ChatGPT Hoodie Continuity Audit: Build a Scene-Change Log

A ChatGPT hoodie continuity audit can turn two scene notes into a useful change log. Supply what was meant to stay fixed and what was intentionally different. The output should identify unrequested changes and unanswered questions without treating every lighting difference as proof that the product changed colour.

Use the indoor-to-outdoor continuity guide to plan the pair. This companion documents the handoff between versions. It is not an automated garment-authenticity test or evidence that a generated scene reproduces physical lighting exactly.

Prepare the before-and-after notes

Separate the producer's intended change from a reviewer's observation. “Only background changed” may be an instruction, not a verified result. Keep the original garment reference available for both scenes rather than using the latest concept as the sole source for the next version.

Copy the change-log prompt

Build a continuity change log for this hoodie scene pair. Use my source anchor, fixed-feature list, permitted changes and observations. Return: confirmed intended differences, possible unrequested garment changes, visibility limits and the next source check. Preserve candidate IDs. Do not infer colour accuracy from appearance alone, invent hidden construction or treat a creative scene as a real product test. Distinguish an instruction from a verified outcome. Do not approve publication; prepare a concise handoff for human review.

Ask for one source pointer beside each product-related entry. If the model cannot connect a statement to the supplied notes, it should mark that statement unsupported. A complete-looking log is less useful than a smaller log that accurately identifies its evidence.

A filled fictional example

Imagine the brief keeps a beige pullover and navy trousers fixed while moving from an indoor room to an outdoor setting. Your notes say the outdoor image appears lighter and may show an added front opening. The change log should separate the lighting question from the possible construction change.

The added opening needs a comparison with the real pullover reference. The lighter appearance alone does not establish a new product colour. This is a fictional example, not a measured lighting experiment or a claim that a particular tool produced those errors.

Use two narrow follow-ups

Keep the lighting observation unresolved. Rewrite the front-opening issue as a source-comparison action, citing the pullover construction note.

Show which entries describe planned changes and which are verified observations. Do not merge the two lists to make the audit sound complete.

Review the revised log with the actual images. If a reviewer confirms an alteration, update the evidence entry with that finding. If the detail is merely hidden, preserve that uncertainty. The model should not replace a cautious human correction with a stronger unsupported conclusion.

Check the final handoff

The pair needs stable source IDs, a readable list of deliberate scene changes and a small queue of unresolved product checks. Keep any new source file attached to its own version. An approval of one reviewed pair should not silently carry over when one panel is replaced later.

Keep the final change log with both reviewed candidate versions. If one scene is replaced later, repeat the source comparison for that pair instead of copying the earlier approval. This makes the record useful beyond the person who wrote the first brief.

TryBuy product composition and construction facts can be copied from the exact current listing, but they do not establish colour fidelity or weather performance in a scene. Keep the final decision with a human source comparison. For general instructions and follow-ups, consult official OpenAI guidance.