ComfyUI Node
PSN Latent Forensics
Inspects latent tensor metadata and sampled values for NaNs, Infs, and unexpected shape/dtype changes. Pure-logic implementation operating on plain Python values (no torch).
PSN Latent Forensics
- integrity_report
- is_valid
◄run_idrun_001►
◄actual_shape1,4,64,64►
◄expected_shape1,4,64,64►
◄actual_dtypefloat16►
◄expected_dtypefloat16►
◄sampled_values[]►
◄max_sampled1024►
CategoryParametric/Image
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| run_id | STRING | run_001 | Run id. Default run_001. |
| actual_shape | STRING | 1,4,64,64 | Actual shape. Default 1,4,64,64. |
| expected_shape | STRING | 1,4,64,64 | Expected shape. Default 1,4,64,64. |
| actual_dtype | COMBO | float16 | Actual dtype. One of 5 choices. Default float16. |
| expected_dtype | COMBO | float16 | Expected dtype. One of 5 choices. Default float16. |
| sampled_values | STRING | [] | Sampled values. Default []. |
| max_sampledopt | INT | 10241–100000 | Max sampled. From 1 to 100000. Default 1024. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| integrity_report | STRING | — |
| is_valid | BOOLEAN | — |