How to make AI history videos from a researched script

To make an AI history video, start from a researched script, then make one picture per scene: a generated still in the period's look, or a real archival photo you have the right to use. Animate each picture into a short clip that opens on it, voice the script, and cut the clips to the narration. The AI pictures are reconstructions, so label them and keep the facts in the narration.

The Sume steps below come from the Image API, Video generation, and Timeline 1.0 docs and the TTS schema in the Sume API reference, read on 2026-09-28.

What does an AI history video need?

Five parts, in this order. The script comes first because every picture illustrates a line of it.

How do I make period scenes with AI?

Write one image prompt per scene and name what dates it: the year, the place, the clothing, the buildings, the light, and the camera or painting style of the era. Send each to POST /v1/images in the shape of the final video, 16:9 for YouTube long-form or 9:16 for Shorts. Review every still before you spend on video: a generated image is invented, not recorded, and it can put the wrong uniform, flag, or building in a scene.

The Image API returns stills at signed URLs, so save a copy of each approved still and send it as a frame from a public HTTPS URL you control.

How do I animate old photos and maps?

Send the picture as the first_frame in frame_images on POST /v1/videos, and describe only the motion in the prompt: a slow push-in on a map, a gentle pan across a crowd, smoke drifting from a chimney. The clip opens on your picture and moves from there. Can AI animate old photos? covers scanning a print; Restore old photos with AI repairs damage first.

Movement has to come from this step. A still placed straight into a Timeline slot is held static, and any motion on it is ignored.

How do I narrate and cut the video?

Voice the script with TTS 1.0 and ask for timestamps: { "words": true }: the result lists each word with start and end seconds, so you know when each sentence begins and can start its shot there. Then one Timeline 1.0 render lays the clips on the narration; in current code the render takes its sound from the narration (plus an optional music bed) and drops the audio of each clip. The default output is 1080×1920 for Shorts; set output.width and output.height for 16:9. Every URL in the render must be a Sume-hosted file, such as the clips and narration from your earlier Sume jobs. For a video longer than a few minutes, How to make an AI documentary video shows how to voice and join chapters.

How should I label AI history scenes?

Say on screen, or in the description, which pictures are AI reconstructions and which are archival. Keep the claims in the narration, where you can source them, and let the pictures illustrate. Whether you may use a given archival photo depends on its rights, which only you can check for your use.

• Do not present a generated scene as real footage or a real photograph.
• Do not generate a portrait and call it the likeness of a real person.
• Keep your research notes with the script, so a correction reaches the narration.