Example assignment: a producer sends a selected product shot to an editor. The background needs another pass; the approved lighting and framing should stay. The file shows the visual target. The handoff identifies the source version, the inputs worth keeping and the decision that makes this version the starting point.
You can use the shot log CSV for the first record. Fill it from a job you can inspect today. Leave unknown settings explicit and add detail when it changes what the next person can do.
1. Decide what the recipient needs to do
- Review the result. Send the selected version, the question to decide and any relevant comparison. A reviewer may need a short explanation of the change without access to the full generation project.
- Continue production. Include the retained prompt, model information, source assets and editable workflow or project where available. Name missing dependencies before the receiving artist opens it.
- Publish or register the work. Agree the required delivery specification, usage permissions, credits and any disclosure or registration information with the person responsible for that step. A final-file approval does not answer every one of those questions.
Record the recipient, their task and the delivery deadline. That gives you a practical boundary for the package. Internal rejected concepts, confidential references and full prompts should travel only when they belong in the agreed handoff.
2. Keep the source record your tools already provide
Inspect the original export before writing the same information again. ComfyUI can keep a workflow in generated media and export a separate JSON graph. Frame.io lets you export review comments in formats including CSV, plain text and JSON. These can be useful parts of the package; check what the particular export contains and which media version it describes.
Retain the original generated file separately from the graded, retouched or compressed delivery. Use a stable shot identifier across the source, chosen version and review record. If a tool exposes only a model name, keep that name. Do not invent a version number or recover a seed from memory. A later rerun is a new test, not evidence of an earlier setting.
3. Connect the technical record to the human decision
A prompt can describe the intended image. The selection note explains why one actual result became the starting point for the next person. Write the reason while choosing the version, with enough specificity to guide a revision: what stays, what changes and who can accept the change.
| Record | Keep from the source | Add or check with the team |
|---|---|---|
| Version | Original output, file identifier and the final delivery copy. | Which version was selected, its review link and any later changes. |
| Generation | Available prompt, model, settings and workflow export. | Missing inputs or dependencies and where an artist can access them. |
| Human work | The editable project or retained intermediate files. | Who composited, retouched, graded or edited which part. |
| Decision | The relevant review comment or written approval. | Why this version was chosen and what the next revision must preserve. |
| Delivery | The files and supporting records agreed for the job. | Recipient access, permitted use and any outstanding approval. |
Existing comments, metadata fields or workflow notes may already hold the reason. Keep a clear link to that decision instead of creating a conflicting second approval. Distinguish an internal creative selection from the client's acceptance and from permission to use source material.
4. Write a delivery note someone can act on
Use this delivery note for the background-revision example as a starting point. Replace the placeholders with your files, people and decisions.
- Use this version
- [Shot ID / selected version / file or review link]. Selected by [person] on [date]. Client approval: [link or still pending].
- Preserve this choice
- Keep the lighting direction and product framing from the selected shot. Replace the background only. Record any change that affects those constraints for review.
- Continue from here
- Source record: [prompt / workflow / project link]. Required inputs: [accessible files]. Missing information: [specific gap or none identified].
- Return for review
- Send [deliverable] to [reviewer] by [deadline]. Keep the new version distinct from the approved starting point.
5. Open the package as the next person
- Open the actual recipient link with their access level. Find the selected file and verify that the visible version matches the delivery note.
- Read the reason for the choice. Can you identify the permitted change without asking the original artist?
- Download the agreed files. If production continues elsewhere, check that the required source files and dependencies are accessible there.
- Check the information being shared. Confirm that internal notes, reference material and generation details appear only where intended.
- Record unresolved questions with an owner. Send a new version for changes after approval rather than silently replacing the reviewed result.
For a production restart, test the retained workflow in the receiving environment. Document whether it opens and runs, any substitutions and the result. Saved settings help an artist continue; they do not guarantee an identical render across models, versions or environments.
Keep delivery requirements attached to their actual scope
Check the current client contract and the destination's requirements for this job. For a US copyright registration application containing AI-generated material, the Copyright Office's guidance addresses disclosure and the human-authored contribution. That is a particular registration process, not a universal delivery checklist.
Where publication raises an AI disclosure question, use the companion guide to EU AI Act Article 50 to identify the relevant obligation and responsible role. Keeping production information supports that assessment. The record does not grant rights, apply a public disclosure or establish legal compliance.
Use BTW when the record needs to stay with the work
Behind The Workflow is built for AI creative teams who want the selected asset, its generation context and the reason behind a choice together. Bring one real shot, check the details retained by its supported capture or generation path, and add the human decision. Review the resulting share before delivering it.
A folder, workflow notes and a review tool can already do this well when the links are maintained. Test BTW against that actual setup: ask the receiving person to find the starting version, its source record and the next instruction in each. The useful outcome is a handoff they can use.