AI video for a history channel: one recurring host, a 60-second shot plan and its credit bill

The problem with most AI history shorts

Search for how to make a faceless history channel and you get the same workflow on twenty pages: write a script with a chatbot, check the facts, generate a voiceover, lay AI images or stock clips under it, cut it in an editor. One tool vendor alone publishes close to twenty near-identical guides on this. They cover the voice and the slideshow. None of them covers the thing that makes a history channel recognisable from one short to the next: a host the viewer sees, who comes back every episode, standing in Rome one week and in a 1920s newsroom the next.

That host is the hard part, because the era changes everything around them. New costume, new light, new set, every episode. This page is how we plan it, shot by shot, with the credits each shot bills.

Split the host into two blocks

Every prompt for a host shot has a character block and a scene block.

The character block describes the person and never changes, word for word, across the whole channel. It covers face, hair, build and age, and nothing they wear:

Theo, a man in his late forties, lean face, short grey beard trimmed close, deep-set dark eyes, receding salt-and-pepper hair cut short, thin silver-rimmed round glasses.

The scene block carries the era: the costume, the place, the light and what Theo does.

Scene: a Roman street market at midday, 1st century. Theo wears a plain undyed wool tunic and a brown cloak pinned at the shoulder. He turns from a fruit stall to the camera and speaks.

The glasses are deliberate. A history host changes clothes every week, so the face needs one object that stays when the clothes go. Round glasses read in a toga and in a trench coat. Some guides call this a "signature prop"; we treat it as part of the face.

The references follow the same split. Theo's reference sheet shows him head and shoulders in a plain dark crew-neck, front, three-quarter and side, under even light. Nothing on the sheet is an outfit the engine could cling to. If the sheet showed him in a tweed jacket, the jacket would creep into every era. Our outfit change guide goes into this in more detail.

The eight-shot plan

Sixty seconds, eight shots. The host opens, the era takes over, the host returns. This is the plan we use:

#	shot	length	audio	engine	credits
1	Theo close-up, the hook line	4 s	voiced	Kling O3 Pro	120
2	era wide, the street or battlefield, no faces	8 s	silent	Veo 3.1 Fast	160
3	Theo medium, in costume, first claim	12 s	voiced	Kling O3 Pro	360
4	era action, crowd or ships, no face read	8 s	silent	Veo 3.1 Fast	160
5	insert: a coin, a map, a letter	6 s	silent	Veo 3.1 Fast	120
6	Theo medium, the turn in the story	12 s	voiced	Kling O3 Pro	360
7	era wide, aftermath	6 s	silent	Veo 3.1 Fast	120
8	Theo close-up, sign-off	4 s	voiced	Kling O3 Pro	120
	total	60 s			1,520

Why this split. Theo is on screen in four shots and speaks in all four, so those run on Kling O3 Pro at 30 credits a voiced second: it takes up to five characters as elements and renders up to 15 seconds in one shot. The era shots have no face to hold, so they need no references. That lets them run silent on Veo 3.1 Fast at 20 credits a second, in its 6 and 8 second lengths. Without references Veo serves 4, 6 or 8 seconds; with references every Veo shot is 8.

The wides carry no dialogue on purpose. A wide shot leaves the face only a small patch of the frame, which is where faces drift. Our wide-shot guide explains why we keep the host out of them.

The same episode on one engine

If you would rather keep one engine for everything:

plan	credits for 60 s at 720p
eight-shot mix above	1,520
all Kling O3 Pro, host voiced, wides silent	1,632
all Kling O3 Pro, everything voiced	1,800
all Seedance 2.0	2,700
all Seedance 2.5	4,200

Seedance bills by resolution with sound included, so silent wides save nothing there. Seedance 2.5 does earn its rate on one kind of shot: a long continuous move through a scene, up to 30 seconds in one take, with up to 30 reference images. For a battle that sweeps past the host, that one shot can be worth 70 credits a second.

Narrator on camera or off

There are two ways to run the voice.

The host speaks on camera. What the plan above does. The four Theo shots are voiced renders; his mouth moves with his lines. The wides carry ambient sound only, or the next line of narration laid over them in the edit.

The host narrates off camera. Render every shot silent and record the narration over the cut. On Kling O3 Pro that drops Theo's 32 seconds from 960 credits to 768, since silent is 24 credits a second. The cost is that Theo never visibly speaks, which makes him a picture in the corner of a slideshow. We keep at least the opening and closing shots voiced so the viewer sees who is talking. Our narration page prices the trade in full.

Either way the voice has to be the same every week. On Kling O3 Pro a character can carry a cloned voice: 50 credits once to clone your own, and the voices in our bank cost nothing. Our page on keeping a voice across episodes covers the rest.

What changes per era, what does not

stays fixed across the channel	changes every episode
character block, word for word	costume, written in the scene block
reference sheet, plain top	location and period
round glasses	light: torchlight, daylight, gas lamps
voice	music bed and ambient sound
shot plan shape: host, era, host	the hook line

When an era pushes Theo's face around (candlelit interiors make faces read harder and more angular), add "soft, even light on his face" to the scene block before you touch anything else.

Two rules we keep on facts

First, AI renders invent detail. A Roman street will have whatever the engine thinks Rome looked like. We keep claims in the narration and treat the picture as illustration, and we never put a specific inscription, date or name in a rendered sign, because the engine will misspell it. Put text on screen in the edit, where you control it.

Second, YouTube asks creators to disclose realistic content that was made or significantly changed with AI. A stylised host on an illustrated set reads as illustration; a photoreal "archive" shot of a real event can mislead a viewer. We label it.

The studio keeps Theo saved with his references and his voice, so each new episode starts from the shot plan, and the credit quote for every shot shows before it renders.

FAQ

How do I make an AI history channel with the same host every episode?

Write the host's face, hair and one signature item, such as round glasses, as a fixed block, and put the costume and era in a separate scene line. On ReelDouble the saved host carries the same references and voice into every episode.

How much does a 60-second AI history short cost to make?

On ReelDouble our eight-shot plan costs 1,520 credits at 720p: four voiced host shots on Kling O3 Pro and four silent era shots on Veo 3.1 Fast. All Seedance 2.5 would be 4,200 credits.

How do I change the costume of an AI character for different time periods?

Keep the reference images neutral, in a plain top, and write the period costume only in the scene description. If the references show a specific outfit, that outfit tends to appear in every era.

Should an AI history host talk on camera or narrate?

Voice at least the opening and closing shots on camera so viewers see who is speaking. Era wides can run silent with narration over them, which on ReelDouble cuts Kling O3 Pro shots from 30 to 24 credits a second.

Do I need to disclose AI in YouTube history videos?

YouTube asks creators to disclose realistic content made or significantly altered with AI. A photoreal recreation of a real event is the clearest case, so label it in the upload settings.