Ideogram launched Object Remover this week — a dedicated tool for erasing people, text, signage, cables, watermarks, and other clutter from a photo while reconstructing what was behind them, rather than leaving a smeared patch or an obviously synthetic fill.
How it actually works
The workflow is a three-step loop: upload a photo, brush over the object to mask it, generate a cleaned result. What separates a good object remover from a mediocre one is what happens in that third step — the system has to infer lighting, texture, shadows, grain, and perspective from the surrounding pixels well enough that the fix doesn’t read as obviously edited. Ideogram’s own materials claim the tool preserves exactly those details: shadows, reflections, and subtle lighting shifts that generic erasers tend to flatten out.
The numbers Ideogram is leading with
Ideogram is positioning Object Remover directly against the field on RemovalBench, a benchmark for exactly this task, claiming the top score against FLUX Erase (Black Forest Labs), GPT Image-2, and Nano Banana 2 — real competitors, not strawmen. It’s also leading on price: roughly $0.03 per request via the API (/v1/remove-object endpoint), which Ideogram describes as the lowest among tested alternatives, with average processing around 9.8 seconds.
Object removal has quietly become one of the more commercially useful corners of generative image AI — it’s the unglamorous work behind cleaning up product photography, removing people from real-estate listings, stripping watermarks from stock photos, and general photo cleanup that doesn’t need a full generative rebuild of the image. A dedicated, benchmark-leading, cheap tool for that specific job is a more concrete competitive move than another general-purpose image generator entering an already crowded field.
Why this is a distinct move from “add editing to the base model”
Most major image-generation platforms bundle some version of object removal as one feature among many inside a broader editing suite. Ideogram shipping it as a standalone, benchmarked, separately-priced tool — with its own landing page and its own API endpoint — signals a bet that treating a narrow, extremely common editing task as its own product, tuned and measured specifically for that task, beats folding it into a general “edit anything” interface. Whether competitors respond by sharpening their own object-removal quality specifically, rather than treating it as a minor feature of a bigger model, is the thing worth watching next.