Inspect¶
Inspects any value flowing through a workflow, formats it for readability, and optionally writes a structured log entry. It also generates a best-effort image preview from tensors or embedded/base64 image data and shows the inspected text directly on the node UI.
Usage¶
Use this node to debug and understand intermediate data as it moves through a workflow. Connect any output to its input to view a readable summary, log it (if enabled), and, when possible, see a preview image. Commonly placed after model calls, data transformations, or image-producing steps to verify structure and content.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| value | True | WILDCARD | The value to inspect. Accepts any type, including strings, JSON-like structures, tensors, objects with an 'image' tensor attribute, or base64-encoded image strings (data URLs or raw base64 for JPEG/PNG). | {"result": {"label": "cat", "confidence": 0.92}} |
| write_logs | True | BOOLEAN | If true, writes a formatted inspection block to the server logs. | true |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| value | STRING | A formatted string representation of the inspected input. JSON is prettified when detected; non-JSON strings are returned as-is; other structures are serialized to JSON-like text. | { "result": { "label": "cat", "confidence": 0.92 } } |
| image_preview | IMAGE | An image tensor preview derived from the input when possible. If the input is an image-like tensor or contains base64 image data, a preview is produced; otherwise a blank image is returned. | Tensor of shape [1, 64, 64, 3] float32 in [0, 1] |
Important Notes¶
- The node displays the inspected text directly in its UI panel after execution.
- If the input is a tensor with image-like shape, it is used directly for the preview. Otherwise, the node attempts to decode base64 image data from strings or JSON fields.
- Unsupported or unrecognized image inputs yield a default blank 64×64 RGB image.
- The 'value' output is always a string (not the original object). If you need the original data downstream, branch the graph before this node.
- Write logs toggles server-side logging of a formatted inspection block for easier debugging.
- The node accepts objects with an 'image' attribute that is a tensor; that tensor will be used for the preview.
Troubleshooting¶
- Preview shows a black/blank square: ensure the input is an image-like tensor (shape [B,H,W,C], [H,W,C], or [C,H,W]) or a valid base64-encoded image (data URL or raw JPEG/PNG base64).
- Text shows as a raw string or '[object Object]'-like content: provide valid JSON strings or JSON-serializable structures for clearer formatting.
- No text appears in the node UI: confirm the node executed successfully and that upstream nodes produced data; also check that logging is not required for UI display.
- Unexpected tensor preview shape: ensure tensors are in image-compatible layouts with channel count 1, 3, or 4.
Example Pipelines¶
