Your China-based production partner has started processing a collection of product assets when the campaign offer changes. Someone requests cancellation. More files arrive afterward. Should the team accept them, discard them, or commission the whole collection again?
ONCE recommends three separate milestones: the stop request, the end of processing, and reconciliation against the revised brief. A cancellation can limit execution without deciding whether an individual output remains useful.
What the documented behavior supports
Adobe’s Workflow API feature guide says cancellation prevents new assets from starting, while assets already in progress may complete. It also provides per-asset execution results. This applies to the Firefly Creative Production Workflow API.
Anthropic’s Cancel a Message Batch reference describes a canceling state in which non-interruptible requests may finish. Individual results establish which requests were canceled. That is a message-processing API rule, not a universal video-generation guarantee.
Both sources were checked on November 2, 2026. Their precise publication dates were not established. They support a limited conclusion: the batch label alone cannot describe every item’s final outcome. They do not establish identical charging or refund policies across services.
ONCE’s stop-and-reconcile record
This proposed production method separates execution state from the decision to use an asset. It is not a platform feature or a claim about ONCE client results.
| Record | What to retain | Question answered |
|---|---|---|
| Stop request | Original batch and brief revision, request time, response and final processing state | What was stopped, and what still needs tracking? |
| Item results | Task identifier, input revision, actual output and final execution outcome | What is available to inspect? |
| Revised-brief decision | New brief, affected elements, reviewer and retain-or-rework decision | Can the output belong in the revised delivery? |
Two successfully returned assets can need different treatment. A clean product close-up may remain suitable; a promotional image carrying the old offer may need work. Leave unresolved items in a pending-investigation list. File arrival alone is insufficient permission to deliver or publish.
Hold affected delivery while cancellation settles
Once the change is confirmed, remove affected assets from the release list and notify the production and campaign owners. Send the cancellation request, then reconcile final results using the provider’s supported process. This is an operating recommendation; it does not assume every service supplies the same pause or rollback controls.
Keep late-arriving results attached to their original task and brief revision. A new filename does not make an old-brief output compliant with the new assignment. Link replacement work to the original task and avoid submitting the full collection again while the first run is still unresolved.
A hypothetical 100-item change order
Suppose a brand changes an offer while 100 asset tasks are being processed. After final reconciliation, assume 42 have outputs and 58 did not finish execution. Review finds that 24 outputs satisfy the revised assignment and 18 require regeneration.
This is an illustration, not a client case, test or promised platform outcome. It assumes all 100 deliverables are still required, the retained 24 passed the relevant checks, the other 18 cannot be repaired through a simpler edit, and there are no unresolved or additional failed items.
| Disposition | Illustrative count | Action |
|---|---|---|
| Retain after review against the new brief | 24 | Carry into the delivery list with original provenance |
| Completed but requiring regeneration | 18 | Regenerate under the revised brief |
| No completed result | 58 | Produce under the revised brief |
The next generation scope is 76 items, rather than an automatic rerun of 100. That count is conditional. An overlooked old price could invalidate a retention decision; a reduced delivery scope could reduce the remaining work. Reconcile the requirement as well as the files.
What buyers should put in the change procedure
Ask what “stop immediately” can mean in the actual service: stop further submissions, remove queued tasks, or interrupt active work. Where the answer is unknown, verify it before promising it. Record cancellation response, reconciliation and revised delivery timing separately.
Keep platform charges, human work already performed and work added by the revised brief distinct. A cancellation status cannot establish that no cost occurred or that the client owes a particular amount. Fixed-scope services and pass-through usage arrangements need different treatment under the agreed scope and change procedure. No universal refund percentage follows from these documents.
A practical planning formula is: additional change effort = result reconciliation and review + necessary preparation of retained assets + repair or regeneration of affected items + production of missing items + any actual rescheduling work. Count each component once and identify work already included in the service. This is a scope model, not a price benchmark.
For a brand working with a Chinese manufacturer and an overseas campaign team, give product changes a named owner. Confirm the new offer and product facts before replacement work starts. Your production partner needs one revised assignment, rather than conflicting instructions attached to different files.
For ongoing content services, the useful operational asset is a record linking changed requirements, returned outputs and the revised delivery. Such records could support better estimates over time; this article does not claim measured margin gains. Explore ONCE’s AIGC articles for related creative-production decisions.
Sources and limits
The Adobe and Anthropic references linked above were checked on November 2, 2026. The reconciliation method, effort formula and 100-item example are ONCE’s recommendations and hypothetical analysis. Check the actual provider and service agreement for cancellation and charging behavior; these examples do not establish a rule for every AI tool.
