Skip to main content

Hi there! Use this place to report bugs or request features!

We’ll see you guys around!

3 comments

Log in to comment and vote

Comments3

  • John Simpson

    •

    May 16

    using claude.-Example prompt:

    15-second video. UGC-style. Single actor, single setting, three cuts. Clean visual composition with no graphics or text overlays of any kind. The actor's voice carries the message — the video stays purely cinematic, like a documentary shot rather than a social media post with captions.

    Setting: bright modern kitchen, white worktops, kettle and mug visible in soft background, warm natural light.

    Actor: confident 30-year-old British man from London, well-built athletic frame, short hair, plain black t-shirt. Stood at the kitchen worktop.

    Voice: deep, calm London accent. Conversational pace, like he's having a chat with a mate. Speaks slowly, with clear pauses between each sentence. Not rushed. Not a sales pitch.

    INFRNO product tub: only appears in the final shot, held naturally at chest height. Not visible before the closer. The label reads "INFRNO" — this is the only readable text anywhere in the entire video. No other words appear on screen at any point.

    Dialogue (paced slowly with natural pauses):

    Beat 1 (0–4s): Actor looks at camera. "Started on half a serving." (pause) "Eighteen months later, you're taking three."

    Beat 2 (5–10s): Same shot, gestures naturally. "That's tolerance." (pause) "Same amount stops working. So you take more." (pause) "Until you can't train without it."

    Beat 3 (11–15s): Picks up the INFRNO tub naturally from the worktop, holds at chest height. "Inferno's non-stim. No tolerance." (pause) "In TikTok Shop."

    Brand pronunciation: "inferno" (like the Italian word).

    The issues I'm seeing:

    One — Captions auto-generate on screen despite explicit instruction not to. Happens on roughly half my generations.

    Two — Accent drifts to American or mid-Atlantic around the 8-10 second mark on continuous dialogue. Holds the London accent at the start, loses it on the back half.

    Three — Pacing is too fast even with "slow" and "(pause)" markers in the dialogue. Speech compresses to fit the 15 seconds rather than breathing.

    Four — Some words get mispronounced or substituted with similar-sounding ones (e.g. "scoop" rendered as "slate," "comedown" rendered as "comma-down").

  • John Simpson

    •

    May 15

    On talking head videos (single actor, no cuts, speaking direct to camera), the accent drifts back to American around 8-10 seconds in, even when prompted with strong UK accent enforcement (London, Manchester, Yorkshire). Tried celebrity voice references, explicit accent instructions, repeated reinforcement -accent holds for the first half then defaults to American on the back half., any ideas please

    • Alexander Le

      Team•

      May 16

      Thanks for the heads up John. I’ll look into it. Currently, it’s not a well supported use-case, but with a bit of adjustment from my side it should be possible.

      How do you currently prompt it? Would you mind sending or attaching a screenshot? 😄