Test record
Cut the subject out of a photo in one tap.
🔬 Tech tested · app test pending
Date14 Oct 2026
Tool and tierOpen-source stand-in only: rembg 2.0.85 (onnxruntime 1.31.0) on a CPU. We have NOT tested the iPhone Lift-subject feature or any web/phone app.
Test dataFictional or public-domain only
Exact prompt
(none: tap-by-tap, no prompt)
Sample input
- Input photo: “Cat November 2010-1a” by Alvesgaspar, Wikimedia Commons, CC BY-SA 3.0 · download input.jpg
Raw output
- Output on a flat yellow background (our modification) · download output_on_yellow.jpg
What went wrong, and limits
Whiskers and fine fur edges were lost or softened. First run took 59.7 s because it downloaded a ~1 GB model; later runs would be faster. The result is from open-source software, not from the apps we recommend.
Runnable test code
The script we ran (content/remove-photo-background/test_code/remove_bg.py)
# Test code actually run on 2026-10-14 (box CPU, rembg 2.0.85, onnxruntime 1.31.0, Pillow 12.3.0).
# Input: 'Cat November 2010-1a' by Alvesgaspar, Wikimedia Commons, CC BY-SA 3.0 (960 px wide rendition) saved as input.jpg
import time
from rembg import remove
from PIL import Image
s=time.time(); o=remove(Image.open('input.jpg')); o.save('output.png')
print('secs',round(time.time()-s,1),o.size,o.mode) # observed: 59.7 s incl. ~1 GB model download, (960, 1282) RGBA
bg=Image.new('RGBA',o.size,(255,214,102,255)); bg.alpha_composite(o); bg.convert('RGB').save('output_on_yellow.jpg') # flat yellow background (our modification)
Full working log
# Evidence: remove-photo-background Status: TECHNICAL ONLY (open-source stand-in). Consumer tools pending. - Date: 2026-10-14, box CPU - Tool: rembg 2.0.85 (onnxruntime 1.31.0), default model; first run downloaded ~1.0 GB model - Input: input.jpg = 'Cat November 2010-1a' by Alvesgaspar, Wikimedia Commons, CC BY-SA 3.0 (960x1282 rendition) - Output (unedited): output.png (RGBA cut-out); output_on_yellow.jpg (composited on flat colour, a modification); before_after.jpg - Runtime: 59.7 s incl. model download - Result: clean subject mask; ears, paws, tail edge good. Whiskers lost/softened; ledge removed correctly. - Licensing: derivative must be credited and shared CC BY-SA 3.0 (done on site /about.html and caption). - Needs human test: same photo through remove.bg free + iPhone/Android built-in tools.