I just had a quick look at the dataset as well.
Some great pointers here already;
- Use the Ha only (you can just leave the type of accent/data at "H-alpha or S-II from a narrowband filter"; ST will extract the red channel automatically if given an RGB dataset)
- Ideally image with a true Ha pass filter, rather than having O-III "polluting" (in small amounts) the signal as well.
- Use the Galaxy preset and then a smaller Detail Size parameter to only accentuate smaller detail.
As you know, however, something like AutoDev is allergic to correlated (e.g. non Poissoninan/shot) noise and will bring it out as detail. Since 1.8 there is a correlated noise filter in the Wipe module. Use it to reduce the worst of the correlated noise in the shadows. You should notice that whenever you use it on "regular" datasets with correlated noise, you almost don't need to use the "Ignore Fine Detail <" parameter in AutoDev anymore; it is that effective.
You should now have a much easier time "matching" the datasets in the NB Accent module;
Once in the second screen, start off with the Galaxy preset, set Strength and Gamma.
Trhottele Detail Size until the knots become visible, but don't go higher.
If your accents are correlate very poorly with the visual spectrum data (e.g. there is really nothing visible in the visual spectrum data where there is something visible in the accent data), then ease off the Brightness Correlation (or turn it off altogether).
So you'd get something like this for example (screenshot is from 1.9 development version, but 1.8 should allow you to get close); Hope that helps!