Nano Banana 2 — 完全ガイド

Nano Banana 2 is Google’s Gemini-class image stack for fast text-to-image, image-to-image, and conversational edits. Public guides (including Google Cloud’s Nano Banana prompting overview and community write-ups on Gemini 3.1 Flash Image) emphasize structured language, camera vocabulary, and iterative follow-ups rather than tag soup.

1. Write prompts like a shot brief, not a keyword bag

Use short, grammatical sentences that cover subject, action, environment, lens or camera angle, lighting, materials, and aspect intent. A practical skeleton is: who/what + where + time of day + lens feel (e.g. 50mm, shallow depth) + light quality (softbox, golden hour) + color palette + any on-image text verbatim.

Prefer one clear hero subject before adding busy background detail.
Name the camera distance: wide establishing, medium portrait, macro product.
State negative space or layout needs (“top third empty for headline”) when designing posters.

2. Reference images: identity, product, and style locks

Gemini-family workflows commonly combine one or more references with text. Use references to lock a face, garment, packaging silhouette, or palette; then describe the change you want (“same character, side light, rainy street, keep outfit”). If colors drift between generations, restate the palette hex or plain-language color names in each follow-up.

3. Aspect ratio, composition, and readability

Pick the ratio for the destination (1:1 feeds, 9:16 stories, 16:9 hero, 21:9 banner). When editing from an existing frame, say whether to preserve composition or re-crop. For UI mockups or packaging, explicitly request legible typography and safe margins—many guides note that spelling out text content beats vague “add a title.”

4. In-image text and logos

Quote the exact string you need, specify placement (“upper-left corner, small caps”), and describe contrast (“dark text on warm gray”). Avoid tiny micro-copy if you need readability; ask for “large, high-contrast headline” instead.

5. Iterate in small steps

After a first pass, refine with surgical edits: “softer fill light only,” “widen crop 10%,” “swap background to brushed concrete, keep subject pose.” Keeping prior constraints in each message reduces drift compared with rewriting the entire prompt from scratch.

6. Draft → polish workflow

Many practitioners start at a moderate resolution or simpler lighting to validate composition, then add complexity or upscale once the layout is locked. On Pixocto, watch credit hints per model and resolution tier before batching variants.

7. Safety, rights, and brand

Do not request real-person likenesses without rights, counterfeit packaging, or hateful content. Respect Pixocto Terms of Service and upstream provider policies; treat outputs as creative assets that may still need human QC for client work.

8. Using this on Pixocto

Open the image workbench, choose the Nano Banana 2 route, attach references if needed, set aspect ratio and resolution, then generate. Use My Creations to compare versions; when you are satisfied, export or continue with short edit prompts.