Dreamina is our first pick for a reference-led object swap: start with the video you want to adapt and images of the replacement. Runway adds a useful edited-keyframe preview before applying a change across the clip. Kling and Higgsfield also offer reference-guided editing, while Beeble gives filmmakers a mask-based way to control retained and generated regions. Pika and Luma provide creative transformation routes; After Effects suits work that needs deliberate tracking and compositing.
The right tool depends on what “replace” means in your shot. Making a bottle disappear, substituting your own bottle, and redesigning the whole scene require different controls. The tools below work with existing footage, but they do not all preserve it in the same way.
- Choose the control your object needs
- 1. Dreamina: Bring the replacement into the brief
- 2. Runway: Approve the keyframe before the video
- 3. Kling AI: Name the video and reference explicitly
- 4. Higgsfield: Choose Objects swap for a local change
- 5. Beeble SwitchX: Decide which area gets generated
- 6. Pika: Use the current Swap Anything app
- 7. Luma: Control motion and structure with Ray3.2
- 8. After Effects: Remove, track and composite when precision matters
- Build the swap around one clearly visible object
- Fix the failure that actually happened
- Questions before you replace a video object
Choose the control your object needs
For a specific product, a reference image matters more than an elaborate adjective-filled prompt. For a sign or phone screen with exact text, tracked graphics may be more dependable than generative redrawing. For a prop partly hidden behind fingers, the foreground hand needs attention as well as the replacement.
A background change has its own selection problem: hair, transparent edges and subject separation. If that is the real job, see our guide to AI video background replacement. Here, the central task is replacing an object inside the scene.
1. Dreamina: Bring the replacement into the brief
Dreamina’s Recreate workflow combines a reference video with visual references. Object Swap is the relevant mode when you want to specify a new product, prop, outfit or other element while aiming to retain the surrounding shot.
How to start. Open Dreamina Recreate, select an available model and Object Swap, then add the source video and at least one replacement image. Explain what should change and what should remain. Review the generated clip in the task center and use re-edit to refine the request.
Choose the appropriate model. Documented standard Recreate uses Seedance 2.5 and accepts one 4–30-second video; Recreate Lite uses Wan 3.0 and accepts 4–15 seconds. Availability varies by region and account. Prepare the intended segment before upload: the documented launch workflow does not provide segment selection or trimming, and an overlong input is rejected.
Why choose it. The workflow starts with both the existing motion and the replacement’s appearance. That fits a product adaptation more directly than generating a similar new shot from text. Judge the result against both inputs; a plausible new object still needs the right shape and interaction.
2. Runway: Approve the keyframe before the video
Runway’s Edit Studio gives the replacement a useful intermediate stage. You can edit a clear frame, inspect the proposed appearance, then apply the selected version to the moving shot with Aleph 2.0.
How it works. Open Edit Studio from Apps, upload the clip, and select a keyframe that clearly shows the target. Write a short change instruction and add a reference image if useful. Preview the adjusted image, refine it if necessary, then generate the video. A range control can limit the edit to part of the clip.
Why choose it. If a replacement looks wrong while stationary, propagating it through motion adds work without solving the original problem. The preview lets you correct its size, shape or placement first. Extra motion is a separate optional instruction; a normal prop swap does not necessarily need it.
Conditions. Current documentation lists conventional aspect ratios, 480p–1080p input, 24–30 fps and no more than 10 cuts. Upload guidance calls for videos longer than two seconds and shorter than 30 seconds. Video generation costs 28 credits per edited second with a 56-credit minimum; keyframe image iterations are charged separately. ProRes and PNG-sequence generation have additional plan and cost conditions.
3. Kling AI: Name the video and reference explicitly
Kling’s O1 editing workflow combines video, text and image or element references. It is useful when the replacement has a specific appearance that a generic instruction would not communicate.
How it works. Supply the existing video in the editing workflow, add the relevant replacement reference, and use the interface’s reference tags in the instruction. Describe the target and the intended change clearly. An element built from several views can provide more appearance information than one ambiguous photograph.
Why choose it. Tagged references help separate the source footage from the item you want to introduce. A prompt can identify the object’s role as well as its color: “replace the bottle held in the right hand with the referenced bottle” is clearer than “add this bottle.”
Conditions. The official O1 guide documents one 3–10-second video, up to 200 MB and 2K input, with up to four images or elements combined when a video is present. Those are O1-specific conditions. Do not confuse its video-reference feature for generating another shot with editing the original shot; the output task is different.
4. Higgsfield: Choose Objects swap for a local change
Higgsfield Genjutsu separates Objects swap from Motion transfer. That distinction matters: the first targets a chosen element, while the second uses the movement to rebuild more of the scene.
How it works. Select Genjutsu in the video model picker, choose Objects swap, and upload the clip and replacement images. The current standard workflow accepts a 4–30-second video and up to 30 reference images. A short prompt can specify the change when references or presets need more direction.
Why choose it. This suits variations around a product, prop or outfit when the original performance and shot timing already work. One clear reference may be enough for a simple substitution; several useful views can better describe a more complex item.
Trade-off. A prompt-targeted edit still needs an object-by-object review. Similar items, a prop disappearing behind someone, or a cut to another angle can make the target less clear. Split complicated sequences into shots and name the visible item in each, rather than assuming one broad request covers every appearance.
5. Beeble SwitchX: Decide which area gets generated
Beeble SwitchX uses an alpha mask to distinguish retained source regions from generated regions. It offers a different kind of control from relying on the model to infer the target solely from words.
How it works. Supply the footage, define the mask and add a prompt or reference for the desired look. Beeble’s current explanation uses white for retained source areas and black for fully generated areas. Confirm that convention before submitting; an inverted mask changes the wrong part of the frame.
Why choose it. The mask gives you a visible boundary to inspect when the object is close to a face, clothing or another important feature. A replacement brief can also specify the surrounding light and visual style.
Trade-off. Retained does not necessarily mean identical color: SwitchX can relight and restyle source areas to blend with the new scene. If untouched pixels are a delivery requirement, inspect that behavior and consider conventional compositing. Its workflow is useful precisely because you can make the generated area explicit, rather than treating preservation as a vague promise.
6. Pika: Use the current Swap Anything app
Pika’s current Swap Anything app accepts a clip, a description of what changes and optional references. It provides a direct route for creative changes to props, products, clothing or cast.
How it works. Upload the video, name the element to replace, and add references for the intended appearance. Generate the modified version, then compare its new look with the original movement. Keep unrelated change categories empty when your aim is one prop.
Why choose it. A task-oriented app is convenient when you want to start with the swap rather than select a general video model first. For a playful social clip, you may value a bold transformation more than keeping the source pixel-for-pixel.
Trade-off. Older guides may call the feature Pikaswaps and describe settings from an earlier interface. Start from the app and inspect the available model and generation settings. A claim about an older Pika plan or resolution is not automatically a condition of the current app.
7. Luma: Control motion and structure with Ray3.2
Luma’s current Ray3.2 Modify Video V2 workflow transforms existing footage while using its motion and spatial layout as guides. It is useful when a prop change belongs within a broader creative treatment.
How it works. Open Modify Video and choose Ray3.2, upload the source, and describe the replacement. Motion controls concern movement and camera dynamics; Structure controls how tightly the result follows the original shapes and composition. Begin with strong structure adherence for a local prop change, then loosen it only if the intended new shape needs more freedom.
Why choose it. This separates keeping the action from retaining the object’s old geometry. Multi-keyframe guidance can also direct how an appearance changes through the clip.
Conditions. Modify Video supports up to 20 seconds at 24 fps, 15 seconds at 30 fps, or seven seconds at 60 fps, with outputs up to 1080p. Fast, blurry or heavily occluded action is harder to track. Review the product and surrounding scene: performance preservation is not an untouched-pixel guarantee.
8. After Effects: Remove, track and composite when precision matters
After Effects is a compositing route rather than a single prompt that invents the whole replacement. It is useful when the object must follow exact motion, display exact artwork or sit behind a separately controlled foreground hand.
How it works. Mask and track the original object. If it needs to disappear, Content-Aware Fill can reconstruct the covered area using surrounding frames; difficult backgrounds may need a corrected reference frame. Add the replacement as a separate element, track its movement and perspective, then use masks or rotoscoping to restore foreground occlusions.
Why choose it. A phone screen, flat book cover or sign often needs exact text. Tracking and placing the real artwork keeps that artwork under your control. The work shifts from “generate a convincing version” to “make this specific asset sit correctly in the shot.”
Trade-off. You supply more craft: tracking, edges, motion blur, lighting and contact shadows. Content-Aware Fill alone removes an object; it does not automatically insert your product. For removal without substitution, our guide to removing unwanted objects from video covers the cleanup decision separately.
Build the swap around one clearly visible object
A useful first attempt is a short, continuous shot with a visible target and a replacement reference photographed from a compatible angle. You learn more from a controlled substitution than from a long montage containing several different objects.
Prepare the source and replacement together
Trim the source to the relevant action before uploading when the tool has no trimming control. Keep enough lead-in and ending to see the object enter, move and leave. A crop that hides the hand or object halfway through the action makes the result harder to judge.
Use a clean replacement reference that shows the silhouette and important details. A front view helps with a label; an angled view communicates thickness. For a product that turns during the shot, one front-facing photograph leaves the side and back open to invention. If you are still preparing a recognizable product asset, our guide to consistent product images for ads explains which details need to remain fixed.
Write the change and the retained action separately
A practical starting instruction is:
Replace the white mug on the table with the blue ceramic mug in the reference image. Keep the person, hand movement, table and camera movement. Keep the fingers in front of the mug when the person lifts it. Match the original light and contact shadow.
Adapt the object and action to the clip. Use the tool’s actual reference labels or tags where available. “Replace” is more useful than “add” when the original item must disappear; otherwise the request can leave both objects in the scene.
Where a keyframe preview is available, solve the stationary appearance first. Check the replacement’s size against the hand or nearby furniture, the angle of its base and its position in the scene. Then propagate that version into motion.
Review the whole action and the downloaded result
Inspect the beginning, the strongest movement, every overlap and the ending. A bottle that looks correct before a hand crosses it can return with a different label afterward. Watch for a sudden size change, a handle appearing on the wrong side, or a reflection that still shows the original object.
Play the clip at its intended speed as well as inspecting key frames. Flicker and brief disappearances can be harder to notice in isolated stills. Export the candidate and check the actual file’s duration, framing and final appearance before putting it into a longer edit. If the task is to keep only the camera move while remaking the shot, use the separate workflow for remaking a product video with its camera movement.
Fix the failure that actually happened
The old object remains, or only some appearances change. Make the target description more specific and divide the sequence at cuts. A book, a blank card and a package may look similar to a person but require separate descriptions. For a mask-based tool, verify which region is selected and check the mask when the target reappears.
A hand merges into the replacement. The task includes an occlusion: some fingers belong in front, while part of the object sits behind them. Choose a clearer frame, reduce the requested change in shape, or restore the original hand as a foreground composite. A more detailed product photograph alone does not explain every hand-object interaction.
The object changes shape or branding during movement. Use references that reveal the relevant views and simplify the shot. When exact lettering or packaging is essential, composite approved artwork or use a real shot of the replacement. A plausible generated label is not a faithful label.
The background or person changes too. Reduce the transformation scope, choose the local swap mode and simplify the prompt. If a method inherently redraws too much, switch to a mask-controlled or conventional compositing route rather than repeatedly asking for “everything identical.”
The new object fits, but the shot still feels wrong. Check its shadow, reflection, motion blur and points of contact. Replacing a reflective prop can require changes outside its silhouette. A rigid tracked graphic can work for a screen or book cover; a deforming handbag or splashing liquid needs a different solution.
Questions before you replace a video object
Is object replacement the same as video inpainting?
The terms overlap, but the intended result matters. Removal reconstructs what should be behind the old object. Replacement adds another object with its own appearance and interaction. A tool that only removes the original may need a second compositing step.
Can a replacement keep the original camera movement?
Reference-video editors use the source motion as a constraint, but inspect the generated clip rather than treating the feature description as an exact-motion guarantee. If you need the original footage outside the edit under direct control, choose an appropriate mask or compositing workflow.
What if the object is a person?
Character replacement adds facial identity, hair, clothing and performance to the comparison. Start with our guide to replacing a video character while keeping motion. A successful prop swap does not establish that a tool will preserve a person’s face or performance.
Start with Dreamina when you want a reference-led version of an existing clip, and make the first request one clear substitution. Use a keyframe preview, mask or tracked composite when that control solves the shot’s real difficulty. The useful result is the replacement that works through the entire action, not just the frame that looks good first.