ComfyUI Extension: ComfyUI-SaveMem

Authored by hnmr293

Created

Updated

1 stars

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

Save Image/Latent to Shared Memory

Looking for a different extension?

Custom Nodes (2)

README

ComfyUI-SaveMem - Save Image/Latent to Shared Memory

日本語 ver.

A custom node for ComfyUI that writes images and latents to shared memory (multiprocessing.shared_memory.SharedMemory).

This is intended for use via API.

The nodes are added under hnmr/image and hnmr/latent.

Nodes

SaveImagesMemory

node image

Receives images (multiple images are also supported) and writes them as PNG format binary to the shared memory with the specified name.

| in/out | name | type | Description | | --- | --- |--- | --- | | in | name | STRING | Name of the shared memory to write to | | in [opt] | dummy_input | INT | Dummy argument to force redrawing |

Please create the destination shared memory on the receiver side in advance.

Memory format:

- 4 bytes: Number of images
- 4n bytes: Length of each image binary
- rest: Binary data of each image

Each values are stored in native endian.

SaveLatentsMemory

node image

Receives latent of any shape and writes it to the specified shared memory using torch.save.

| in/out | name | type | Description | | --- | --- |--- | --- | | in | name | STRING | Name of the shared memory to write to | | in [opt] | dummy_input | INT | Dummy argument to force redrawing |

Memory format:

- 8 bytes: Length of following binary
- rest: Binary data

Each values are stored in native endian.

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

Learn more