How to Make Long AI Videos Without Stitching 8-Second Clips
By Steven Van ·
Synthesys says its agent turns one brief into a 10 to 30 minute video, with a storyboard and credit cost shown before anything renders.
Synthesys has published a guide to making AI videos of 10 to 30 minutes from a single brief, without joining short clips in an editor. The post argues that most Synthesys alternatives stop at 4 to 15 seconds per generation, so a 30-minute video means 120 to 225 clips before voiceover, captions, music and editing. The details are in the vendor's guide.
The clip arithmetic
Using each tool's published limits, checked on 28 September 2026, the guide counts clips for 1,800 seconds of video:
- Google Veo 3.1 at 8 seconds: 225 clips. With Extend, each chain tops out at 148 seconds, so 13 separate chains.
- OpenAI Sora 2 at 12 seconds: 150 clips.
- Runway Gen-4 at 10 seconds: 180 clips.
- Higgsfield at 15 seconds per shot: 120 shots.
The workflow
The Synthesys agent takes one brief and builds the video in these steps:
- Pick a format and length. Videos run from about 10 seconds to 30 minutes, in 9:16, 16:9 or 1:1.
- Describe the video in one sentence.
- Review the script and storyboard the agent writes. For long videos, the guide suggests checking chapter order first, then the lines.
- Check the credit cost shown on the approve button. Nothing renders until approval. Scripts, storyboards, voiceover, music and chatting with the agent are free.
- The agent voices the video, generates footage, adds captions and music, and assembles the cut.
- Edit by chat: reorder scenes, rewrite a line, change the voice or music, or ask for another language.
What Synthesys saw in its own test
On 28 September 2026, the team ran a 10-minute faceless YouTube explainer on compound interest in the Faceless Story format, with default settings. The storyboard came back in about two minutes as 62 scenes and a planned length of 832 seconds (13 min 52 s), roughly 39% over the request. One chat message asking for 90 seconds produced a 29-scene, 120-second plan in under 50 seconds, about a third over. Neither plan was rendered.
The team then rendered a 60-second version. The storyboard had 12 scenes. The estimate under the prompt box said about 60 credits, but the approval card asked for 120, and that is what was charged. A wrong figure in scene 10 was corrected through chat in under a minute at no cost, since nothing had rendered. The finished 1:01 video (1080×1920, with narration, captions and music) was in the editor 6:32 after the brief. The team says the voice was right, while the captions and some of the footage were not.