Inspect¶
Inspects and displays any value flowing through a workflow. It formats text/JSON for readability, summarizes tensors, and attempts to render an image preview (including decoding base64-encoded images). Optionally writes a detailed summary to the system logs.
Usage¶
Use this node anywhere you need to debug or understand intermediate data. Connect any upstream node output to Inspect to view a human-readable summary, pretty-printed JSON (when applicable), and an image preview if the value is an image tensor or a base64-encoded image.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| value | True | WILDCARD | Any value to inspect (e.g., strings, numbers, dicts, lists, JSON strings, torch tensors, or objects). JSON strings are detected and pretty-printed; tensors are summarized; base64 image strings are decoded for preview when possible. | {"user": "alice", "score": 42} |
| write_logs | True | BOOLEAN | If true, writes a formatted summary of the value (with separators) to the logs. | true |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| value | STRING | The inspected value as a string. For JSON inputs, this will be pretty-printed. For other types, a readable string representation is provided. | { "user": "alice", "score": 42 } |
| image_preview | IMAGE | An image tensor preview derived from the input, when possible. If the input is an image tensor or a base64 image string, a corresponding preview is produced; otherwise, a default blank image is returned. | Tensor with shape [1, H, W, 3] and float32 values in [0, 1] |
Important Notes¶
- Output Node: This node is designed to display information directly in the UI and can act as a terminal inspection point in a workflow.
- JSON handling: If the input is a valid JSON string, it is parsed and pretty-printed; otherwise the raw string is shown.
- Tensor handling: Tensors are summarized in text and, when they represent images, are shown as an image preview.
- Base64 images: Strings starting with data URL prefixes (e.g., "data:image/..."), JPEG markers ("/9j/"), or PNG markers ("iVBOR") are treated as base64 images and decoded for preview.
- Fallback preview: If an image cannot be derived from the input, a small blank image is returned as a preview.
- Logging control: Set write_logs to false to avoid verbose logs when inspecting large or frequent values.
Troubleshooting¶
- Image preview shows a blank image: Ensure the input is either an image-like tensor (with valid dimensions and channels) or a valid base64-encoded image string (e.g., "data:image/png;base64,
"). - JSON did not pretty-print: Verify the input is valid JSON. If you pass a Python-like string (single quotes or non-JSON syntax), the node will display it as a raw string. Prefer passing a dict/list object or a valid JSON string.
- Large outputs slow down logging: Disable write_logs to prevent heavy log I/O when inspecting large tensors or long strings.
- Unexpected text in the node display: The UI shows a stringified version of the result. If the input is complex or includes non-serializable objects, provide a simpler representation (e.g., convert to JSON-serializable structures).
Example Pipelines¶
