GCP Storage Upload Model

Once the download side of the GCP Cloud Storage Integration pack (koirpraw/ComfyUI-GCP-storage-decouple) is working, the obvious next question is: how do models get into the bucket in the first place? GCP Storage Upload Model is the in-graph answer. You give it a local file path, it uploads the file to your bucket, and returns the gs:// URL it landed at.

How it works: model_path is the local path to the file - the node verifies it exists and raises FileNotFoundError if you typo it - and bucket_path is where in the bucket it goes, default models/. An optional bucket_name overrides the env-configured bucket. The file is streamed up via blob.upload_from_filename, the same code path the pack's auto-sync uses, and you get the gcs_url (STRING) back: gs://bucket/models/whatever.safetensors.

When would you actually reach for this? Three cases. You trained a LoRA or finetune on the instance and want it in the bucket before the VM dies. You pulled a model down to iterate on it and want the corrected version synced back up. Or you're seeding a fresh bucket and this beats opening a terminal. One case it's not great for: moving a big existing model collection up. A single 6 GB checkpoint over a normal uplink is slow, and the upload isn't resumable - a dropped connection restarts it. For bulk transfer the README points you at gsutil -m cp -r ./models/* gs://your-bucket-name/models/, or the pack's selective_migrate.py script for hand-picking files. Use the node for the occasional file, gsutil for the migration.

Worth knowing for context: this pack also has a transparent mode - set COMFYUI_ENABLE_GCP_STORAGE=true and it patches ComfyUI's file handling so models and outputs sync to the bucket automatically, with a local cache. The upload/download nodes are the explicit, manual layer on top of that. This one matters because a bucket full of models is only useful if the download side can reach it - so the gcs_url you get back is exactly the string you'll paste into GCP Storage Download Model (or a loader) on the next instance you spin up.

Install is the pack standard: ComfyUI Manager → search "GCP Cloud Storage" (pack title "ComfyUI GCP Cloud Storage Integration"), or clone https://github.com/koirpraw/ComfyUI-GCP-storage-decouple into custom_nodes and restart. Then pip install google-cloud-storage (the pack's requirements.txt covers it) and configure GCP_PROJECT_ID, GCP_STORAGE_BUCKET, and credentials via GOOGLE_APPLICATION_CREDENTIALS or GCP_CREDENTIALS_JSON.

Troubleshooting: 403 Forbidden means the service account can't write - grant it object admin on the bucket. 404 means a wrong bucket name. DefaultCredentialsError means credentials are misconfigured. And if you get a runtime error saying the file wasn't found, check model_path - the node won't invent a path for you. This one touches your real cloud credentials, so keep the service account scoped to just this bucket, and don't run ComfyUI exposed to the internet without auth.