Extended AD When the Cut Is Too Fast

Generated social cuts are often a wall of speech on a wall of picture. There is no pause long enough for standard audio description, because the generator was not asked to leave one. WCAG 2.2 Success Criterion 1.2.5 (Level AA) still expects description during existing pauses. When those pauses are not there, the AA technique runs out. The next one is extended audio description: freeze the picture, speak, then resume.

That is Success Criterion 1.2.7, Level AAA. It is not required to claim AA. It is the practical way to keep the sense of a fast cut once the visuals matter. The work is editorial: where to freeze, how long to hold, and how to keep a picture master and an AD master from drifting.

Standard AD stops where the silence stops

Standard AD writes into gaps. Fast generated ads, UGC avatars, and native-audio takes often have none: the voiceover filled the runtime, then the picture was cut to it. Cut rate on TikTok, Reels, and Shorts is a retention choice and an AD constraint. A 0.4-second price super mid-sentence is visible to a scroller and invisible to an AD listener.

Try these in order:

1. Fold the visual facts into the hero VO and recut. Price, URL, the gesture that proves the product, spoken in the main track. WCAG notes that if the soundtrack already conveys the important video information, no extra description is necessary. Cheapest fix, and the one most spots should have used at script.
2. Open gaps in the picture master. Hold a frame, pad between VO sentences, let a super sit on a silent bar. Standard AD (1.2.5) becomes possible again. Fine for a landing-page film; often wrong for a 15-second hook.
3. Make an extended-AD version. Freeze at beats with no gap, play the extra description, resume. Ship a second file, labeled, not forced on every viewer. WCAG's note on 1.2.7: use it when the sense of the video would be lost without the extra description and the pauses are too short.

Do not whisper over the VO and call it standard AD. Netflix allows talking over dialogue only as a last resort. A cut with no gap is a missing version, not a last resort.

Freeze, hold, or recut: pick per beat

Extended AD is not one filter. It is a per-beat decision.

Beat type	What the picture does	What the AD version does	Risk
On-screen price, legal, URL with no silence	Continues under VO	Freeze on the super; read it; resume	Freeze is visible; duration grows
Action that is the proof (pour, click, before/after) under VO	Continues	Freeze at the readable frame of the action; describe; resume	If you freeze too late, the proof is already gone
Decorative cut-to-cut montage	Continues	Skip, or describe one representative frame	Over-describing kills pace
Dialogue-on-dialogue (two people, no air)	Continues	Do not freeze unless a visual aside is plot-critical	Freeze during speech feels like a stall
End card with type and music only	Already a hold	Treat as a standard gap; no freeze needed	None

"Freeze" is a still of the current frame for the AD line, then the timeline resumes. "Hold" is a living locked-off shot you extend. Holds look less like a technical pause; use them when you can still recut the shot.

Runtime will grow. A 15-second wall-to-wall cut that needs about six extra seconds of holds becomes a 21-second AD cut. That is expected. Do not keep the AD cut at 15 by speeding the describer; the freeze exists to make time.

WCAG's lecture example is the same pattern: the professor talks while drawing, erases, draws again. The video pauses between problems so the drawings can be described, then resumes. A product montage is that lecture with worse manners.

Two versions, one picture clock

The AD version is the picture master plus time inserted at known frames. If the picture master swaps shot 4, the AD master swaps the same shot 4 and re-times the freeze attached to it.

Keep a shared clock. Versely's default video frame rate is 25 fps, not 24. Write freeze points as decimal seconds plus "25 fps" on the brief. Mixing 24 fps notes into a 25 fps timeline is how a freeze lands two frames into the next super.

A minimum brief:

Picture master: refill-ad-15s-v3.mp4 (25 fps, 15.00s)
AD master:      refill-ad-15s-v3-ad-ext.mp4 (same picture, inserted holds)

Freeze table (picture-master time):
1. 00:03.12  hold 2.0s  AD: "She lifts the empty bottle. A crack runs down the glass."
2. 00:07.00  hold 1.6s  AD: "On-screen: 30-day refill, 24 dollars."
3. 00:13.20  hold 2.2s  AD: "Logo. On-screen: refill dot example dot com."

Rules:
- Do not re-time any picture except the listed holds.
- Holds are freeze-frames of the listed frame, not new shots.
- Hero VO is untouched. Duck only under the AD line, then restore.
- If v4 of the picture master changes a listed frame, stop and return this table.
- Export both files. Do not replace the picture master with the AD master.

That table is the contract. The AD writer fills the spoken lines. The editor owns the holds. If either party "just pads the end," the supers and the words fall out of sync.

Hand the picture master over like any generated cut: clean editor handoff, freeze table in the same brief as the locks. A brief that survives handoff is the envelope; extended AD is one more constraint block, not a Slack voice note.

Building the AD master in the editor

You need trim, hold, speech, and mix, not a specialist AD tool.

1. Lock the picture master. No more picture notes after the freeze table is written.
2. Mark freeze frames. In Versely, trim or cut the video to isolate each segment between holds.
3. Generate AD lines as their own audio, one file per freeze, with style_instructions for a calm, reporting delivery. Not one 21-second read you hope matches the holds.
4. Rebuild as segment, hold+AD, segment. Stitch the clips, or assemble in the video editor as one EDL.
5. Mix AD onto the held frames only. Duck the original under the AD line, then restore. Do not kill room tone.
6. Preview before you pay for the export. The editor's 480p preview pass is free and carries a short per-user cooldown; use it to hear whether a hold is long enough. The final export is a single charge. The 480p preview loop is where you catch a freeze that starts one shot late.

Publish two files. Platforms with a selectable description track can take a described mix; most brand placements cannot. Label the extended cut on the page. If the team later slows the public cut, throw away the freeze table and write standard AD into the new gaps.

FAQ

Does AA require extended AD on every fast social cut?

No. 1.2.7 is AAA. AA requires 1.2.5, which uses existing pauses, and exempts video whose soundtrack already carries the important visuals. A hook with no unique visual beyond the VO does not fail for lack of freezes. A hook whose only price is a super, with no pause to read it, fails the user even when it can still argue AA. That is why you make the second cut.

Can I extend the video with a generative extender instead of freezing?

Only if the extender holds the same frame's information. A generative extend that invents new motion is a new shot, and the AD line you wrote for the original frame will be wrong. For accessibility holds, freeze or recut. Save video extend for picture work, not for making time to talk.

Who signs off the AD master?

The same person who signs off the picture master, against the freeze table, not against "it sounds fine." Each hold starts on the listed frame, the AD line matches the super, and the public cut was not overwritten. If v4 of the picture lands after sign-off, the table is stale.

Should the extended cut run in the same paid placement as the picture cut?

Usually not. Freeze-frame AD changes pacing and length. Use the picture master in the slot it was designed for, and put the extended cut where a user can choose it (site embed, product page, "described version" link). Forcing freezes into a 15-second buy is how both versions get rejected.