Hand GPT Image 2.5 Flare stills to a video timeline with an explicit asset contract

A storyboard frame is not a finished video shot. For GPT Image 2.5 Flare, that distinction should be explicit in the handoff: use the model for the still-image asset, and give the editor responsibility for how that asset occupies time.

OpenAI’s Flare model documentation says it accepts text and image inputs and produces image outputs. Its modality table marks audio and video as unsupported. This is the documented Flare variant’s boundary, not a claim about every GPT Image 2.5 variant or a report of a generation test. The source was captured on September 10, 2026; that is an observation date, not a release date.

The practical decision: if the deliverable requires a moving shot, don’t mark the shot complete when the Flare image is approved. Mark only the still asset approved, with the timeline work still outstanding.

Give the editor an asset contract, not an implied clip

Here is a proposed handoff record for a hypothetical storyboard still. These are production notes owned by the application or editor—not fields returned by Flare and not image-generation API parameters.

Field Example editorial entry Why it belongs in the handoff
Asset / shot frame-07-v2 / shot-07 Identify the exact still and intended timeline position.
Media kind Still image from GPT Image 2.5 Flare Keep image approval separate from video approval.
Display duration 3 seconds, editor assigned Specify how long the image should be held; the duration is not a model output.
Crop / framing 16:9 timeline crop; preserve the face and sign Give the editor a framing requirement to check against the actual asset.
Transition Hard cut in and out Make the connection to adjacent shots an explicit choice.
Motion responsibility Static hold; any later movement requires a separate edit and review Avoid mistaking a motion-oriented prompt for delivered motion.
Audio responsibility Separate sound pass; none supplied by this image step Prevent the still asset from implying a soundtrack.

The three-second hold and crop are examples, not recommended defaults for every project. Choose them for the intended sequence, and leave them visibly unresolved if the editorial decision has not been made.

Separate the still approval from the motion approval

For a static storyboard or slideshow treatment, review the actual crop and the chosen hold duration in the timeline. The image may be approved while its placement still needs adjustment.

If the brief instead calls for a pan, character movement, or an interpolated sequence, route that requirement to a separate editing or generation step. Keep the original still reference, identify the downstream output, and review that output before calling the shot finished. In particular, check whether the intended action happens, whether important framing survives, and whether the transition into the next shot works. These are proposed acceptance checks, not claims that a particular tool can deliver the effect.

Flare’s documented image output does not by itself establish native motion, timing, sound, or temporal consistency. An appealing still cannot settle those questions. Conversely, adding motion later would not demonstrate that Flare generated video—it would demonstrate a separate stage in the workflow.

The useful boundary is small but consequential: Flare supplies the image; the timeline owner supplies its duration and sequence behavior. Keep both responsibilities visible before approving the final video delivery.

I’d add a replacement rule to the handoff contract: approval should follow the exact still revision, not just the shot ID.

For GPT Image 2.5 Flare, the documented image-output boundary means the generated asset and the editor’s timed treatment are separate deliverables. That creates an integration failure worth checking: a new still can replace an old one under the same shot name while the timeline keeps an approval that belonged to the old framing.

My proposed application-side rule is to store the still’s immutable asset revision alongside the timeline revision used for approval. If the still changes, retain the previous review as history but mark the dependent crop and timeline approval as needing review. Don’t silently carry approval forward because the replacement also counts as an image. These are application bookkeeping choices, not Flare response fields.

A useful acceptance test would be to approve a hypothetical shot-07 using frame-07-v2, then replace it with frame-07-v3, where the sign has moved toward an edge. The test passes only if the application exposes the changed dependency and requires the crop to be checked again before export approval. Merely loading the replacement successfully is not sufficient.

This does not require Flare to generate or understand a video timeline. It makes the downstream integration accountable for which particular Flare still was actually reviewed in that timeline. I haven’t run this test; it is a proposed regression case for the handoff workflow.