Choose the caption bubble to begin. If you want to upload a custom bubble or alternative image then submit the caption first, then edit the comment and attach a "Place img". Once the image is approved you can use it by editing the caption again.
I put a java uploader for galleries that scales down images before sending them to ESR (saving time on the upload). Unfortunately it doesn't do the best job and I guess you can see here you get a mspaint effect :).
It's still better than uploading a huge image to ESR and letting the site scale it down, since uploading takes so fkin long on UK connections.