Introducing VOID Video Inpainting on WaveSpeedAI
VOID Video Inpainting — remove objects from video with mask-guided AI inpainting. Quad-mask or auto SAM-3 masks, optional Pass 2 refinement for temporal consistency. Now live on WaveSpeedAI.
Clean Object Removal in Video — Finally a Working One
Removing an unwanted object from a video has historically been the hard problem in post — the frame-by-frame rotoscope, the clone-stamp dance, the temporal flicker. Most AI attempts have produced shimmering, unstable results that only look good in single frames. VOID Video Inpainting takes a different approach: mask-guided inpainting with optional Pass 2 refinement that locks temporal consistency. We’re excited to announce that VOID Video Inpainting is now live on WaveSpeedAI.
What Is VOID Video Inpainting?
VOID Video Inpainting is a mask-guided video object-removal model. You provide:
- An input video.
- A mask — either supplied manually (a quad mask) or auto-generated via SAM-3.
VOID produces a clean version of the video with the masked region inpainted, filled by context-aware content that matches the surrounding scene across time, not just per frame.
Key Features
Quad-Mask or Auto-SAM-3 Masks Supply your own mask, or let the built-in SAM-3 integration generate one automatically from a bounding region or click prompt.
Optional Pass 2 Refinement
Enable enable_pass2_refinement to run a second refinement pass that sharpens temporal consistency — dramatically reducing flicker on difficult shots.
Adjustable Denoising and Guidance
Tune denoising_steps, guidance_scale, and temporal_window_size for the quality/cost tradeoff your shot needs.
Temporal Window Control Set the number of frames the model reasons over at once — larger windows preserve motion coherence on fast-moving content.
Production REST API Not a research demo — a hardened endpoint ready to drop into post-production and editing pipelines.
Real-World Use Cases
On-Set Mistakes — Unwanted Gear, Crew, Signage
Remove that boom mic, that reflection of the director, that passing truck — without a rotoscope artist.
Social / UGC Cleanup
Creators can wipe unwanted people, logos, or background clutter from phone-shot video.
E-Commerce Video Cleanup
Remove tags, fingers, or reflections from product videos on a per-SKU basis.
Archival and Documentary Restoration
Clean up archival footage — wires, trash, damage — while preserving the original aesthetic.
VFX Plate Preparation
Pre-clean plates before CG element insertion. The difference between a clean plate and a messy one is the difference between a 2-hour comp and a 2-day one.
Privacy and Compliance
Remove identifying marks, faces, or license plates from training data, documentation video, or publication material.
Getting Started on WaveSpeedAI
- Upload your source video.
- Provide a mask — quad-mask, pre-made mask video, or let SAM-3 generate one.
- Tune refinement settings — Pass 2 for best quality, skip for speed.
- Submit — production REST API, no cold starts.
Full schema on the model page.
Pricing
- Base rate: $0.05 per second of source video.
- Pass 2 refinement: 2× base (highly recommended for publishable work).
- Supplied mask video: +$0.05 per second.
A 10-second clip with Pass 2 enabled and auto-generated mask runs $1.00. Add a supplied mask video and it’s $1.50.
Why Run VOID Video Inpainting on WaveSpeedAI
- One API across the video stack. Chain VOID inpainting with generation, upscaling, and editing models through one endpoint.
- No cold starts. Critical for interactive post-production tools.
- Transparent per-second + add-on pricing. Predictable billing for studios.
- Production-scale throughput. Fan out batch jobs across a full shoot’s worth of plates.
Pro Tips
- Always try Pass 2 on hero shots. The temporal-consistency win is worth 2× the cost.
- SAM-3 masks are great starting points. For tricky edges, review and manually refine before submitting.
- Smaller masks inpaint cleaner. Generous masks give the model less context to lean on.
- Use a larger temporal window for fast motion. Action shots benefit from the extended horizon; static shots don’t need it.
- Run low-res proofs first. Dial in mask and settings on a cheap pass; kick the final at full res + Pass 2.
Start Creating Today
VOID Video Inpainting is the cleanest object-removal pipeline we’ve seen packaged as a single API call — with real temporal consistency, not just per-frame magic.
Try VOID Video Inpainting now on WaveSpeedAI and remove unwanted content from video in one call.
