구글 Flow 무료 블로그 이미지

들어 있는 파일

SKILL.md
README.txt

SKILL.md 앞부분 미리보기

---
name: flow-nano-banana-images
description: Make blog images free with Google Flow and Nano Banana on a free plan: 0-credit safety check, prompt rules, download, quality check. Requires Aside. 구글 Flow 무료 나노바나나 블로그 이미지.
---

# Flow + Nano Banana Blog Images (free plan)

Use this to generate the images for a blog post in Google Flow with Nano Banana, **without spending credits**.

## Requirements
- **Aside** (browser automation tool) with a browser profile already logged in to the user's Google account. Never type or store a password; if Flow shows a login screen, ask the user to log in themselves.
- A Google account whose Flow plan shows **"0 credits will be used"** for image generation. Free plan limits can differ by account and can change; check on your own account first.

## Safety check before every batch (do not skip)
1. Open `https://flow.google.com`, then **New project** (a new project per post keeps images separate).
2. Click the model chip next to the prompt box. Choose **Image**, aspect ratio **1:1**, model **Nano Banana**, count **x1**.
3. The popup must say that generation uses **0 credits**. If it shows any credit cost, an upgrade prompt, or a paid plan, **stop** and tell the user. Never click upgrade or buy buttons.
4. Do not use the "agent" mode; use the plain image prompt box.

## Prompt rules (works for any topic)
Write each prompt in English with these six parts: camera/technique (for example "photorealistic photo, 35mm lens"), **count** of subjects, the action or state, the light, the composition, and the negatives.
- Always end with: `No people, no text, no logos, no watermark, no letters on any object.` Image models love to invent text on signs and dials; also add `blank unlabeled dial, only tick marks, no numbers` when a gauge appears.
- Say what is in frame and where ("centered", "upper right"). If the post card needs empty space for a title, ask for a calm dark lower half and no hard horizontal line.
- One idea per image. Real objects: describe their true shape. Historical scenes: lock the period.
- Do not ask for real people's faces or brand logos.
- Plan 7 images per post: 1 hero, 5 section images, 1 card background.

## Submit (what works)
1. Click the prompt box (a `textbox` div), type or paste the prompt.
2. **Click the send arrow with a real mouse click at its screen position.** A plain DOM `.click()` on the arrow button did not submit on a free account; `page.mouse.click(x, y)` did. Take a screenshot to find the arrow (bottom right of the prompt box).
3. Wait about 25 seconds per image. A tile shows a percentage while it renders.
4. Submit a few at a time (3-4), then wait for all tiles to finish before the next group. Newest tiles appear at the top of the list, so map results to prompts in reverse submit order, and **check each picture by eye**.
5. If a tile fails ("unusual activity", no charge), retry it; retried tiles can change order, so re-check the mapping.

## Download
- The finished tile's `<img>` has a signed `flow-content.google` URL (expires). Collect the URLs with the list scrolled to the top, then download each file with a script (for example Python `requests`) into one folder named `hero.png`, `s1.png` ... `card_bg.png`.

## Known limit: visible sparkle mark
On the free plan the downloaded images carry a small sparkle icon in the bottom-right corner (seen on every image in our test). Keep important content away from that corner, crop it out in the title card (a slight zoom does it), and tell the user that AI-made images must be marked as AI in the blog editor anyway.