ObjectRemover

One of the core technical innovations in ObjectRemover is Intelligent Engine Routing — an adaptive decision layer that automatically chooses the optimal inpainting backend for each video removal task.Instead of applying a single model to every clip, the system analyzes scene motion characteristics and routes between ProPainter (propagation-based) and MiniMax-Remover (diffusion-based) depending on the footage.

Why Routing Matters

Video inpainting must satisfy two conflicting requirements simultaneously:

• Per-frame visual quality

• Temporal consistency across the entire sequence

A fixed pipeline often fails here. A large static object against a motionless background needs strong generative capability, while a small object in a dynamic handheld shot demands excellent optical-flow propagation to avoid flicker and ghosting.

How the Router Works

Before processing starts, ObjectRemover builds a lightweight scene motion profile from downsampled frames and the SAM3-generated mask. The key signals extracted are:

1. Background motion (outside the mask) — frame-to-frame change in surrounding areas

2. Internal mask motion — relative movement of the object itself

3. Edge contrast — how sharply the masked region separates from neighboring pixels

Implementation Notes

• Full adaptive routing is active for Standard and Pro quality video removal jobs.

• Low quality jobs default to ProPainter (with tuned large-mask handling).

• Image removal uses a dedicated MiniMax path (no video motion router).

• The entire decision process is fully automatic — users do not select or even see the backend.

Real-World Impact

The routing system noticeably improves results on common creator footage:

• Reduced temporal streaking on follow shots and handheld clips

• More convincing fills for large static objects (people, pets, furniture)

• Better handling of mixed scenes that previously challenged single-model pipelines

We continue to refine motion thresholds and routing heuristics as underlying models evolve.For the complete explanation, including more details on the motion profiling pipeline and FAQ, read the full post:

https://objectremover.video/blog/intelligent-engine-routing-video-object-removal

Hey IH community,

A few weeks ago I shared how ObjectRemover can handle up to 50-second videos with solid temporal consistency. Today I want to talk about something equally frustrating — photo object removal.You’ve probably seen the viral comparison (like the one from@Rainmaker1973): same photo, Samsung Galaxy S25 Ultra AI erase vs iPhone AI erase… and the results look noticeably different.That’s exactly why I expanded ObjectRemover to support still images (JPG, PNG, WebP).

The Problem with Phone AI Erasers

• Different OEMs use different inpainting models

• Results are device-locked and inconsistent

• Usually only “erase”, no clean extraction option

• You lose control over original files and quality

What ObjectRemover Now Offers for Photos

• Works in any modern browser (no phone, no install)

• Remove unwanted objects or Extract subjects with transparency

• Supports up to 4096px resolution and 20MB files

• Same natural language prompt workflow as video (“remove the person on the right”, “extract the product only”)

• One account & credit system for both video and photo

Here’s the full comparison and test I wrote:
→ https://objectremover.video/blog/photo-ai-object-removal-samsung-iphone-comparison

The goal is simple: give creators a reliable, device-independent way to clean up photos — whether it’s removing photobombers, logos, or extracting assets for compositing.Would love to hear your experience:

• Do you still rely on your phone’s built-in AI eraser?

• What’s the biggest pain point when cleaning photos for work?

Looking forward to your thoughts!