Source Footage and Model Provenance Intake (Onboarding)
A checklist with 7 steps: Log every asset the client hands over before any generation begins.
By InnovaAI ResearchPublished
What are the steps?
Source Footage and Model Provenance Intake (Onboarding)
- 01
Log every asset the client hands over before any generation begins
Record file origin, capture date, and the person who supplied each clip or image. A retainer client who sends a folder of phone footage and a folder of licensed stock needs those two streams tracked separately from day one.
- 02
Confirm written usage rights for each uploaded face, voice, and location
Avatar and dubbing platforms reproduce likeness and vocal timbre, so a signed release per identifiable person is the baseline. Flag any asset where the client cannot name the rights holder.
- 03
Map which generator each asset type will route through
Assign avatar-led explainer work, generative B-roll, and template-driven batch output to distinct tools so the team does not default everything to one platform. Note the model version each tool currently runs, since output quality shifts between releases.
- 04
Set a resolution and aspect-ratio floor per delivery channel
A 16:9 training module and a 9:16 social cut need different source minimums. Capture the requirement now so a batch render does not fail after the client has approved a script.
- 05
Document the client's brand kit constraints in a shared spec
Colors, logo safe zones, lower-third fonts, and any banned visual treatments belong in one file the whole delivery team reads. Templated avatar output drifts toward generic without these guardrails.
- 06
Run one test render end to end and get client sign-off on the look
A 20 to 30 second sample surfaces lip-sync, pacing, and voice-clone issues before a full batch is committed. Store the approved sample as the reference for every later revision.
- 07
Record the intake decisions in the project brief the delivery team inherits
Rights status, routing map, channel specs, and the approved sample all travel with the job. Without this, a second editor rebuilds the same decisions from scratch mid-campaign.