hmm, IE8 and RE-0 seems to extend further than the TWFK treble. What sonion driver are you referring too ?

http://www.sonion.com/Products/Transducers/Receivers/~/media/Files/Products/Data%20Sheets/Transducers/Receivers/Receivers%202300/Receiver%202300/2389_v3n.ashx

Specs of 2300 doesnt seem to indicate so.

I have 2323 which shows similar on measurement.

Ok, back to it. I've managed to get good 3D models out of the pictures of the impressions, here are some pointers if you want to do the same.

• You will need either VisualSfM application (which is only licenced for non-commercial usage) or learn to use Bundler and CVMS from OSM-Bundler package. (OSM being Open Street Map). Googling these is easy.
• 30+ pictures from all the angles are necessary for good reproduction.
• Good strong lighting is a must. Flash causes the process to fail.
• Cellphone cameras tend to provide worse results, such as not being able to focus properly on such a small object without sharp edges. Manual focus is recommended.
• Best distance is about 50 cm to 1m.
• My pictures were done with Canon EOS500D, which is in the Bundler database, but it is not important unless you want to keep absolute scale.
• Take pictures one impression at a time - for multiple impressions at once, the results were not acceptable.

I've also attempted to make a pair of "fillers" out of PU by filling PU impressions with a layer of silicone putty and then filling the remaining space with the PU. However, the experiment has failed right now due to silliness on my part, I'll retry next week.

On the other hand, I've ordered way better silicone supplies now, they should arrive late this month. This includes special PU foam mixes (stiff and soft), transparent silicone and PU, high performance condensing silicone, high temperature (300 °C) silicone for future attempts at zinc-based shell, high hardness (Shore-A60) silicone for structure.
Perhaps later I'll order some actual mechanical handheld tools which can work in silicone and PU precisely.
I forgot to mention I also got sulphurless polyclay and plasticine - the other try to make the fills and perhaps clay shells too - which might be an interesting experiment.

Prepared RE-400 for CIEM makeover, that's how they look now.

That white thing is high density, pressed cotton wool, amount deemed optimal by Vas calculation. (That is one weird driver.)
It has been coated with a sealant lacquer and is quite rigid. The cyan thing is a bit of silicone coat, later "waxed" into the surface to make it even more waterproof.

The sound is sweeter, warmer, airier and much more linear. There is a small 1kHz dip present, the highest end lacks only 6 dB to linearity. 2k peak is gone. Subbass is linear and the bass becomes 3D.
Sound is more coherent and smoother, while slightly improving in precision, especially in lower midrange.
The filter and foam had to be removed, they turned the sound really really dark. Medium depth fit with the pictured tips.

New developments with new methods.

Now trying for a pouch mold instead of a full negative - should be easier to work with. This one is with Mold Max 60 condensing silicone, high temperature so should be good even for, say, zinc shell. Very stable silicone and friendly to urethanes if need be.
I'll try to make a 1 mm "skin" for the IEM from Rebound 25 brushable or Dragon Skin 15 FX Pro (both pretty fast) then additional hard fill from Mold Max. This will serve as a "skin" mould - in which it should be easy to install various equipment, and the inside "plug" will make it easier to create replicas.
I'll use PU foam as the rigid support for the pouch, but that should be really optional, as Shore A 60 silicone is quite rigid. Dragon Skin is milky transparent, so various color effects are possible. The surface is not smooth to touch though, but for that I have an additional thing - Smoother.

The other test involves silicone foam - I'll try to make custom foam earphone tips / plugs. It might be possible to create a layer like that on a CIEM to alleviate any kind of misfit. It's supposedly highly resilient and can be colored. That foam is Soma Foama 15. High end stuff. It will also be used as dampening (after cutting off the smooth layer - it's open-cell) and as anti-vibration mounting.
EDIT: It seems that a soft pouch mould is pretty much required for it if you want it to expand beyond. It's still slightly expanding and definitely not fully cured after 30 minutes. Looks nifty though. Less memory than Comply, but way softer. Not sure about tensile strength - it's supposedly relatively hard to tear once fully cured. Demoulding it from even a soft full negative (non-pouch) was very hard.

I've also managed to finally create good 3D meshes out of ear mould pictures via SIFT + Bundler + PVMS2 + MeshLab. (Actually, this trial run was on some other weak machine and I used reduced resolution pictures, so probably lost some points.) It's not yet final, but definitely is viable. I'll also have to try with zoom and/or in macro mode for more details.
Good lighting and focus are a must. I will need some additional object for proper scaling next time. I'll try 3D printing these meshes - of course a 3D printed surface is not really smooth, so it won't be usable as a whole shell - but the meshes are very useful in CAD for designing tube locks, placing air tubes, calculating volumes etc.

Next work will be on applying some freshly acquired Body Double to copy my outer ear and cheek shape. The goal of that subproject is to make a personal ear-and-cheek simulator.