Find out whether time is being spent waiting for a worker, running the ComfyUI graph, transferring media, or updating the Studio screen.
Before you tune #
Choose one known-good Pipe, fixed inputs, batch size one, and a stable project destination. Record the current engine, model, dimensions, and Studio version.
Readiness closes Generate when the engine, graph, terminal output, dependencies, required fields, or project routing are incomplete. Performance begins after that contract is valid: queue wait belongs to scheduling, node time belongs to ComfyUI execution, delivery belongs to media transfer and storage, and interface responsiveness belongs to the Studio screen being used.
Locate the time #
- Confirm the readiness check is clear.
- Note when the job becomes queued.
- Note when ComfyUI begins execution.
- Identify the node or stage that occupies the runtime.
- Note when media reaches Gallery.
- Compare interface responsiveness separately from job time.
Use the timing #
| Observation | Cause to inspect |
|---|---|
| Generate remains closed | Readiness message |
| Long queue before execution | Engine scheduling or competing jobs |
| One slow graph node | Workflow, model, dimensions, frames, provider GPU |
| Engine finishes; media arrives late | Transfer, thumbnailing, or project storage |
| Typing or grid interaction is slow | The visible Studio screen and its data volume |
The generation, Gallery, Pipes, and Transform examples in these Docs show those screens independently. They are diagnostic scenarios, not product speed promises.
Tune one variable #
Change one workload or exposed Advanced value, repeat the same run, and compare the same interval. A successful comparison names the interval that changed.