Tips & Tricks
A few things worth knowing before you spend render time — none of it is required, but each one changes the result in a specific way.
Enhance — using GFPGAN: quality vs. speed
GFPGAN runs a dedicated face-restoration pass over every single frame. It's the difference-maker for sharpness — but that extra pass takes real time, so it isn't free.
A practical workflow: leave Enhance off while you're iterating, then turn it on for the take you're actually going to keep.
Don't overdo Expression Scale
This slider controls how much the face and head move. It's easy to push too far — here's what each end actually looks like:
- 0.7 keeps the head close to still — useful for a calm, composed presentation feel.
- 1.0 is the natural default and the right starting point for most videos.
- 1.2 already moves the head a lot — treat anything above this as a deliberate, expressive choice rather than a default.
Pitch & Speech Rate — how far to push them
ImTalking's voice comes from Microsoft's Edge TTS engine, which takes pitch as a shift in Hz and rate as a percentage change from the voice's natural speed — that's exactly what the two sliders control. Because both are relative to each voice's own baseline, the same number can feel bigger on some voices than others, so treat the ranges below as a starting point, not a rule.
Pitch
Reads as natural intonation variation. Pushing toward the ±50 Hz limit starts to sound artificial, since it's a large shift relative to most voices' baseline pitch.
Speech Rate
Slows down dense text or adds energy without drawing attention to itself. Near the ±50% extremes, speech starts to sound rushed or unnaturally slow.
Both sliders go to ±50 mainly to leave room for deliberate effect — for standard narration, small adjustments almost always sound better than large ones.
For long scripts, generate several short videos instead of one long one
If you have a large amount of text, it's tempting to paste it all in and generate one long video. In practice, splitting it into smaller chunks — a paragraph or a few sentences at a time — and generating each as its own independent video usually works out better:
- Each render finishes faster, so you can check the result and move on instead of waiting on one long job.
- If one section needs a re-take (wrong emphasis, a mispronounced word), you only re-render that piece — not the whole thing.
- Shorter, independent clips are easier to review, reorder, or drop into a video editor afterward if you want to combine them with transitions, titles, or music.