CANVID’s generative AI features just keep getting better. AI Retakes is an amazing new feature, but since it’s still in beta and can be a little picky at times, today I’m going to share some tips from my own experience to help you get the best results.
What is AI Retakes?
In simple terms, AI Retakes lets you replace a section of your voiceover directly inside CANVID — either by typing what you want to say (text-to-speech), or by recording a new version using your microphone.
I recently used AI Retakes while recording a walkthrough video about CANVID’s AI features. Without this feature I would likely have had to retake my video
What happened?
After recording and generating the transcript, I noticed a small but important mistake. In one section, I said “CANVID’s noise cancellation” instead of the correct brand term “CANVID’s AI Voice Clarity.”
Normally, that kind of error would mean re-recording the entire video, syncing the audio, and hoping everything matches again.
But not this time.
With CANVID’s AI Retakes, I simply used the power of AI to generate a new audio clip with the corrected words — and it automatically regenerated my lips and lower face to match the new audio.
Even if the feature is still in beta and actively being improved, it's very useful and can be incredibly time-saving.
To help you get the best results from AI Retakes, here are 4 Pro tips that make this feature unstoppable.
Pro Tip #1: Replace More Than Just a Word
First, you need to find the sentence that contains the words you want to replace.
I highlighted the entire sentence, then clicked “Replace Audio.”
You might ask: “If I only needed to change a word or two, why highlight the entire sentence?”
Here’s why:
Even if you’re changing just a couple of words, it’s usually better to replace the whole sentence (or at least a larger chunk). Doing so helps to generate smoother, more natural-sounding speech that blends perfectly with the rest of your audio.
Moving to the Replace Audio Window
Next, we’re in the Replace Audio window.
I selected Local as the processing mode — this uses the power of your Intel AI PC for AI Retakes.
At that point, it was easier for me to just type the replacement words, so I went with the text-to-speech option—especially since I’d noticed the error some time after the original recording and had already changed my mic setup.
Once selected, CANVID shows a visual summary of the sentence I highlighted. I pressed play to preview it and noticed something: the first word, “CANVID,” was slightly cut off. That needed fixing.
This is where the waveform (for now*) becomes your best friend. You can easily see where your sentence starts and ends — and make precise adjustments.
NOTE: We are working on automatic, even more verbatim transcription with super high accuracy word timestamps, so these type of issues would likely be an exception and not the norm in the future
Pro Tip #2: Give Your Selection Breathing Space
When you’re selecting audio to replace, make sure there’s a bit of spacing before and after the sentence or chunk of words you aim to regenerate.
You don’t want the neighboring words to blend phonetically with what you’re replacing.
If you’re replacing a full sentence, this usually takes care of itself.
So in my case, I simply dragged the handle slightly to the left, using the silent space in the waveform as my guide. The right side looked fine, so I left it as is.
Then I previewed again to make sure everything sounded natural.
Pro Tip #3: Always Preview Before Generating
Before generating your retake, always preview your selection to confirm that CANVID captured the start and end of your audio correctly.
Trust me, this step saves you from having to redo things later.
Once I was happy with my selection, I clicked Next Step: Replace Audio with Text.
Typing the Replacement Line
On the next screen, you’ll see two boxes side by side:
- On the right is your original sentence.
- On the left is where you’ll type (or paste) your new version.
What I usually do is copy the original sentence, paste it into the left box, and then edit the parts I want to change.
Once that’s done, we’re ready for the final rule.
Pro Tip #4: Match (or Exceed) the Original Length
Here’s something important to remember:
The results from AI Retakes are better if your replacement phrase is the same length — or slightly longer — than the original words you’re replacing.
This helps the AI match timing and facial motion more naturally.
The Moment of Truth
Alright, time to put AI Retakes into action!
I clicked Finish, waited a short while for processing (it doesn’t take long), and then — voila! — it was done.
Back in the editor, the sentence block I worked on was now underlined, making it easy to spot.
I pressed play on the preview and couldn’t believe it — I was suddenly saying the corrected phrase exactly as I typed it earlier. It looked and sounded seamless.
For visual reference, here’s a look at that section of the video before and after applying AI Retakes:
Check out this link where I've uploaded the CANVID project file with the Retakes already applied. Feel free to use the project file to try your own retakes if you'd like! And remember, don’t forget the pro tips!
You’ll also find the before-and-after videos I've exported, so you can quickly compare the results.
Final Thoughts
In just a few minutes, I managed to fix a voiceover mistake that would’ve taken hours to re-record and sync manually. AI Retakes not only saved me time but also proved how advanced CANVID’s AI technology has become.
If you haven’t tried it yet, I highly recommend giving it a spin — and remember the 4 golden rules for best results!
P.S. I’ve also summarized what happened in the video above, in case you’re the type who prefers watching over reading!
