|
| 1 | +--- |
| 2 | +title: Creating Videos with AI |
| 3 | +sidebar_position: 5 |
| 4 | +--- |
| 5 | + |
| 6 | +FrameScript is a text-based video editor, so it pairs extremely well with LLM-based coding agents. |
| 7 | +This guide shows how to create a video using a coding agent. |
| 8 | + |
| 9 | +## 1. Prepare a coding agent |
| 10 | + |
| 11 | +First, pick a coding agent. |
| 12 | +Good options include [`Codex CLI`](https://developers.openai.com/codex/cli/), [`Claude Code`](https://code.claude.com/docs/en/overview), and [`Gemini CLI`](https://geminicli.com/). |
| 13 | +Personally, I recommend `Codex CLI`. |
| 14 | + |
| 15 | +## 2. Create a project |
| 16 | + |
| 17 | +Run the following command in any directory: |
| 18 | + |
| 19 | +```bash |
| 20 | +npm init @frame-script/latest |
| 21 | +``` |
| 22 | + |
| 23 | +This will create a project via an interactive prompt. |
| 24 | + |
| 25 | +Move into the created directory: |
| 26 | + |
| 27 | +```bash |
| 28 | +cd <project-path> |
| 29 | +``` |
| 30 | + |
| 31 | +Then run: |
| 32 | + |
| 33 | +```bash |
| 34 | +npm run start |
| 35 | +``` |
| 36 | + |
| 37 | +## 3. Open your editor |
| 38 | + |
| 39 | +Launch your preferred editor in the created project directory. |
| 40 | + |
| 41 | +## 4. Start the coding agent |
| 42 | + |
| 43 | +Open a new terminal and start the agent in the project directory. |
| 44 | + |
| 45 | +## 5. Give clear instructions |
| 46 | + |
| 47 | +This is the most important part. |
| 48 | +FrameScript bundles the editor inside the project directory, which is a bit unusual. |
| 49 | +The AI needs to understand that setup. |
| 50 | +We recommend instructions like: |
| 51 | + |
| 52 | +> First, read `AGENTS.md` to understand the project overview. |
| 53 | +
|
| 54 | +Then follow with something concrete, for example: |
| 55 | + |
| 56 | +> I want to create a video here. |
| 57 | +> First, please do ~~~. |
| 58 | +
|
| 59 | +## When AI works best |
| 60 | + |
| 61 | +AI is often good at adding motion to images or text. |
| 62 | +However, cutting videos or fine-tuning positions can be tricky, because LLMs cannot view the preview. |
| 63 | +For detailed positioning, ask the agent to expose values as variables and adjust them yourself. |
| 64 | +For video cuts, it is usually faster to do a rough pass manually. |
| 65 | + |
| 66 | +## Gallery |
| 67 | + |
| 68 | +In this example, the outlined text, character placement, and radial lines were created with AI instructions. |
| 69 | + |
0 commit comments