no price snapshot, and downloaded files scattered across local folders. This skill pins the entry point to public_model_nano_banana_pro and turns query, upload, submit, poll, and download into a repeatable CLI flow, reducing state drift in hand-written scripts.

How It Works

• Generate and edit: use generate --prompt for image descriptions or editing instructions; add one or more reference images with --image, and specify whether each image supplies subject, layout, texture, or style.
• Parameters and routing: pass real-time imageConfig values with --param key=value, set count with --batch, and choose COST_FIRST, SPEED_FIRST, or SUCCESS_FIRST with --routing; results default to ~/Downloads/AiHive.
• Task management: after submission, keep the taskId and query status with task --task-id; upload --file handles image uploads. The skill does not silently swap models, avoiding accidental substitution with another model.

Boundaries

It requires a valid AI Hive API Key and a Python 3 runtime with requests. Parameter enums, image sizes, and reference-image counts should be taken from the live model configuration. Text that must appear exactly still needs manual review before delivery. Batch generation multiplies cost, so confirm live pricing first; if a model is retired, renamed, or reconfigured, query the live list instead of hard-coding stale values.