AI can create new footage, animate a presenter, improve a recording, or make an illustrated explainer from a script. Choose based on what viewers need to see: an imagined scene, a human performance, real events, or clear steps and ideas.
For a narrated explainer or product demo, explainroo is the best fit. It is a free, open-source kit that lets a coding agent create a video from a script and drawn scenes. Other methods work better when you need AI-generated live-action-style footage or edits to camera footage.
Five ways to make a video with AI
| Approach | What AI does | Best for | Main trade-off |
|---|---|---|---|
| Text-to-video | Creates new moving images from a written description | Imaginative scenes, concepts, and illustrative B-roll | The result depicts a generated scene, not proof that an event happened |
| Reference-guided video | Uses images or video to guide a generated result | Keeping a product, character, or visual style recognizable | References guide the output but do not guarantee exact details |
| Script-to-avatar | Turns a script into a video with a digital presenter and synthesized voice | Training, onboarding, and repeatable updates | The presenter may not deliver the distinctive feel of a real person |
| Performance-driven animation | Transfers a person’s expressions, speech timing, and gestures to a digital character | Animated dialogue that needs human acting | Requires a recorded performance and permission to use a person’s likeness |
| AI-assisted editing | Transcribes, captions, trims, or modifies footage you already recorded | Interviews, tutorials, demos, and social clips | The original footage remains central, and generated repairs need review |
Text-to-video creates images from scratch. Reference-guided video uses visual inputs, such as a character image or opening frame, to guide the result. Google’s Veo 3.1 documentation describes both approaches, along with extending eligible generated clips. Check that the action, product details, and text look right in both types of video. Research benchmarks also find problems with consistency, object interactions, and spatial relationships in generated video (VBench, T2V-CompBench).
Avatar videos start with a script. The software provides a digital presenter, voice, and visual layout; HeyGen’s studio workflow is one example. Performance-driven animation starts with an actor’s recorded performance and transfers its timing and expressions to a character, as described in Runway’s Act-Two documentation.
When the original footage matters, use AI to edit it rather than replace it. Transcription can make spoken-word editing easier, while captions and reframing help prepare clips for different formats. Adobe’s Generative Extend can add frames at a clip’s beginning or end, but it does not generate or extend spoken dialogue.
Use explainroo for narrated explainers and product demos
For an explainer that needs narration and visual examples, explainroo turns a coding agent into a video-making assistant. The agent writes script.md, which contains the spoken words, and scenes.js, which draws the scenes. explainroo then handles narration, timing, visuals, sound, and MP4 export on your computer.
Start by pasting this prompt into a coding agent that can run shell commands, such as Claude Code, Codex, Pi, OpenCode, or Gemini CLI. Replace the brackets with your subject:
Make me a short explainer video about [your topic]. Use explainroo for it: clone https://github.com/vincentsch/explainroo, read its AGENTS.md and follow the steps.
The agent sets up explainroo, creates and checks the video, then gives you an MP4. For a product demo, name the product and point the agent to its code or website. The agent can rebuild the screens and show a pointer clicking buttons or entering text while the narration explains the product.
Kokoro, an open voice model, provides narration in American and British English. Whisper, a speech-recognition tool, marks when each word is spoken so scenes appear on cue. Chrome draws the frames, and the video includes background music and sound effects. Captions come in tall and square formats, as well as 4:5. You can choose among six built-in visual looks and adjust the pace in video.json. The agent checks scene stills and text layout and looks for speech errors, but it cannot watch the finished video.
See explainroo’s examples and documentation. It does not edit existing camera footage or create talking avatars or AI-generated live-action footage. It also has no drag-and-drop editor: describe the video to your agent and ask for changes. Optional AI illustrations cost extra; the rest costs nothing per video with built-in voices, while the coding agent is a separate service with its own pricing.
Choose based on what must stay real
Use recorded footage when viewers need to see a real person, product action, place, or event. Choose an avatar for consistent, script-led communications, and performance-driven animation when expressive acting matters but the final character can be digital. Generate new scenes for illustrative or imaginative visuals. For a narrated explainer or product walkthrough, explainroo offers a direct route from a topic to a scripted, illustrated video.
Whichever method you use, check names, numbers, product details, permissions, and any generated scenes that viewers could mistake for real evidence. A video that looks convincing still needs editorial review.