Hi everyone! 👋

As the author of comfyui-client, I've noticed a common challenge when building external applications: How do we efficiently pass internal ComfyUI data (like Latents, Conditioning, or Tensors) between different workflows or machines via the API?

To solve this, I built a new set of custom nodes: ComfyUI_Base64_IO.

Why pair this with comfyui-client?

Standard ComfyUI nodes generally output images or save files to disk. However, when building sophisticated API-driven apps, you often need to:

1. Distributed Inference: Run lightweight tasks (CLIP/VAE) on a CPU machine, serialize the result, and send it to a high-end GPU server for the heavy lifting (KSampler).
2. External Caching: Generate "Negative Prompt Conditioning" once, get the Base64 string via the client, store it in your DB, and inject it into future requests (skipping CLIP execution entirely).
3. Cross-Workflow Logic: Pass a Latent generated in Workflow A into Workflow B without saving intermediate files to disk.

How it works:

• AnyToBase64: Connect any input (Latent, Conditioning, Image), and it returns a Base64 string. The node is designed to return the string in the API response clearly.
• Base64To[Type]: Decode the string back into a usable ComfyUI object.

Example Usage:

You can now easily fetch intermediate data using comfyui-client and the AnyToBase64 node:

WIP

Check it out here:
👉 https://github.com/lenML/ComfyUI_Base64_IO

Feedback is welcome! Let me know if this helps simplify your API workflows.