Can AI add b-roll to a video automatically?

Yes. Upload the video to Valmera and type "add b-roll where it helps", or name the moments. The agent picks the moments that need a cutaway, researches real footage for all of them as one sequence, checks frames of each downloaded clip, and places the shots on the exact words. It renders a preview and tells you what it used.

Does Valmera generate b-roll with AI?

No. There is no text-to-video and no AI image generation. Valmera's AI researches, picks and places real footage: stock video and photos, real footage or photos of the people, products and places you name, recordings of web pages, and your own clips. When nothing real fits a moment, it keeps the shot on you.

Where does the stock footage come from, and can I use it?

Stock video and photos come from Pexels and Pixabay under each library's own license. Photos of real people, companies and places come from Wikimedia Commons and Flickr through Openverse, each with a license line. Footage fetched from the web is downloaded, not licensed, so clear the rights to anything you didn't shoot.

Does adding b-roll make my video longer?

Not with a cutaway, which switches the picture while your audio keeps playing. A splice inserts the clip as its own scene, which pauses the talk and adds the clip's length. If the video came back longer than expected, say "make that a cutaway and keep my sentence running."

Will my captions show over the b-roll?

Yes. Cutaways render above the footage and below the captions. A spliced clip has no captions of its own, because captions come from your main footage's transcript.

Does b-roll work in a vertical 9:16 video?

Yes. Stock searches follow your output frame, so a 9:16 edit gets vertical footage. A landscape clip spliced into a vertical edit keeps its whole picture on a blurred extension unless you ask to fill the frame or use black bars.

AI B-Roll Generator for Talking-Head Videos

Valmera’s AI b-roll generator doesn’t synthesize footage. It researches, picks and places real footage: stock video and photos from Pexels, Pixabay and Openverse, real footage or photos of the people, products and places you name, recordings of web pages, and your own clips. Upload a talking-head video, type “add b-roll where it helps,” and the agent lays each shot over the picture on the exact words while your voice keeps playing.

The agent checks frames of every downloaded clip before placing it and tells you what it used. Captions stay on top, the runtime doesn’t change, and any cutaway can be swapped or removed with one sentence.

Add B-Roll by Describing It

Upload a talking-head video, ask for b-roll, review the preview. Creating an account and uploading are free. Editing requires a subscription.

Create an account →

See pricing →

How to Add B-Roll to a Video Automatically in Valmera

1
Upload your talking-head video
MP4, MOV, M4V, MKV or WebM, up to 100 GB and 3 hours. Valmera indexes it once: a word-level transcript, shot changes, silences and frames.
2
Add your own clips if you have them
Upload clips and photos (PNG, JPG or WebP up to 50 MB), or paste a link. Your own footage matches your lighting and camera, so the agent uses it before stock.
3
Ask for b-roll
Type “add b-roll where it helps,” or name the moments: “when I mention the Starship launch, show it.” Say what must stay on your face: “not during the punchline at 3:40.”
4
Let the agent research and place it
It maps each cutaway to a moment in the current edit, researches the set as one sequence, checks frames of each downloaded clip and places it on the words, then renders a preview and lists what it used.
5
Review and export
Swap, shorten or drop any shot in chat (“cut back on my next sentence”). When it reads right, export a full-resolution MP4 rendered from your original file.

PROMPTS TO PASTE

Add b-roll where it helps the story, but keep me on camera for the punchlines.

When I mention the Starship launch, show real footage of it for four seconds.

Cover the jump cuts in the first two minutes with the workshop clips I uploaded.

Record a scroll of https://example.com and show it while I talk about the website.

Where the Footage Comes From

Your own material goes first, because it matches your lighting, color and lens. When you don’t have the shot, the agent goes looking, and if nothing real fits, it says so and asks for a clip.

Source	Best for	How Valmera gets it
Your own clips and photos	Anything you filmed: the product, the workshop, the team.	Upload them, or paste a link (Dropbox, Drive, YouTube, Vimeo, TikTok).
Stock video and photos	Generic visuals: a busy street, ocean waves, hands on a keyboard.	Pexels and Pixabay, searched together. Orientation follows your output frame.
Photos of named subjects	A person, company, product or place the speaker names.	Wikimedia Commons and Flickr through Openverse, each photo with its license line.
Footage of named topics	An event or launch the speaker mentions.	find_footage searches the web for real video; the agent names the source.
Recorded web pages	Your landing page, docs or a product page.	A headless browser scrolls the public page or clicks through steps you describe.

How the Agent Chooses the Moments

It starts from the transcript of the current edit, so every time is where the words land after your cuts. For each cutaway it writes a moment: a concrete visual query, why the viewer needs it (proof, context, contrast, a reset, a payoff), a time and a duration. research_broll searches every moment at once and lays the candidates out on one board, so the shots are chosen as a sequence that holds together in color and look. For an important line it tries several routes to the idea: the exact subject, an action that proves the point, a telling detail.

A weak moment can stay on you: when no candidate shows what you’re saying, Valmera keeps the shot on the speaker instead of filling the gap with wallpaper. Nothing enters the project until a pick is downloaded, and the agent checks frames of the actual clip, not just its thumbnail, before placing it.

Cutaway or Splice

A cutaway (add_overlay with fit="cover") switches the picture for a few seconds while your voice, the music and the captions keep running, so the runtime doesn’t change. A splice (insert_media) adds the clip as its own scene: the talk pauses, the video gets longer, and a mid-take splice splits cleanly at a word edge. The agent picks by asking whether the sentence should keep going; one sentence from you switches it.

Settings That Matter

Setting	Default	How to change it
Placement	Cutaway: the picture changes, your voice keeps playing	“Splice it in” pauses the talk for the clip.
Length	2–6 s per cutaway; an image defaults to 4 s	“Hold it for three seconds.” “Cut back on my next sentence.”
Sources	Your uploads first, then stock or real footage	“Only use my clips.” “Find real footage of the launch.”
Orientation	Matches the output frame, so 9:16 gets vertical stock	“Use landscape clips and pad them.”
Fit of spliced clips	Whole picture kept on a blurred extension	“Fill the frame” crops; “black bars” pads.
Sound	Cutaways are silent; a spliced clip plays its own audio (stock clips have none)	“Mute that clip.”
Motion on spliced media	None	“Slow zoom in on that photo”: zoom in or out, pan left or right.

Licensing and Credit

The agent always tells you which shots it used and where each came from, and never presents stock or found footage as yours. Pexels and Pixabay clips come under each library’s own license. Openverse photos carry a license per photo (public domain, credit required or non-commercial only), which the agent passes on when it carries an obligation; no-derivatives licenses are excluded.

Footage found on the web or fetched from a link is different: downloading verifies neither ownership nor license, and a downloadable video is not permission to republish it. Clear the rights to anything you didn’t shoot before you publish.

Tips for Talking-Head Videos

Cover the jump cuts that snag. Cutting pauses from one camera leaves a jump at every join. Cover the mid-sentence ones, punch in on others and leave the rest; the jump cut editor explains the triage.
Show what’s being said, on the word. Name the word in your prompt, not a timecode; timecodes move when you cut again.
Keep the face for the payoff. Never cover a punchline, a reveal or a line delivered straight to camera.
End on a beat in the narration, not when the clip runs out.
Ask for things a camera can see. “City traffic at night” beats “the pace of modern life.”
Don’t overdo it. One to three purposeful cutaways a minute reads as edited; wall-to-wall b-roll pushes the speaker out of their own video.

More on pacing and structure in how to edit talking-head videos, and on the craft in the b-roll glossary entry.

Limits

Valmera doesn’t generate b-roll: there’s no text-to-video and no AI image generation, so a shot that exists nowhere has to be filmed, screen-recorded or replaced with a photo. Overlays don’t track objects; a picture-in-picture holds its position, or a path you set, and won’t follow a moving subject. Spliced clips aren’t captioned (captions come from your main footage’s transcript), and a cutaway’s own audio doesn’t play.

Part of the Same Edit

To show your own product, record a website demo from its URL and cut to the capture. Overlay sizes, positions and fades are in the text and overlays docs, and the YouTubers page shows the whole talking-head workflow.

Using Claude, ChatGPT or Cursor? The same research_broll, search_stock, find_footage, add_overlay and insert_media tools are available through Valmera's MCP server.