How big does an image need to be before it traces well?
Around 1000 pixels on the long edge for a logo. Below about 500 the edges are already soft, and a tracer follows that softness faithfully, because it cannot know the blur was not intended.
A tracer cannot un-blur an edge
Tracing finds where the image changes from one region to another and draws a line there. If the change happens over one pixel, the line is crisp. If it happens over eight, the tracer has eight pixels of ambiguity and has to pick, and it picks the middle.
That is why a small image does not simply produce a small vector. It produces a vector with wobbly edges, because the wobble was in the source and the tracer was faithful to it.
Upscaling first does not help
Enlarging a 300px logo to 3000px before tracing adds no information. The edge is still soft, it is now soft over eighty pixels instead of eight, and the tracer still splits the difference. What it does add is time, and a lot of contour points.
What to do instead
- Find the original. Somebody has a PDF, an AI file or an EPS, and any of those traces perfectly because they are already vector.
- Re-export from whatever made it, at a larger size.
- If it is a scan, rescan at 600dpi rather than 150.
- If none of that is possible, trace in mono, raise the smoothing, and expect to tidy the result in the editor.
The other end of the scale
Very large images are not free either. Detail the trace never sees is detail the vector cannot contain, so nothing is gained by downscaling before tracing, but a 40 megapixel source produces an enormous number of contour points and a file that is painful to open. If the trace has to reduce the image, it says so rather than doing it quietly.