A China-based production team sends convincing character designs. Your brand approves them. The assembled film still feels disconnected: one person reacts before the other has done anything, or a close-up suddenly implies intimacy that the preceding shot never established. Better facial detail may leave that problem untouched.
For AI brand films, ONCE recommends separate approvals for character appearance and character interaction. One establishes who is in the scene. The other tests what causes that person to change. Buy a continuous interaction test before committing the full scene to production.
A current campaign prompts a practical question
Deutsche Telekom's 2 November 2026 campaign announcement describes a Christmas story about a grandfather and granddaughter reconnecting. It credits director Léo Cannone and New Forest with generative-AI production of the visual world and campaign imagery. ONCE checked the announcement on 5 November.
The disclosure does not establish production savings, revision effort or advertising returns. It does raise a useful buying question: when human relationships carry the idea, what should a client approve beyond the character sheet? The process below is ONCE's proposal, not a reconstruction of this production. It is not a frame-by-frame review of the film.
Identity continuity does not establish a relationship
A character can retain the same face and wardrobe while behaving as though the other person is absent. Two individually plausible expressions may still make an implausible exchange. The test is whether information passes between the characters in an understandable order.
Replace an instruction such as “look warmly reassured” with a playable event: the person notices their partner set aside another task, then chooses to move closer. The expression can vary. The reason for changing must remain legible.
The ONCE three-part interaction test
| Part | Define before production | Review in motion |
|---|---|---|
| Stimulus | What A does, and B's state beforehand | Does the initiating event actually appear or sound? |
| Perception | How B can notice it, including position and gaze | When does B receive the new information? |
| Response | What B changes because of that information | Does the response connect back to A? |
These are three dramatic functions, not a rule requiring exactly three shots. Perception can happen within an unbroken shot. Review enough material before and after the action to understand the change, in its intended order. Three selected stills cannot establish timing. If an offscreen sound supplies the stimulus, include it in the test.
A hypothetical product story
Imagine an outdoor brand working with its Chinese manufacturer and production partner on a camping film. One companion is still packing alone. Another clears a place beside the fire and offers a seat. The first person notices, puts the remaining bag down and joins them.
The interaction depends on an invitation being available to perceive before the response. Generating a smile for one character and a welcoming gesture for the other is insufficient if the edit reverses that order. Remove the invitation temporarily and ask whether the same response still plays for an unrelated reason. This is a writing exercise, not an ONCE client result or a claim about product performance.
Review once for comprehension, then with the intended sound
First, remove mood-setting music while keeping dialogue and environmental sounds essential to the event. Ask a target-language colleague who did not make the scene to describe who noticed what and what happened next. This can reveal a misunderstanding; a few internal opinions are not representative audience research or a forecast of effectiveness.
Then restore the proposed soundtrack. Is it supporting a readable exchange, or merely announcing the emotion that the action has failed to establish? If the relationship remains unclear, revise staging, shot choice or the cut before asking finishing work to rescue it.
Check the sequence directly: does a response precede its trigger? Does an eyeline unexpectedly refer to somebody else? Does the end of one movement permit the next shot to follow? None of this requires a universal duration for a human reaction. These questions identify the intended correction without promising that a particular model can execute it.
Let the test determine the next production step
If the exchange reads correctly but identity drifts in one shot, repair that shot and review the sequence again. If B never has a chance to notice A, revise the staging or storyboard. More samples do not supply a missing story event. If precise performance remains unattainable with the proposed route, evaluate filmed performance, controlled animation or a hybrid approach.
Label an attractive but unresolved sequence accurately: appearance approved, performance unresolved. It should not enter a library as an approved scene. Otherwise a later team may repeat a mistake that the brand never accepted.
Specify interaction development as work
Distinguish character development, the interaction test, final shot production and version preparation in the scope. Identify the exchange being tested, the continuous material to be supplied, the reviewer and the changes that affect neighbouring shots. Price actual work rather than borrowing an unsupported industry percentage.
Changing an expression within an agreed intention differs from changing why a character responds. The latter may alter preceding action, edit duration and sound. Equally, a supplier should not treat an interaction missing from its original interpretation as an automatic client scope addition.
For ONCE's longer-term workflow assets, retain the approved intention, continuous test, rejected alternatives and their reasons. This can make future reviews more specific. Any reduction in rework must be established from real project records. Teams preparing a brief can explore ONCE's production solutions and specify the interaction they need proved, alongside product and campaign requirements.
Sources and limits
Primary source: Deutsche Telekom, “Deutsche Telekom Christmas campaign: Take time to connect,” 2 November 2026, checked 5 November. The project was also reviewed through Campaign Brief Asia's 4 November production report; factual production claims here follow the brand's disclosure. The three-part test, two reviews and camping example are ONCE's proposed methods, not validated performance benchmarks. The original cover is conceptual.
