nano-banana-image-gen（纳米香蕉图像生成）

Pixwith Nano Banana 2 — AI Image Generation

Generate and edit images via the Pixwith API using the Nano Banana 2 model. Supports text-to-image and image-to-image with up to 4 reference images, multi-resolution output (1K / 2K / 4K), and flexible aspect ratios.

⚠️ CRITICAL — Do NOT Alter API Response Values

ALL values returned by the API (task_id, result_urls, image_url, upload_url, fields) are opaque tokens. Use them EXACTLY as returned — do NOT add, remove, or change even a single character. Store each value in a shell variable and reuse it directly. A single wrong character in task_id or result_urls will cause errors or broken links.

Setup

This skill requires a PIXWITH_API_KEY environment variable.

If the variable is not set, guide the user through these steps:

1. Go to https://pixwith.ai/api and sign up / log in.
2. Click "Add" to create a new API key and copy it.
3. Add the key to ~/.openclaw/openclaw.json:

{
  "skills": {
    "entries": {
      "nano-banana-image-gen": {
        "enabled": true,
        "env": { "PIXWITH_API_KEY": "key_your_key_here" }
      }
    }
  }
}

Verify by running:

curl -s -X POST https://api.pixwith.ai/api/task/get_credits \
  -H "Content-Type: application/json" \
  -H "Api-Key: $PIXWITH_API_KEY"

A successful response looks like {"code":1,"data":{"credits":500}}.

Pricing

Always inform the user of the cost before creating a task.

Model Parameters

• model_id: 0-41 (fixed)
• prompt (required): Describe the image to generate or the edit to apply.
• image_urls (optional): 1–4 publicly accessible image URLs for image-to-image mode.
• options.prompt_optimization (boolean, default true): Auto-translate prompt to English.
• options.resolution (required): 1K, 2K, or 4K.
• options.aspect_ratio (required): 0 (auto-match input image), 1:1, 16:9, 9:16, 3:4, 4:3, 3:2, 2:3, 5:4, 4:5, 21:9.

Workflow A — Text-to-Image

Use when the user provides only a text prompt and no images.

Step 1: Check credits

curl -s -X POST https://api.pixwith.ai/api/task/get_credits \
  -H "Content-Type: application/json" \
  -H "Api-Key: $PIXWITH_API_KEY"

Verify data.credits is sufficient for the chosen resolution.

Step 2: Create task

curl -s -X POST https://api.pixwith.ai/api/task/create \
  -H "Content-Type: application/json" \
  -H "Api-Key: $PIXWITH_API_KEY" \
  -d '{
    "prompt": "<user_prompt>",
    "model_id": "0-41",
    "options": {
      "prompt_optimization": true,
      "resolution": "1K",
      "aspect_ratio": "1:1"
    }
  }'

Response contains data.task_id and data.estimated_time (seconds).

Step 3: Poll for results

Wait for estimated_time seconds, then poll:

curl -s -X POST https://api.pixwith.ai/api/task/get \
  -H "Content-Type: application/json" \
  -H "Api-Key: $PIXWITH_API_KEY" \
  -d '{"task_id": "<task_id>"}'

• data.status == 1 → still processing, wait 5 seconds and poll again.
• data.status == 2 → done, data.result_urls contains the image URLs.
• data.status == 3 → failed, inform the user.

Present the EXACT result_urls to the user.

Workflow B — Image-to-Image

Use when the user provides one or more reference images plus a text prompt.

Step 1: Upload local images (if needed)

If the user provides a local file path (not a public URL), upload it first.

Upload constraints:

• Allowed formats: .jpg, .jpeg, .png only
• Maximum file size: 10 MB
• content_type must match the file: image/jpeg for .jpg/.jpeg, image/png for .png
• The presigned upload URL expires in 10 minutes

1a. Get a presigned upload URL:

curl -s -X POST https://api.pixwith.ai/api/task/pre_url \
  -H "Content-Type: application/json" \
  -H "Api-Key: $PIXWITH_API_KEY" \
  -d '{"image_name": "photo.jpg", "content_type": "image/jpeg"}'

Set content_type based on the file extension:

• .jpg / .jpeg → "image/jpeg"
• .png → "image/png"

Response:

• data.upload_url — presigned POST data (url + fields)
• data.image_url — the final CDN URL to use in image_urls

1b. Upload the file using the presigned data:

curl -s -X POST "<upload_url.url>" \
  -F "key=<upload_url.fields.key>" \
  -F "Content-Type=<upload_url.fields.Content-Type>" \
  -F "x-amz-credential=<upload_url.fields.x-amz-credential>" \
  -F "x-amz-algorithm=<upload_url.fields.x-amz-algorithm>" \
  -F "x-amz-date=<upload_url.fields.x-amz-date>" \
  -F "x-amz-signature=<upload_url.fields.x-amz-signature>" \
  -F "policy=<upload_url.fields.policy>" \
  -F "file=@/path/to/local/image.jpg"

Fill every field from the upload_url.fields object into the form. After upload, use data.image_url from step 1a as the image URL.

If the user already has a public image URL (starts with http), skip this step.

Step 2: Check credits

Same as Workflow A, Step 1.

Step 3: Create task with images

curl -s -X POST https://api.pixwith.ai/api/task/create \
  -H "Content-Type: application/json" \
  -H "Api-Key: $PIXWITH_API_KEY" \
  -d '{
    "prompt": "<edit_instruction>",
    "image_urls": ["<image_url_1>", "<image_url_2>"],
    "model_id": "0-41",
    "options": {
      "prompt_optimization": true,
      "resolution": "1K",
      "aspect_ratio": "0"
    }
  }'

When editing images, aspect_ratio: "0" auto-matches the input image dimensions.

Step 4: Poll for results

Same as Workflow A, Step 3.

Error Handling

All API responses follow {"code": 1, "message": "success", "data": {...}}. When code is 0, message contains the error. Common errors:

• Invalid API KEY — key is missing, wrong, or disabled.
• Credits not enough — user needs to purchase more credits at https://pixwith.ai/pricing.
• Invalid image format — only jpg, png, jpeg are supported.
• Invalid image url — the URL is not publicly accessible.

Defaults

When the user does not specify preferences, use these defaults:

• resolution: 1K
• aspect_ratio: 1:1 (text-to-image) or 0 (image-to-image)
• prompt_optimization: true