Multi-turn natural-language video editing
Edit a finished clip by talking to it: each instruction builds on the last result, and everything you do not mention stays.
Imgo's video editor runs Google's Gemini Omni Flash (gemini-omni-1.1-flash). Pick a video, describe one change in plain words, and the model returns the edited clip. The next instruction applies to that result, because the edit keeps its state from turn to turn — no timeline, no re-upload, no starting over.
Who this is for
- Marketing teams fixing near-final clips
- Creators iterating on a generated shot
- Editors who need quick variants without a timeline
Why this is hard
One-shot AI edits make you write the perfect instruction up front. When the result is close but not right, you describe everything again and hope the parts you liked survive. In a timeline editor, a change as small as "make it dusk" is a colour-grading job.
How Imgo solves it
Edits that remember
Imgo passes the previous turn to the model, so each instruction edits the latest result. Earlier changes carry forward without re-uploading the video or restating them.
What you do not mention stays
An instruction is a change request, not a regeneration: the subject, motion and audio you did not ask to change are preserved.
A reference image when words fall short
Attach one image — a product, a face, a style — to guide a specific edit, such as matching a jacket or a packaging design.
The 3-step flow
Step 1
Pick a video
Choose a clip from your Imgo library — one you generated or uploaded. Each edit works on one source video.
Step 2
Describe one change
Write a single instruction, such as "make it dusk" or "remove the cables behind the product". One clear change per turn is easier to check than five at once.
Step 3
Refine, then download
Send the next instruction to build on the result, or continue from an earlier turn to try another direction. Every result lands in your library.
Recommended tools
The Imgo surfaces this use case routes into.
Prompts you can copy
Replace every bracketed variable with your own product, audience and setting before running one.
Day to night, one step at a time
Turn 1: Make it dusk, with warm streetlights coming on. Turn 2: A little darker, and add light rain on the pavement. Turn 3: Keep everything else; make the rain softer and add the sound of rain.
Clean up a product shot
Turn 1: Remove the cables and clutter behind the [product]. Turn 2: Replace the back wall with [background]. Turn 3: Slow the camera move to half speed; keep the [product] and its label unchanged.
Restyle without losing the subject
Turn 1: Restyle the whole clip as [style]; keep the person's face and movement. Turn 2: Less saturated — closer to the original colors. Turn 3: Use the reference image for the jacket; change nothing else.
FAQ
What does multi-turn mean here?
Each instruction applies to the result of the previous one. The model keeps the state of the edit between turns, so you never restate earlier changes or re-upload the clip.
Which model does the editing?
Google's Gemini Omni Flash, called by its stable model id gemini-omni-1.1-flash, inside Imgo's video editor. It is live on Imgo today.
Will it change parts I did not mention?
It is built to preserve them, but check faces, on-screen text and audio after each turn. If something drifts, name what must stay fixed in the next instruction.
Can I go back to an earlier version?
Yes. Every turn's result is saved to your library, and you can continue editing from any of them instead of only the latest.