If the sound doesn't change with velocity within the source instrument (VST or keyboard) then there would be no value in sampling multiple velocities.

If MODX doesn't have the effect or a close enough character to it - this would influence the with or without FX question. If you never use the without-effect sound then less is lost by sampling it. Depends on the effect, though as some effects are more "dynamic" or conditional than others. If you sample in the effects then a benefit is you free up MODX's insertion effects for something else. This is a finer point which may also deal with element "packing" and other possibilities. Options I think are beyond your simplistic approach.

If you want to spend the time doing the EG work - then this gives you more flexibility to either change statically (change the EGs different from the original source) or dynamically (higher velocity changes EGs or motion sequence changing EGs or ...). This is a tradeoff of how long you want to spend in sound design vs. having something ready-to-use. I don't have any input, generally, on which approach is better.

Big ones first:

You can sample one note per octave. That's 11 semitones of shift required, which isn't much of a stretch.

If you want more accuracy, go down to a fifth. I don't think there's much benefit from going any tighter than this.

Velocity's maximum is 4. I just do this, each time, because I do a lot of envelope shaping that's highly responsive to velocity, and would actually prefer 8 velocity samples, but that's not possible.

Because the envelopes on the Montage/MODX aren't very good (IMNSHO), I'd sample them from the source, especially if they're velocity responsive.

After that, on FX... that's a tricky one. Up to you, as to how you want to later mangle it.

If you do an octave x 4 velocity samples, and the samples aren't lots of seconds long, you might able to do one with FX and one without FX, not eat up to much sample memory, and then put one in one Element, and another in the other, and blend between them as you like via Assign Knobs and/or SuperKnob.

This is the really tragically under served and under utilised super power of Montage/MODX - and why it's epically sad that they didn't include a super fast SD card slot for loading and unloading GB of these. The USB rate of these things is ridiculously slow.

Thanks!! Very helpful!

I forgot about one more important question!

Came to my mind after reading this:

and the samples aren't lots of seconds long

How long should I sample?
Again, I need to take an assumption here, let's say for the sustained part.
I downloaded some samples from the web, and they were 10s+ long, which I found very long for an instrument (for example a flute or violin like instruments) and I cut them down to around 3s.

@Jason

If you want to spend the time doing the EG work - then this gives you more flexibility to either change statically (change the EGs different from the original source) or dynamically (higher velocity changes EGs or motion sequence changing EGs or ...). This is a tradeoff of how long you want to spend in sound design vs. having something ready-to-use.

I guess I have to take this hard route! Now thinking about it, I prefer to sample a sustained/flat part then do the shaping on the MODX. @Jason, learned this from you in previous post abour Mono/Legato/Re-triggering in AWM2, the instruments I need to sample are mostly played legato, so in order to achieve that, I need to hide what you called "artifcats" from the original EGs and FXs.

@Andrew

You can sample one note per octave. That's 11 semitones of shift required, which isn't much of a stretch.

Cool! That would make my life easier, especially that I need to spend some time with what I just wrote above (EGs and FXs).

Sorry to repeat a question here, it seems one was not seen:

How long should I sample?

For sounds that "start" and "end", e.g. drum sound, then that's pretty obvious.

But for other instruments for which I'm sampling their "Sustained part" only, e.g. a flute blowing, what is the ideal length for quality/memory @16 bit? 3s? 5s? Any general rule here?

I didn't answer because sample length is a personal choice. For looping samples, sample robot detects the loop point and should tolerate shorter samples. And then how much tail you have or attack. These are things you need to decide how much you need. The goal would be to have the bare minimum to preserve memory. But you may have competing goals and need to decide what tradeoff you're going to employ.

Note that if the source sound itself has 150ms of attack, 150ms of release, and the "looping" section itself is 150ms (you'd have to somehow "figure out" something about the original sound if possible) then there would be no reason to sample for 5 seconds because the original content is only 450mS.

I cannot possibly know anything about your recorded content.

Thanks Jason! Your answers always help.
One thing I noticed is that even if the wave has no attack or release, or I am intentionally only sampling the sustain part, there is still some sort of loop where the wave shape repeats, so I guess I should aim for this first repetition to keep the size at minimum.