Add an Object to a Video with AI (Real Test)
To add an object to a video with AI, use an AI video editor that takes the clip you already have and draws it again with the new thing in it. On November 7, 2026, we tested Riffkit's Swap on an 11-second clip of a woman at a desk and asked, in words only, for two additions: sunglasses on her face and a small potted plant beside her keyboard. Both were there at every half-second check, from her first step into the frame to the last frame, and her face, moves, caption and sound held. The sunglasses came out with a thin metal frame rather than the black one we asked for. It cost 1,440 credits on Seedance 2.0 at 720p and took about 7 minutes. Disclosure: we make Riffkit.
Which AI tools add an object to a video you already have?
Most of them take a short clip and a written instruction. They differ in how long a clip they take and in what happens to the sound. Per each tool's own page, checked November 7, 2026:
| Tool | How you ask for the new object | Clip length it takes | Sound |
|---|---|---|---|
| Runway Edit Studio (Aleph 2.0) | Words (plus an optional reference image) on one frame, which it carries through the clip | Up to 30 s, 1080p | Not stated |
| Kling 4.0 video editing | Words plus reference images; its guide lists adding an element | Up to 5 clips, 3 to 30 s each, 30 s in total | Its guide says music and sound effects are generated from the visuals; its edit examples ask it to keep the original audio |
| Google Flow, video to video (Omni Flash 1.2) | Words, from the edit tool on a clip | Up to 10 s per edit, or up to 15 s with a Google AI Plus, Pro or Ultra plan | Not stated |
| Riffkit Swap | Words in "What to change", plus photos named with @ | Up to 45 s, 100 MB (we tested 11 s) | The clip's own track, when only what's on screen changes |
Pick by your clip. A clip of a few seconds fits every tool here. Past 10 seconds (15 with a Google AI plan), Google Flow edits it one segment at a time, and past 30 seconds, Swap is the only tool in this table that takes it. If the sound matters, as with a voice-over, music or a punchline, use a tool that keeps your track, or plan to lay the original audio back on. Pika's add tool, Pikadditions, was available through fal's API until fal removed it on October 29, 2026; Pika's own home and pricing pages don't list it, so check inside Pika before you plan on it.
What others found when adding objects
Others have tested adds too. The Rundown's guide (updated July 23, 2026) added a spaceship to a parking-garage clip in Higgsfield; its first prompt changed too much of the scene, and weaker results struggled with depth. The AI Filmmaker tested Google Flow's add tool in October 2025: a beer landed where it was asked to go, and a hat didn't fit the head. Segmind's own guide to Seedance 2.5 Edit (November 2, 2026) added a toy train and left the rest of the frame close to untouched, but the train only rolled in about 4 seconds into a 5-second clip; on the guide's other edits, the sound came back resynthesized rather than as the original file.
The test: sunglasses on her face, a plant on the desk
The source is the same 11-second vertical clip we used in our object replacement test, made with AI and posted on our TikTok in September: one shot, no cuts. A woman in a headset hurries in with a laundry basket, types on a black keyboard, fake-laughs at a video meeting with her face close to the camera, then backs away. The only sound from her is the laugh, and the caption reads "my only contribution / to team meetings".
We picked two additions that fail in different ways. Sunglasses sit on a face that moves, turns and comes close to the lens. The plant sits still, while her hand and the basket pass right by it.
Settings: Swap, Seedance 2.0, 720p, no photo, no character. We typed this into "What to change", word for word:
Add two things and change nothing else. First: she wears a pair of black sunglasses with a thin frame for the whole clip, resting on her nose under the headset and staying on her face as she leans in close and walks back. Second: a small green potted plant in a white pot sits on the desk beside the keyboard, in the same spot and at the same size from the first frame to the last. Keep everything else exactly as it is: the same woman, her face, expressions, hair, headset and microphone, top, scrunchie, the laundry basket and the clothes in it, the keyboard, the desk, the room, the window blinds, the lamp, the aquarium, the camera angle, the framing, the timing, the sound, and the caption text and its position. Add nothing else: no extra people, no other objects, no new text.
What it cost: 1,440 credits, the same as the quote: the clip runs 11.04 seconds, which bills as 12 seconds at 120 credits a second, the Seedance 2.0 Swap rate at 720p.
What held, checked every half second
We wrote down what counted as a pass before the render came back, then checked 23 stills from the result, one every half second, plus six moments side by side with the original and close-ups of her face.
| What we checked | Result |
|---|---|
| The sunglasses | On her face in every still she's in, 22 of 23 (the first, at 0 seconds, is before she walks in): from her entrance at half a second, through the close-up from 4 to 5.5 seconds, to the back of the room in the last frame. The arms run under the headset's ear cups |
| The plant | A small succulent in a white pot, beside the right end of the keyboard in all 23 stills, in the same spot and at the same size relative to the keyboard as the camera moves |
| Things passing it | Around 2 seconds, her hand on the basket handle is behind the plant, and the plant hides it; the basket passes behind the plant without going through it |
| The person | Unchanged apart from the sunglasses: expressions and their timing, hair and claw clip, headset, top, two necklaces, the ring, the scrunchie |
| The room | Unchanged: the basket and clothes, the black backlit keyboard, the desk, the blinds and sunlight, the lamp, the armchair, the aquarium |
| Camera and timing | Unchanged, and no cuts added (cut-detection software found none) |
| Caption | "my only contribution / to team meetings" at every check, same size and spot |
| Sound | The original track, lined up with the source with no offset |
What drifted
Two details came out differently from the brief:
- The frame. We asked for black sunglasses with a thin frame. They came out as dark-lensed aviators with a thin silver-gray metal frame.
- The plant. We asked for a small green plant without naming one, and got a succulent. Name the plant if it matters.
We found nothing extra in the frame. That's one run on an AI-made clip, so read it as one data point, not a rule. Our earlier tests add two more points. In two runs of our outfit test, sunglasses were part of a streetwear photo and we wrote nothing, on Seedance 2.0 Fast at 480p: the clothes went on both times, the sunglasses didn't. Here, described in words, they stayed on for the whole clip. And in our Christmas tests, added lights, a tree and a garland held, while presents we asked to replace boxes appeared beside the boxes instead.
How to write the addition
Say what it is, where it goes, and that it stays. For something worn, say how it sits ("resting on her nose under the headset") and that it stays on through the moves ("as she leans in close and walks back"). For something placed, name what it sits next to and add "in the same spot and at the same size from the first frame to the last". Then list what must stay, and end with what to leave out. A pattern you can reuse:
Add [the object, with color and size] [where it goes], [how it sits or how the person wears it], staying [in the same spot / on her face] for the whole clip. Keep everything else exactly as it is: [the person, clothes, room, camera, timing, sound, caption]. Add nothing else: no extra people, no other objects, no new text.
For an exact look, such as your own product with its label, add a photo and name it with @. More on wording in write what changes, not a prompt.
Adding your product
To put your product in someone's hand, replace what they already hold: Swap keeps the person's moves, so your product takes the place of the thing in their hand. Our lipstick product swap shows that route with a product photo.
All our tests used clips made with AI. We haven't tested phone footage of a real person yet, and on a Seedance engine the vendor's content review checks each clip first and can turn one down.
If your clip has several shots
Our test clip is one continuous shot. In our Christmas room test, added lights and a tree held through two jump cuts in the same room, but when one render spans shots in different places, the change can land mostly in the opening shot: when we turned a 12-shot clip from day to night, Seedance 2.0 turned only the first shot. If one shot matters most, cut it into its own clip and run it on its own, then check every shot.
How to add an object to your video with Swap
- Add your clip in Swap, up to 45 seconds and 100 MB.
- Describe the addition in "What to change": what it is, its color and size, where it goes, and that it stays for the whole clip.
- List what must stay: the person, the room, the camera, the timing, the sound and any caption.
- End with what to leave out: "Add nothing else: no extra people, no other objects, no new text."
- Add a photo for an exact look. Save it under References (or Products), pick it in the Swap form under "Reference images" (or as the product), then type @ in "What to change" to name it.
- Check the price, then select Riff it →. Before you post, check the new object in every second of the clip.
You can also run Swap from ChatGPT, where Riffkit is in the plugin directory (add it), or from Claude with the Riffkit connector; how to make videos in ChatGPT and Claude has the setup for both.
What it costs, and how to start free
Swap bills per second of finished video, rounded up to whole seconds: on Seedance 2.0 that's 60 credits a second at 480p and 120 at 720p. A run that fails before any video comes out costs nothing. Plans start at $15 a month for 3,000 credits; at that price, our 1,440-credit test comes to about $7.20.
You can sign up free, with no card, and new accounts usually start with free credits. A free account can Swap on Seedance 2.0 Fast or Seedance 2.0 at 480p, a lower resolution than the 720p we tested. A clip under 8 seconds comes back whole; a longer one, like this 11-second clip, usually comes back as its first section, and a plan finishes the rest. Videos on a free account carry a small Riffkit watermark; any plan removes it from every video, earlier ones included.
Rights
Add objects to footage you own or have the rights to. Putting sunglasses on someone doesn't change who holds the rights to their likeness, their voice or the footage. A video that won on someone else's account still belongs to them: to riff its formula into your own story, use Adapt, which makes new footage (where the line sits).
Open Swap and try an addition on a clip of your own.
FAQ
How do I add an object to a video with AI?
Use an AI video editor that redraws the clip you already have with the new object in it. In Riffkit's Swap mode, add the clip (up to 45 seconds), describe the object, where it goes and that it stays for the whole clip in "What to change", list what must stay, and end with "add nothing else". In Riffkit's November 7, 2026 test on an 11-second AI-made clip, a small potted plant beside her keyboard was there at every half-second check and sunglasses were on the woman's face in every check from the moment she walked in, and her face, moves, caption and sound held, on Seedance 2.0 at 720p for 1,440 credits.
Can I add an object to a video with AI for free?
Yes, to try it. You can sign up for Riffkit free with no card, and new accounts usually start with free credits. A free account can use Swap on Seedance 2.0 Fast or Seedance 2.0 at 480p. A clip under 8 seconds comes back whole; a longer one usually comes back as its first section, and a plan finishes the rest. Videos made on a free account carry a small Riffkit watermark.
Can AI add sunglasses to someone in a video?
Yes. In Riffkit's November 7, 2026 test on Seedance 2.0, a written brief put sunglasses on a woman in an 11-second AI-made clip, and they were on her face at every half-second check from the moment she walked in, including when she leaned close to the camera and when she walked to the back of the room, with the arms running under her headset. The frame came out thin silver-gray metal rather than the black we asked for. Phone footage of a real person isn't tested yet.
Does adding an object with AI change the rest of the video?
It can, with tools that redraw the whole clip, such as Riffkit's Swap, so list everything that must stay and check the result. In Riffkit's November 7, 2026 test, the woman's face and expressions, her headset and clothes, the room, the camera, the timing, the caption and the sound all held, and nothing extra appeared. In earlier Christmas tests, small extras nobody asked for, such as a pocket and a necklace, did get in.
Do I need a photo of the object to add it to a video?
Not for a generic object. In Riffkit's November 7, 2026 test, words alone added sunglasses and a potted plant; the details came out only roughly as described (a metal frame instead of black, a succulent for "a small green plant"). For an exact look, such as your own product with its label, save a photo of it in Riffkit, pick it in the Swap form and name it with @ in "What to change". In two earlier Riffkit runs on Seedance 2.0 Fast at 480p, where sunglasses were part of a streetwear photo and nothing was written, the clothes went on both times and the sunglasses didn't, so describe the addition in words as well.
Will the video keep its original sound when AI adds an object?
In Riffkit's Swap mode, yes: when only what's on screen changes, Swap keeps the clip's own track. In Riffkit's November 7, 2026 test, the result's audio was the original track, lined up with the source with no offset. With other tools, check: Kling 4.0's guide says music and sound effects are generated from the visuals and doesn't say whether an edit keeps the clip's own track, and Segmind's Seedance 2.5 Edit guide found the sound came back resynthesized rather than as the original file.
Can I add an object to a video from ChatGPT or Claude?
Yes, with Riffkit, which runs its Swap mode from inside ChatGPT or Claude: in ChatGPT, Riffkit is in the plugin directory; in Claude, you add the connector, with setup at riffkit.ai/mcp. Ask the assistant for an upload link, add your clip on the Riffkit page it opens, and say what to add and what to keep; the assistant shows the price before anything is spent. Setup for each assistant is on riffkit.ai/mcp.
Keep reading
Make a Christmas AI Video from a Clip You Already Have
Make a Christmas AI video from a clip you already have: four Seedance 2.0 tests with lights, a tree, pajamas and snow, the exact prompts, and what slipped.
Change the Person in a Video with AI (Real Test)
Change the person in a video with AI: eight tools compared, and our 11-second test where words changed her hair but not her face. Why a photo helps.
Replace an Object in a Video with AI (Real Test)
Replace an object in a video with AI and keep the person, room, camera and sound. Our keyboard test, checked every half second, plus the tools and the cost.