ComfyUI Node
CV Array → Latent
Wraps a raw float ndarray as a ComfyUI LATENT ({'samples': [1, C, H, W]}) so a cv2-produced array can re-enter a latent pipeline - e.g. feed the plate from 'CV Temporal Reduce' (run on a LATENT batch) into 'VAE Decode'. The array's LAST axis becomes the latent channel axis; values are kept as float32 and NEVER quantized (latents are unbounded). This is the inverse of feeding a LATENT into any OpenCV node, which unwraps frame 0 to [H,W,C]. The channel count must match what the downstream VAE expects (16 for qwen_image_vae, 4 for SD 1.5 / SDXL).
CV Array → Latent
- nparray
- latent
Categoryimage/CV/low-level
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| nparray | NPARRAY | Float ndarray [H,W] or [H,W,C] (C = latent channels; a 2-D array becomes single-channel). Values are taken as-is - no 0..1 or 0..255 scaling. Use 'Inspect CV Data' to check the shape/dtype. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |