Today, I’m letting AI do my job. While I make myself a coffee, an AI video agent finds, picks and cuts YouTube Shorts in DaVinci Resolve, all on its own. AI is going to take all our jobs anyway, so let me show you how DaVinci Resolve AI automation can work for you, too.
Full video: https://youtu.be/FO4KWphUP9w
The task is simple: we have a video of a lecture, and we want five finished, rendered YouTube Shorts from it. Each one should end with the words “The End”. I don’t touch the timeline, the Media Pool or the Deliver page. The AI does everything.
Part 1: Connect DaVinci Resolve with AI
First, we hitch the horse to the wagon. You need the Studio version of DaVinci Resolve and an AI assistant. You can use Claude, ChatGPT, Gemini, or even a local model.
In the File menu, choose Setup AI Assistant and pick your assistant. In my case, it’s Claude Desktop. It asks for an installation, and that’s it. Your DaVinci Resolve MCP connection is ready.
From now on, the AI can control DaVinci Resolve by itself. You can even close Resolve. The AI will launch it when it needs it.
Part 2: How to Write the Prompt
The deputy can’t read minds, but he sure can read instructions. You have to tell the AI exactly what you want. The more precise you are, the better the result.
In the AI chat, I type: “We’ll create a clip in DaVinci Resolve. I’m attaching the prompt.” Then I drag in a text file with the prompt that describes exactly what I want. I hit Enter, and the AI connects to DaVinci Resolve and gets to work.
You’ll find the full prompt at the end of this article. Use it as inspiration, not as a template to copy blindly.
Part 3: What the AI Does, Step by Step
Best seat on the ranch: I just sit back and watch the hired hand work. I recorded the whole process on two monitors. DaVinci Resolve on the right screen was controlled by the AI the entire time, with no human input.
Project setup
First, the AI launches DaVinci Resolve and creates a new project. Before any timeline exists, it sets the project to 4K vertical format with a frame rate matching the source video. This order matters, because some project settings can’t be changed once a timeline is created.
Importing the media
The AI imports two files into the Media Pool: the lecture itself, and a short end clip with “The End” that goes at the end of every Short.
Transcription and choosing the best moments
DaVinci Resolve transcribes the lecture audio into text. The AI reads the transcript and looks for moments that would grab someone who has never seen the lecture: surprising facts, strong statements and clear explanations.
The rules for each Short
- Each clip runs 25 to 35 seconds.
- It must make sense from the very first sentence.
- Every cut lands between sentences, with a little room at the start and at the end.
- Only after all five clips are picked does the AI load the exact timing of each word, so every cut hits the right frame.
Building the timelines
For each Short, the AI builds its own timeline. Each timeline holds only the chosen clip, followed immediately, with no gap, by the end clip.
Smart Reframe for vertical video
The lecture was shot in landscape, but a Short is vertical. This is where a built-in DaVinci Resolve AI tool comes in handy: Smart Reframe. Resolve finds the speaker by itself and follows him in the vertical frame.
Rendering
Finally, the AI switches to the Deliver page, loads my go-to YouTube render preset for each Short, renders all five and saves them one by one to my drive.
YouTube texts included
That’s AI video editing from start to finish. On top of that, the AI prepares YouTube text for each Short: title ideas based on several templates, a short description, a few hashtags and keywords.
Part 4: Publishing the Finished Videos
Look over the herd before market: we don’t sell the lame ones. Now I have five finished Shorts on my drive. I play each one and check whether it’s okay.
In practice, the quality is decent. I usually toss one or two as unusable and upload the rest to YouTube. While they upload, I fill in the details right away, because the titles and descriptions are already prepared.
Write Your Own Prompt
The main thing I want to tell you is this: don’t blindly copy my prompt. Write your own. It doesn’t have to be perfect. Just try it. You’ll see it’s surprisingly easy. All you need is to know what you want and tell the AI. It will figure out exactly what needs to be done.
Next time, we can look at some other DaVinci Resolve magic. That is, if AI doesn’t take our jobs first. In that case, it might be more interesting to talk about how to rest more instead of working, and where to get the money for it.
Before You Use My Prompt
This prompt is here to demonstrate the process. Feel free to use it as inspiration, but adjust it to your own setup first:
- File paths: replace
D:\AI davinci\video.mp4,D:\AI davinci\theend.movand the output folderX:\with your own locations. - Render preset: “YT-H265” is my own preset. Use the name of your preset, or describe the render settings you want.
- Transcription language: change “English” to the language spoken in your video.
- Tool names:
run_scriptandrun_script_unsafeare the tool names used by Claude with the DaVinci Resolve MCP connection. Other AI assistants may name them differently. - Other connectors: if you have other connectors enabled in your AI assistant, the line “Do not use web search or any other connector” keeps the AI focused on DaVinci Resolve only.
You don’t need my prompt at all. You can create a similar one on your own: just tell the AI in your own words what you want, and ask it to turn that into a prompt. It doesn’t have to be perfect.
One more thing from practice: I get better results when I give the AI smaller tasks. A complex task like this one tends to produce more problematic results. For example, a Short may include parts that don’t belong on screen, or, when there are several speakers, Smart Reframe doesn’t always focus on the person who should be in the frame. So treat this as a demo of what’s possible, not as a one-click production workflow.
The Full Prompt
Work in DaVinci Resolve Studio using the DaVinci Resolve tools. Complete the whole workflow on your own and do not ask me anything unless something fails. If the sandboxed run_script tool throws an error, use run_script_unsafe, but only for DaVinci Resolve operations. Use only the DaVinci Resolve tools. Do not use web search or any other connector. Note: each script call has a 60-second limit. For long operations (transcription, Smart Reframe, rendering), start the operation, then check its status in repeated separate calls until it is finished. A timeout does not mean the operation failed. 1. SETUP - Launch DaVinci Resolve if it is not running. - Create a new project named "short test AI". If that name already exists, use "short test AI 02" (03, and so on). Never overwrite or delete an existing project. - BEFORE creating any timeline, set the project to vertical 4K: 2160x3840, with the timeline frame rate matching the source video's frame rate. Set the transcription language to English. - In the new project, set Retime and Scaling → Scaling to "Fill". 2. IMPORT - Import these two files into the Media Pool: D:\AI davinci\video.mp4 D:\AI davinci\theend.mov - Create a timeline named "FULL - source" containing the full lecture. Keep it untouched as a backup. - Report the resolution, frame rate and duration of both files. 3. TRANSCRIBE AND ANALYZE - Transcribe the audio of the lecture clip in English. - To save tokens, first read only the segment-level transcript (start timecode + text) and pick candidates from it. Load word-level timings only for the chosen passages when you need exact cut points. - Transcript timecodes are source timecodes. Convert them to frames using the clip's Start TC before cutting. - Choose the 5 most interesting passages for YouTube Shorts: - Each passage is 25–35 seconds long, not counting theend.mov. - Each passage must work on its own from the very first second: the opening sentence has to make clear what it is about. Do not start with "so", "and", "well", "okay", "as I said" or with references to earlier parts of the lecture. - Each passage must end at a logical conclusion, so the Short delivers a complete and useful piece of information. - Prefer surprising facts, strong claims, clear explanations and moments appealing to a general audience. - Every cut starts and ends at a sentence boundary. Add about 0.3 s of padding at the start and a few frames at the end, without catching the neighbouring words. - The passages must not overlap. 4. EDIT For each of the 5 shorts, create a separate timeline named "Short 01" through "Short 05". Each timeline contains: - only the selected passage of the lecture (video and audio in sync, nothing else from the lecture), - its own copy of theend.mov placed directly after the passage, with no gap. theend.mov must appear at the end of every short, i.e. 5 times in total, once per timeline. - Make sure the timeline being edited is the active (current) timeline, then run Smart Reframe on the lecture clip only (not on theend.mov), so that Resolve automatically finds and follows the speaker in the vertical frame. - The scripting API does not allow setting Object of Interest or Mode, so Smart Reframe runs with its default settings. Do not try to replace it with manual scaling or positioning. - If Smart Reframe fails on any clip, report which one and leave that clip untouched. - Set the Scaling of theend.mov to "Fit" on the clip itself, so it fits the 2160x3840 frame without cropping. - In the final report, remind me to check the reframing on all 5 shorts and, if needed, switch the Mode to "Pan only" in the Inspector and press Reframe manually. 5. RENDER - Switch to the Deliver page. - For each Short timeline, load my render preset "YT-H265". - After loading the preset, make sure the render resolution is 2160x3840, because the preset must not override the vertical format. - Set the output folder to X:\ and the file names to 01, 02, 03, 04, 05, matching the short numbers. - Before rendering, check that X:\ exists and has enough free space. - Add all 5 render jobs, start rendering, wait until it finishes and confirm the result for each file. 6. OUTPUT (in chat, English) For each short, write: - the short number, - source timecode range, length, and the first and last few words of the passage, - 3 title options (see TITLE TEMPLATES), - a short video description (2–3 sentences), - 10 relevant hashtags on one line, - 10 relevant keywords on one line, separated by commas. Separate the shorts from each other with a horizontal line (---). TITLE TEMPLATES For each short, pick the 3 templates that fit the passage best and write one title per template: - Type A: [Problem] + [Action/Solution] Example: "Asteroid on Collision Course? NASA Has a Plan" - Type B: [Topic]: [Specific Question]? Example: "Asteroid Mining: Could We Really Move a Space Rock?" - Type C: [Mainstream belief] + [Denial/Surprise] Example: "You Think Asteroids Are Too Big to Move. They're Not." - Type D: [Surprising number or fact] + [Context] Example: "A 7-Meter Rock Could Teach Us to Save Earth" Title rules: - Max 60 characters per title; put the hook in the first 40 characters. - Every claim, number or surprise in a title must actually be said in that passage. No invented facts, no exaggeration beyond what the speaker says. - Plain title case, no emojis, no ALL CAPS words. At the end, briefly list any problems you ran into and anything you did differently from these instructions.
Source video used for this demo: “State of NASA” Events Highlight Agency Goals for Space Exploration – watch on YouTube.