How to Edit Video With a Text Prompt

Step 1: Choose a Prompt-Based Editor

Not all AI video editors support true prompt-based editing. The tools that currently offer this capability each implement it differently, so choosing the right one depends on what kind of edits you need.

Descript's Underlord is the most mature prompt-based editing system. You can give it instructions like "remove all filler words," "tighten the pacing by removing pauses," "add studio-quality audio enhancement," or "generate b-roll suggestions for this section." Underlord works on top of Descript's transcript-based editor, which means it has deep context about what is being said in the video and can make intelligent decisions about where to cut and what to keep.

Kapwing offers prompt-based editing through its AI assistant, which handles tasks like "add subtitles with a bold white style," "crop this video to vertical for TikTok," or "remove the intro and jump straight to the main content." The prompts are more focused on formatting and structural changes than creative decisions.

VEED provides prompt-driven subtitle generation, translation, and basic editing commands through its AI tools. You can describe the caption style you want in natural language, and VEED applies it across the video. More complex editing prompts are supported through VEED's text-to-video features.

Some newer tools like Runway and Pika support prompts for generative editing, where the AI creates or modifies visual content based on descriptions. These tools are better suited for video generation than editing existing footage, but the line between the two is blurring.

Step 2: Upload Your Footage

Upload your video file to your chosen editor. Most browser-based tools accept MP4, MOV, and WebM formats. File size limits vary by platform and plan, with free tiers typically capping uploads at 500MB to 2GB and paid plans allowing larger files.

After uploading, the AI processes the video in several ways. It transcribes all spoken audio into text, detects scene changes and speaker transitions, and builds an internal representation of the video's content and structure. This processing can take anywhere from a few seconds for short clips to several minutes for longer videos. Wait for this step to complete before writing prompts, because the AI needs this analysis to understand your instructions in context.

Some tools, like Opus Clip, also accept YouTube URLs directly, which skips the upload step entirely. If your source material is already published online, check whether your editor supports URL-based import to save time.

Step 3: Write Your Editing Prompt

The quality of your prompt directly affects the quality of the output. Good prompts are specific, actionable, and focused on a single change at a time. Vague prompts like "make this better" or "fix the pacing" give the AI too much room for interpretation, which usually produces disappointing results.

Effective prompt examples include:

• "Remove all pauses longer than 1.5 seconds throughout the video"
• "Add word-by-word animated captions in white text with a dark background box"
• "Cut the first 30 seconds and start from where I say 'let us get into it'"
• "Remove every instance of um, uh, like, and you know"
• "Reformat this landscape video to 9:16 vertical, keeping the speaker centered"
• "Split this into clips at each natural topic change"
• "Enhance the audio to reduce background noise and normalize volume"

For best results, give one instruction per prompt and review the output before adding another. Stacking multiple complex changes into a single prompt increases the chance that the AI misinterprets one of them, which can create cascading problems that are harder to undo than individual edits.

When using Descript's Underlord, you can also reference specific parts of the transcript in your prompts. For example, "remove the section where I talk about pricing and skip directly to the demo" tells the AI exactly where to cut based on content rather than timestamps. This contextual understanding is one of the advantages of editors that build on transcript data.

Step 4: Review and Refine the AI Output

Always review the AI's edits before exporting. Preview the full video from start to finish, paying attention to cut points, caption accuracy, and audio transitions. Common issues to watch for include:

• Jump cuts where the AI removed a pause or filler word mid-sentence, creating a visual jump that looks jarring
• Caption errors on proper nouns, technical terms, or brand names that the AI did not recognize
• Over-aggressive silence removal that cuts into natural breathing pauses and makes speech sound rushed
• Reframing errors where the vertical crop loses important visual information at the edges of the frame

Most editors let you undo individual AI edits or manually adjust specific changes without redoing the entire prompt. In Descript, you can click on any part of the transcript to restore deleted content. In Kapwing, you can adjust individual clips on the timeline after the AI has made its changes. Use these manual controls to fix specific issues rather than rewriting the entire prompt.

If the AI's interpretation of your prompt was fundamentally wrong, rewrite the prompt with more specific language and try again. Iterating on prompts is normal and expected, especially as you learn how your particular tool interprets different phrasings.

Step 5: Export the Final Video

Once you are satisfied with the edits, choose your export settings. For YouTube, use 1080p or 4K resolution in MP4 format with H.264 encoding. For social media platforms, vertical 1080x1920 resolution at 30fps is standard. Most AI editors offer platform-specific presets that handle these settings automatically.

Check the export preview for watermarks if you are on a free plan. Some editors that do not watermark the editing preview do add watermarks during export. If your chosen tool adds a watermark on free exports, you will need to upgrade or switch to a no-watermark editor before publishing.

Export times depend on video length, resolution, and whether the editor processes locally or in the cloud. Browser-based editors render in the cloud, which is usually faster than local processing on modest hardware but can be slower on very long videos during peak usage periods.