ComfyUI Offline Batch

Read ../../references/COMFYUI_OFFLINE_BATCH.md and comfyui.

Before handing memory from a local LLM to ComfyUI, ensure the batch contains everything the runner needs:

• final API-format workflow JSON for every job
• stable source IDs and job IDs
• all final prompt/text values
• all seeds, sizes, durations, frame/audio parameters, model selections, and workflow inputs
• exact project input files plus deterministic upload/patch mappings
• job dependency edges
• output directories and expected output kinds
• timeout/retry policy that never changes creative inputs
• already-approved paid-route consent where applicable

Run live schema/model validation before arming the resource handoff. Missing nodes/models or unresolved placeholders are blockers while the LLM is still available.

The offline runner may retry the identical job for transient transport/server failures. It must stop and return control to the LLM for any semantic change, prompt rewrite, model substitution, missing creative decision, or continuity repair.