How to Make a Documentary-Style Video With AI (2026)

To make a documentary-style video with AI, research and lock the script first, record the narration second, and only then make the pictures: period-style stills, restored and gently animated archive photos, short image-to-video b-roll, and AI-drawn maps or timelines, all cut together under the voice with music and captions. On VdoBloom a narrated 3-minute history or nature episode built this way costs about 480 credits, roughly $24 at the Lite rate of $0.05 per credit. The rule that matters more than any model: label what you recreated, and never fake footage of a real event or a real person.

Updated October 2026.

Disclosure: VdoBloom is our platform.

What “documentary-style” means for an AI video

The YouTube explainer documentary (a shipwreck, a forgotten inventor, a migration season) runs on a calm narrator, a sourced script, period-looking visuals and a slow pace. AI is good at the visuals and the voice. It is not a research tool you can trust alone, and it cannot give you real footage of the past. So facts come from sources you can cite, and pictures come from generators you openly call reconstruction.

Step 1: Research and write the script

Start from sources, not prompts: three to five references per episode (books, archive pages, papers, period newspapers), noting which claim comes from which. A chat model helps with structure and with turning notes into narration that reads well aloud. In AI Studio’s Chat mode a message costs 1 credit on Gemini 3.6 Flash and up to 5 on Claude Fable 5. Do not let it invent dates, quotes or numbers. Check every fact it touches against your notes.

Narration runs about 150 words a minute, so 3 minutes is roughly 450 words. Mark a shot change every five to seven seconds; the shot count becomes your budget. Our b-roll guide covers turning a script into a shot list.

Step 2: Record the narration before any pictures

The voice track is the clock every shot has to fit, so make it first in Text to Speech. ElevenLabs Multilingual V2 and Turbo 2.5 cost 5 credits per 1,000 characters. Gemini 3.1 Flash TTS (30 voices, 70+ languages) and xAI TTS cost 6. A 450-word script is about 2,700 characters, which is 15 credits on ElevenLabs. Pick a calm voice, then cut slow sentences now, while cutting is free. The AI voiceover guide covers pacing and levels.

Step 3: Build the archival look with still images

Most history shots can be stills with a slow zoom, the cheapest way to fill the screen. In the AI image generator, write one style line and paste it into every prompt so the episode looks like one piece: for example “1920s silver gelatin photograph, soft grain, slight vignette, faded contrast” or “19th-century engraving, cross-hatched ink on cream paper”. Nano Banana 2.1 costs 3 credits at 1K and Seedream 5 Flash costs 2, so you can reject weak frames freely.

Prompt for places, objects and atmosphere: a harbour at dawn, a ledger on a desk, a crowd seen from behind. Avoid realistic “photographs” of named historical people doing specific things; that reads as a document, and it is not one.

Step 4: Restore and animate real archive photos

When you have real photographs (public-domain archive images, or family photos you have permission to use), repair them in Restore Old Photos. It fixes scratches, fading and blur, stays black and white unless you colorize, and costs 5 credits per image. Colorizing is interpretation; say so on screen.

To give a photo gentle movement, use Animate Old Photos. You pick a preset motion rather than typing a prompt, choose Seedance 2 or Seedance 2 Fast, and choose 5, 8 or 10 seconds. A 5-second clip on Seedance 2 Fast at 480p is 24 credits, or 31 with the built-in restore step turned on. The tool asks you to confirm you have the right to use the photo before it runs. Use movement sparingly; a slight head turn works once and feels uncanny on every photo.

Step 5: Generate b-roll

For shots that need real motion (waves on the hull, smoke over a factory, a herd crossing a plain), animate an approved still in Image to Video instead of starting from text. WAN 2.6 Flash costs 20 credits for 5 seconds at 720p (30 at 1080p) and is image-to-video only. When no still exists, Text to Video with Seedance 2 Fast is 24 credits for 5 seconds at 480p and 50 at 720p. Keep one subject, one action and one slow camera move per prompt. More on that family in where to use Seedance 2.

Step 6: Maps, timelines and infographics

The AI Infographic Maker turns notes into an infographic, timeline, flowchart, comparison or slide, with your title printed exactly. It costs 5 credits on GPT Image 2.5, or 1 credit on Muse, which looks facts up on the web while it draws. For an old-style map, prompt the image generator for “antique parchment map of the North Atlantic, hand-lettered labels, sepia ink”. Treat every AI map and chart as a draft: check coastlines, borders, dates and spellings against a real source before it goes in the edit.

Step 7: Music and sound effects

A low score does most of the emotional work. The AI Music Generator makes a full piece for 12 credits on MiniMax Music 2.6 (instrumental is an option) or 6 on ACE-Step. The AI Sound Effect Generator adds room tone, wind, crowd murmur or a ship’s bell for about 0.8 credits per second, so 5 seconds is 4 credits.

Step 8: Edit, caption and render

Assemble everything in the Timeline Editor. Stills can hold for up to 30 seconds with a slow Ken Burns zoom in or out. There are two audio tracks (narration on one, music and effects on the other), up to 50 text items for dates, place names and source credits, and optional auto-captions burned in at render. The render fee is a flat 10 credits, and captions add the per-minute AI Captions rate: 2 credits a minute, minimum 4. Be aware of the limits: one video track, no colour grading, a 30-clip and 180-second cap per project, and a 16:9 output of 1280×720. That fits a 3-minute episode; longer films need sections joined in a desktop editor. Full walkthrough: stitching AI clips in the Timeline Editor.

Worked example: a 3-minute episode budget

Example: “The night the lighthouse went dark”, 450 words and 30 timeline shots: 12 b-roll clips (60 s), 2 revived photos (10 s), 13 zooming stills at about 7 s (91 s) and 3 graphics (18 s), 179 seconds in all.

Step	Tool and setting	Quantity	Credits
Script help	Chat, Gemini 3.6 Flash	~10 messages	10
Narration	ElevenLabs Multilingual V2, 2,700 characters	2 takes	30
Period stills	Nano Banana 2.1, 1K	20 generated, 12 kept	60
Archive photo repair	Restore Old Photos	3 photos	15
Revived photos	Animate Old Photos, Seedance 2 Fast 480p 5 s, restore off	2 clips	48
B-roll	WAN 2.6 Flash, 720p 5 s	12 clips	240
Maps and timeline	Infographic Maker, GPT Image 2.5	3 graphics	15
Score	MiniMax Music 2.6	2 tries	24
Sound effects	Mirelo, 5 s each	6 effects	24
Render and captions	Timeline Editor, 3 min with auto-captions	1 render	16
Total			482 (about $24.10)

B-roll is half the bill. Swap four animated clips for stills with a zoom and you save 80 credits. Re-render the b-roll at 1080p and it rises by 120. The faceless YouTube cost-per-minute guide works through the same trade-offs for faster-paced channels.

Ethics: what to label and what never to make

• Label reconstructions on screen. A small “AI reconstruction” tag on generated shots, plus a line in the description, keeps viewers from mistaking them for archive material.
• Do not fake real events or real people. No realistic footage of a real disaster, battle or crime as if it were filmed, and no real person shown saying or doing what they did not. Use reconstruction for atmosphere, not as evidence.
• Use the platform disclosure. YouTube asks creators to disclose realistic altered or synthetic content, including generated scenes that never happened and extra AI footage of real places. Script help, captions and restoration do not need it. Answer the “AI use” question at upload honestly.
• Cite sources. List references in the description and credit archive photos to their holding institution.
• Get consent for private photos. Animating a family member or a private person needs permission from them or their family.

Verdict by use case

• History explainers: stills with Ken Burns, restored archive photos, labelled maps; animate sparingly.
• Nature and science: more image-to-video b-roll; WAN 2.6 Flash keeps clips cheap.
• Family history: Restore and Animate Old Photos, with consent and labels.
• Long-form: generate assets here, join 3-minute sections in a desktop editor.

FAQ

Can AI write a documentary script for me?

It can structure and tighten one, but it states wrong facts confidently. Check every claim against your sources.

How much does a 10-minute AI documentary cost?

At the mix above it is roughly 1,600 credits if every second is new. Reusing shots and holding more stills lowers it.

Can I animate a photo of a real historical figure?

Animating a real archive photo with subtle motion is common, as long as it is labelled. Generating new footage of that person doing or saying something is the line not to cross.

Do I need a subscription?

No. Credit packs are one-time, from $10.39 for 200 credits, and they never expire. That suits a channel that publishes when the research is done.