How to generate images with Nano Banana 2 in Figgshield
A practical walk through making images with Google's Nano Banana 2 from inside the agent you already use.
Cancel anytime before day 7 · no lock-in
What Nano Banana 2 looks like through Figgshield
Figgshield is an MCP server your AI agent talks to. Point the agent at https://api.figgshield.ai/mcp, ask for an image in plain language, and Nano Banana 2, Google's model, does the drawing. You never leave the tool you were already writing in.
Pricing is at cost. We do not add a markup on the model, so one image is 8 credits at 1K or 2K and 16 credits at 4K, the same price whether you are testing an idea or shipping the final. The only money Figgshield keeps is on credits that nobody spends.
The same credit balance also covers video, so there is nothing separate to buy when you decide to animate a shot later.
Connect it to the agent you already use
The connector works with Claude (desktop and Claude Code), Cursor, Windsurf, and ChatGPT in Developer Mode on a paid plan, plus any MCP-capable agent. There is no free tier, so the $5 seven-day Pro trial with 150 credits is the way in, and cancelling before day 7 means you pay nothing more. Full numbers sit on the pricing page.
Getting from zero to your first image takes four steps:
- Add the server URL
https://api.figgshield.ai/mcpin your agent's MCP settings. - Start the $5 Pro trial to load your 150 credits.
- Ask the agent, in plain words, for the image you want.
- Download the file from the private link it hands back.
Write a prompt Nano Banana 2 will follow
Nano Banana 2 rewards specifics. Name the subject, the framing, the light, and the medium, then say what you do not want. "A ceramic coffee mug on a pine table, soft window light from the left, shallow depth of field, product photo, no text" beats "nice mug photo" every time.
Keep one idea per generation while you hunt for the look, then lock the parts that work and change a single variable at a time. At 8 credits per 1K or 2K image, iterating stays cheap enough to be worth doing properly.
You can also send a reference image to steer colour or composition. That reference is used in memory for the single request and never stored, so it is gone the moment your image comes back.
A structure worth copying into your prompts:
- Subject: what is actually in the frame
- Composition: shot type, angle, distance
- Light: source, direction, mood
- Style: photo, illustration, or 3D render, plus any reference you can legally use
- Exclude: text, logos, extra fingers, whatever keeps creeping in
Choose 1K, 2K, or 4K on purpose
Resolution is the main lever on cost. 1K and 2K both cost 8 credits, so reach for 2K by default when you want the extra detail at no extra price. Save 1K for fast drafts and thumbnails, where the smaller file is genuinely handier.
4K costs 16 credits and earns its place on hero images, print, or anything a viewer will zoom into. For a social post that lives at phone size, 4K is usually money you did not need to spend.
A rough plan works well: draft at 2K, pick the winner, then regenerate that one keeper at 4K only if its final home calls for it.
Take a finished image into a short clip
When a still is working, you can animate it without starting over. Every generated image has an id, and passing that id into a video request gives you image to video, so the clip begins from the exact frame you already approved.
Video runs on Kling 3, with a faster Kling 3 Turbo tier for when you care more about speed than the last bit of quality. A five-second clip is 35 credits, drawn from the same balance as your images.
Generated files, image and video alike, sit on a private link for 7 days and are then deleted, so save anything you want to keep before the week is out.