cuttingfloorStart free
Demo · ChatGPT

Let AI pick your best take, then make the thumbnail

A tech creator gives ChatGPT four takes of his channel intro. cuttingfloor has Jev, its own vision model, watch every quarter second of each take, keeps the best one, then picks the frame where he looks best for a thumbnail.

01

Before

The clip as it came in · 0:40

intro-takes.mp4 · 0:40

What needs to go

Take 1: “Hey, I’m Jake, and today I’m testing whether an AI can pick my best take.” Take 2: the same line, dragging out “an… AI…”, then a wince. Take 3: the same line, starting off at the monitor, flat. Take 4: the same line, a hard squint halfway, then he cracks up. “It actually looks at the frames, not just the words.” (smiles)

Struck in red: 3 things the agent cuts.

02

The edit

In ChatGPT, with cuttingfloor connected

Take 1: “Hey, I’m Jake, and today I’m testing whether an AI can pick my best take.” Take 2: the same line, dragging out “an… AI…”, then a wince. Take 3: the same line, starting off at the monitor, flat. Take 4: the same line, a hard squint halfway, then he cracks up. “It actually looks at the frames, not just the words.” (smiles)

can-ai-pick-my-best-take.mp4 · 0:10
ChatGPTcuttingfloor

  1. Which of these intros is my best take? Use it and cut the rest.

    I’ll read what you said in each take and have Jev look at your face every quarter second, then keep the winner and put your closing line after it.

    • Used cuttingfloor › media.import five clips
    • Used cuttingfloor › media.analyze every word, timed
    • Used cuttingfloor › media.analyze Jev looked at 160 frames
    • Used cuttingfloor › judge.ask stumbles, squint, laugh
    • Used cuttingfloor › render.frames looked at each take
    • Used cuttingfloor › edit.batch take 1, then the closer
    • Used cuttingfloor › audio.set_master louder, even
    • Used cuttingfloor › render.frames checked the cut
    • Used cuttingfloor › export.video vertical, 0:10

    Take 1 wins: you’re on the lens in 32 of 32 frames, eyes never closed, and it has the most energy. Take 2 drags out “AI” and winces, take 3 starts on your monitor and is the flattest, take 4 squints mid-line and ends laughing. Want a thumbnail and a title?

  2. Now pick a thumbnail frame where I look good, and put a title on the video.

    I’ll have Jev check every quarter second for open eyes, looking at the lens, a smile and sharp focus, build the thumbnail from the best frame, and add a matching title and captions.

    • Used cuttingfloor › media.analyze Jev scored 41 frames
    • Used cuttingfloor › render.frames grabbed the best one
    • Used cuttingfloor › create.still built the thumbnail
    • Used cuttingfloor › judge.ask is his face clear?
    • Used cuttingfloor › code.run word-by-word captions
    • Used cuttingfloor › create.still a matching title
    • Used cuttingfloor › render.frames checked the text
    • Used cuttingfloor › export.video vertical, 0:10

    Your best frame is 5.1 seconds in: eyes open, smiling, on the lens. The thumbnail puts “CAN AI PICK MY BEST TAKE?” under your chin, and Jev confirmed your face is clear. The video opens with the same title, with captions that light up each word. Want a 6-second cut for a story?

Shown in your agent: Claude, ChatGPT or Kazi. cuttingfloor does the editing and hands back the file; your agent shows it to you.

03

After

Ready to post · 0:40 → 0:10

can-ai-pick-my-best-take.mp4 · 0:10

What the agent did

  1. Wrote down every word of each take, with when it was said.
  2. Had Jev, cuttingfloor’s own vision model, look at each take every quarter second: looking at the lens, eyes closed, wincing, energy.
  3. Checked which takes drag a word out, squint or crack up.
  4. Picked take 1: on the lens in all 32 frames, no blinks, the most energy.
  5. Trimmed it to where he speaks and put his closing line right after it.
  6. Matched the closing line’s volume, then made it all louder and even.
  7. Had Jev score 41 frames for open eyes, a smile, sharp focus and a closed mouth, and took the best.
  8. Built a thumbnail from that frame and checked the title doesn’t cover his face.
  9. Added the title for the first seconds and captions that light up each word, and exported it vertical.

Cut a dance video to the beat, with no copyrighted music · Edit a two-person interview into a Short · Turn a beauty product review into a 15-second Short · Edit a real estate listing video with ChatGPT · Remove the pauses and restarts from a talking video · Add a talking head to a slide explainer, then make it a Short · All demos