Workflow
The Overflow Rule: Deciding What Renders Local vs Cloud
ยท RenderBob team
Without a clear rule, "should this run local or cloud?" becomes an artist's judgement call at 9pm on a deadline, and judgement calls under pressure are inconsistent and expensive.

The failure mode of a hybrid pipeline is usually a missing decision. Without a clear rule, "should this run local or cloud?" becomes an artist's judgement call at 9pm on a deadline, and judgement calls under pressure are inconsistent and expensive. Write an overflow rule: a policy the pipeline enforces so nobody has to decide in the moment.
A good overflow rule is built from a few explicit triggers.
VRAM ceiling
If a job's peak memory exceeds what a local node can hold, after quantization and memory flags, it goes to cloud. This is the "it OOMs locally" case, decided in advance rather than discovered mid-render.
Queue pressure during a delivery window
If owned nodes are full during a nominated delivery window, additional jobs burst to cloud so iteration never stalls when it matters most. Outside delivery windows, the same jobs can wait for local capacity to save money.
Resolution and length thresholds
Set explicit thresholds. Above X resolution or Y frames, route to cloud, so the heaviest final passes go straight to capacity that can hold them, while previews and iteration stay local.
Security policy
Some clients' work never leaves owned infrastructure regardless of the above. The rule encodes that too: for tagged NDA jobs, cloud is not an eligible destination.
Two guardrails keep the rule honest. First, a spend ceiling on overflow: a hard cap, idle shutdown, and budget alarms, so bursting can never produce a surprise invoice. The studio owns its cloud account and sees the bill directly. Second, an override with an owner: someone named who can send a job to cloud outside the rule, and a record of when they did, so exceptions are visible rather than silent.
Write the rule down and let the pipeline enforce it. Otherwise bursting stays a 9pm judgement call, and you have two farms instead of one elastic one.
More from the blog
- Two Lanes of AI Editing: Mechanical Cleanup and Narrative Assembly
AI editing tools split into cleanup that follows story decisions and assembly that proposes them. A third question, local or cloud processing, now cuts across both.
- "Just Regenerate": The Bad-Seed Workaround Culture in AI Video Editing
While reference-guided video stays unreliable, working practice is blunt: do not commit a clip until you have looked, delete a stuck result, and switch models when one keeps missing.