Nano Banana Image Prompt and Generation

Convert an image idea into a clear Nano Banana prompt, explain the important choices, and optionally generate the approved image.

Loadout capabilities

Use <integration-tag>direct:fal_tools</integration-tag> for image generation. The primary operation is direct:fal_tools:fal_text_to_image, which currently uses Google Nano-banana 2 at 1K and saves the result to the asset library.

Inspect the live schema and price before execution. The current catalog description lists one generated image at $0.08, or 8 Aident credits; runtime billing and confirmation remain authoritative.

Prompt workflow

1. Identify the subject, intended use, audience, and visual hierarchy.
2. Infer ordinary details, but ask when brand colors, exact text, identity, or legal rights are material.
3. Write one primary conversational prompt. Add alternatives only when the user asks for optiIntegration · Falying on a long negative list.
4. Check that subject, setting, lighting, composition, style, palette, and quality expectations do not conflict.
5. Explain the three to five choices most likely to affect the result.

Use prompt-recipes.md for portraits, products, landscapes, and editorial graphics.

Prompt structure

Use natural sentences in this order:

Create [subject and action].
Place it in [setting and relevant context].
Compose the scene with [framing, angle, and focal hierarchy].
Use [lighting and palette] to create [mood].
Render it as [visual treatment] for [intended use].
Keep [critical constraints].

Prefer concrete visual evidence. For example, use "soft window light from camera-left with shallow depth of field" instead of "beautiful professional lighting."

Generation and iteration

If the user wants an image:

1. Show the final prompt before the first paid call when approval has not already been given.
2. Inspect and execute direct:fal_tools:fal_text_to_image separately through Loadout.
3. Return the generated asset ID or URL and the exact prompt used. Do not save the returned image a second time.
4. For revisions, change only the requested dimensions. Use direct:fal_tools:fal_edit_image when the live schema accepts the existing asset; otherwise generate a new image after explaining that it is a new paid call.

Quality and safety

• Do not promise exact typography, logos, celebrity likeness, or identity consistency.
• Confirm rights and consent for real people, trademarks, copyrighted characters, and private reference images when unclear.
• Refuse deceptive, exploitative, sexual, or privacy-invasive image requests prohibited by platform policy.
• If generation returns no image, report the result honestly and do not call it successful.

Attribution

This is a native Loadout re-authoring of the Nano Banana prompting approach documented in the pinned source in UPSTREAM.md. The preserved source license is in LICENSE.txt. It intentionally omits unsupported platform-specific parameters.