dsh-comfy-video-studio

The true power of ComfyUI lies in freedom: you can define near-arbitrary image-to-video pipelines via API JSON. But once a workflow is running, how do you turn it into a persistent, multi-server-schedulable, crash-recoverable page you keep coming back to? dsh-comfy-video-studio exists precisely for that. It ships without any built-in model templates. Instead, it imports your ComfyUI API JSON directly, auto-detects constant parameters, file uploads, enumeration options, and SaveVideo outputs, and generates a standalone dynamic workflow page.

The core capability spans the full pipeline from import to execution. At import time the plugin calls the /object_info endpoint of the target server to check for missing nodes, invalid enumerations, and port-type mismatches, producing a runtime compatibility report. After saving you can rename the page, reorder and regroup fields, and flag advanced parameters at any time. For multi-server setups each ComfyUI instance gets its own 1-to-3 active-slot configuration; overflow tasks enter a local FIFO queue while the plugin continuously monitors latency, VRAM, and remote queue depth. All video outputs are persisted, with one designatable as the primary preview; after a Harness restart, pending tasks and existing prompt_id monitors are automatically restored.

Built for developers who already use ComfyUI for video generation and want to distill experimental API workflows into long-lived tool pages inside Harness. It is especially useful when you run multiple ComfyUI servers with different hardware configs and need explicit control over concurrent submission and task recovery, where hand-rolled orchestration scripts would quickly become unwieldy.