远程执行分布式计算序列化(safetensors)ComfyUI扩展
在远程 ComfyUI 实例上执行图中节点，支持按输入、从某处开始、通过 JSON 或懒切换模式远程运行并选择序列化方式。
💡 将计算密集型子图发送到远程实例执行，减轻本地负载。
🍴 2 Forks💻 Python🔄 2025-12-01
https://pan.quark.cn/s/e98a62d17551
📦 requirements.txt
websockets>=12.0.0,<15.0.0
requests
📄 README
Nodes to run nodes on remote ComfyUI instances.
Included Nodes:
Remote Run Input Graph: Runs everything going to its inputs on the specified remote instance
Remote Run Start From Here ->: Used together with above node, makes it only run its input nodes up to hitting a start node
Remote Run JSON: Runs the JSON on the specified remote instance
Remote Run Lazy Toggle: Simple toggle switch that runs parts of the graph either only locally or only remotely
Serialization:
Safetensors: default, uses safetensors, fine for serializing latents and conditionings but may not support more complex types
Torch Unsafe: normal torch pt, should have widest compatibility, probably best if you have full control or trust of the remote instance.
Torch Safe: minimal safe implementation of torch pt, safe but only supports a small subset of types
Offloaded text encoder and VAE to remote instances
Same but with lora and toggles to avoid loading unnecessary models: