SL Project news week 44/3: mesh deformer

Testing is continuing with the latest release of the Mesh Deformer project viewer, which can be used to deform mesh items to either default or custom human shapes. While the pool of test items remains small, people appear to be testing using their own creations, with at least some feedback being given to the JIRA (STORM-1716), which remains open to comment. If you are testing the deformer using the latest project viewer, please be sure to provide feedback on your results – be they with default shapes or custom shapes – to the JIRA.

Some problems with breast fittings might be down to an incompatibility between Avastar and the viewer, which is currently being corrected

Most of the results obtained to date appear to be satisfactory, although some issues still remain with custom shapes. Darien Caldwell, working with Gaia Clary, has identified one issue which exists specifically with the Avastar add-in for Blender co-produced by Gaia.

Avastar is a Blender add-on for Second Life mesh creators and animators which provides a wide range of capabilities, including (for mesh creation): SL shape import into Blender, SL shape sliders support, support for attachment bones, and so on.

The issue has been that Avastar’s sliders have been based on a scale of 1-100, whereas the viewer’s sliders operate on a scale of 0-100 , leading to some scale miscalculations within Avastar which in turn have led to issues with mesh fitting over body parts such as breasts. According to Darien Caldwell, she and Gaia now have this “pretty well nailed” and an update to correct Avastar will apparently be out shortly (Update: please see Magus Freston’s comments at the end of this article).

This still leaves the broader deformation issue, as reported recently, which is still being looked into, and awaiting some feedback from Qarl.

Other issues outside of these which have arisen with the deformer have been largely the result of unrealistic expectations – that it will, for example, mimic facial morphs or hand movements closely or some changes to feet. However, in these situations, it is important to remember that the deformer was never developed to deal with these, as it works off the avatar’s bone structure, and facial features and hands don not have any bone structure within the avatar associated with them.

Time Frame for the Deformer

While progress with the deformer continues to look good, there remains no ETA as to when the code will appear in the release version of the official SL viewer.

The major reason for this is the ongoing problems with the Beta release channel for the viewer (of which more in the next update for this week!). As it stands, the deformer is positioned roughly at the back of the queue of releases which are being held as LL work to resolve the current crash issues with the Beta viewer. This means that, at least until the Beta issues are resolved, there is no official ETA for the deformer code reaching the release viewer. However, the latest revisions are starting to be incorporated into some TPVs.

In the meantime, and if you have been testing the project viewer, please remember to give feedback via the JIRA.

Performance Concern

While it is not actually an issue with the deformer per se, commenting at the Content Creation User Group on Monday 29th October, Siana Gearz highlighted a potential problem with mesh clothing utilising the deformer and avatar physics.

The concern is that the deformer uses the same morph-based schema as is used by the avatar physics system. This means that the GPU has to do a lot of additional calculations for the polygons in an item of mesh clothing to simulate movement (such as “bouncing boobs”) when avatar physics are in use. This obviously leads to a performance hit. So long as the polygon count in clothing is kept low, the impact is minimal, but the concern is that clothing build using high polygon counts to provide detail could have a larger impact on the viewer.

One possible way for this to be avoided, should it become an issue, is for clothing makers to optimise their mesh clothing with lower poly counts – and the forthcoming materials processing capabilities should go a long way towards helping with this.

Related Links

9 thoughts on “SL Project news week 44/3: mesh deformer

  1. I am not really keeping on top of the details of the Mesh Deformer project (since I dont make mesh clothes) but I read blog postings like this just to keep my finger on the progress.

    But, the last part of this blog regarding performance and clothing mesh creators making more poly-count efficient clothing to help reducing sim load caught my attention. Especially this past week that relates to this subject.

    I predict a lot of mesh creators are uploading beautiful mesh models into SL and are either blissfully unaware or simply dont care about efficiency of their models. I suspect its is more the former. Most mesh creators were not part of the inner circle of the LL development / deployment of mesh onto the SL grid so they don’t understand intimately the impacts of huge polycount model vs one that is very efficient.

    Added to that, likely most mesh creators do not follow / interact with the SL Mesh Forum discussions and learn what would be considered to be an efficient mesh model as well as what techniques can be used to substantially reduce polycount with things like texture techniques.

    In fact, I met a mesh creator who makes mesh art sculptures. He gave me one to rez. It was a beautiful model with a relatively simple shape but it rezzed at 71 LI at only about 3m size. I am sure I could have made that model under 5 and still look almost identical in quality. BUT at least with these models, the inefficient models are very clear and as SL customers start making mesh efficiency a part of their buying decision – the mesh art creators will be forced to reduce polycount.

    BUT… mesh clothing creators have NO PENALTY. I saw another mesh creator’s actual model / frame of a mesh clothing item. The design was GORGEOUS but I could see that this clothing will have a huge polycount. Large flat surface areas has a huge polycount in them where it could be reduced substantially.

    LL needs to incorporate in the upload some kind of “efficiency Rating” system. I am not sure how, but it has to be IN THE UPLOADERS FACE to tell them – “this model will work but is not efficient for its size”.

    LL need to place Rezzing Penalties on attached/worn clothing like they have done to non-attached mesh models. It was because of these rez penalties for mesh that I stopped making mesh landscape features since they are simply too LI penalizing compared to my old sculpty models. But clothing should have imposed penalties for anyone that wears poor mesh efficiency clothing models.

    LL Marketplace should also have very clear standards to report mesh model metrics that show a model’s efficiency so that consumers will start buying the better models and encourage mesh creators to make better models in order to be more competitive.


    1. It’s a tough nut to crack, and I actually debated on raising the comments from the Content Creator’s group, as the last thing we need is another round of ARC-policing style activities with people pointing fingers at high-poly clothing as a casue of “lag”. On the other hand, if there is going to be an issue, better to try and give a heads-up.

      Hopefully, the situation will be largely avoided as a result of the materials processing capabilities which will be rolled-out to SL in the near future. These should allow very high levels of detail to be added to mesh items for relatively low poly counts in the models themselves. Many creators are looking forward to materials arriving, so I’m certainly hoping those making clothes will be grabbing the capability with both hands :).


  2. I genrally dont like the policing by penalties either and I also believe that the initial LL mesh development/policy team (from advice from an inner circle of top professional mesh creators) went overboard on the rez penalties for mesh as well. It saddens me that because of the huge penalties placed on the SIZE of the model (regardless of the polycount) that I have had to abandon making new mesh versions of my landscape terrains. There is so much more I could have created in “beauty” for waterfalls and rock formations using mesh over sculpties – but rezzing them would not have been feasible compared to 1 prim for a sculpty.

    I think LL missed and is missing the boat to promote mesh adoption and encourage the uploading of beautiful AND EFFICIENT models. Its not surprising which approach they took. LL is a culture of Engineering / Tech geeks and they were adviced during the development of mesh for the SL grid by more technology deep 3D modeller (tech geeks). As such, when you are really good with a hammer – you will solve all your problems with nails.

    LL has never been good at business and understanding their customer base. As such, it never dawned on them that they should have dramatically tempered down the technical mesh penalties (which made sense to them) and spent FAR MORE TIME in developing and promoting education to the greater community of ameteur mesh creators in SL like me and 90% of the other mesh creators that only learned Mesh from Googling and reading forums and word of mouth and experimenting.

    LL should have created a TRAINING COURSE for all mesh creators… to provide all the critical knowledge any mesh creator should know when making mesh. It only had to be maybe a set of 3 or 4 two hour inworld or CBT training sessions. During this course, it could have explained all the techniques to make a model more efficient and what would be consider a better vs a poor developed model.

    To encourage all mesh creators to take the course and LEARN PROPER TECHNIQUE, any creator that completed the course would become a certified Mesh Creator of SL. Their Avatar would be registered with this skill. Then, in Marketplace there would be a flag to check the mesh model was uploaded by a certified mesh creator (which would display a small crest/designation).

    Customers would use this to know that AT LEAST the creator took the time to take the course and understands all the basic techniques on creating and uploading a GOOD mesh.

    To further encourage creators take the course, their respective avatar would be awarded a reduced upload fee during the upload process when the process sees that the avatar is a certified mesh creator. They could have even gone further by using their tech skills and added a field on a mesh object that the model was created by a certified mesh creator and as such its rez cost is reduced by 50%.

    As such, by focusing on the HUMAN aspects of the problem and focusing on TRAINING AND EDUCATION of the masses, LL would have substantially increased their adoption rate of mesh and more quickly addressed GOOD MESH CREATING TECHNIQUES.

    As for the materials system saving the day…. I cant wait for it either BUT again, it doesnt mean a hill of beans if LL does not educate the mesh creators like me that only learn how to create mesh by googling for reference links and forums. I have got around the materials system by baking it on via PS…. but many others will think that with or without a materials system, they need very high poly counts so that they can show all the wrinkles in their clothes.

    Its all about EDUCATION…. not POLICING… but I know LL only understands the latter.


  3. I still find amazing how many don’t even know who many scripts they are wearing (LL v3 about land tab, scripts/avatar scripts even tells where they are placed on waht one is wearing!) or that they can remove a lot of them, mainly recizeable ones!
    Still i agree that some buldiers should be warned and i bealive the best is to not buy their products, cause when i see still clothing that uses scripts and does not allow them to be removed in 1st place (be cause they didnt add that feature to the hud menu, or cause the object is no mod) i feel the urge to show how those products will never be used on sims that dont allow scripted avatars!
    Same will happen i think with Hp mesh, sooner the sim owners will staret banning avatars who are passing the limits of the reasonable!
    Still, the last i want is see the Lab creating another rule, cause in ultimate case, is not the customers, but those who provide a service, thaty should make sure its hardware can support its needs!


  4. Being the guy who started the materials project, I’ve spoken to a number of clothing makers in the past regarding polycounts and how materials will help them drastically reduce how many polygons their clothing makes up. The general feeling is “I know what I’m doing, and there isn’t any texture that will make my clothing look as detailed as this.”

    Even when provided with *direct comparisons* of what a high poly model looks like before it’s made low poly with a normal map applied and after, many content creators I’ve spoken to just can’t seem to grasp what is possible with materials yet, and are maintaining this “My way is better, because I’ve been doing it this way since I learned how to do mesh!” mentality.

    Naturally I don’t speak for all content creators, but when a number of them refuse to learn a new tool that could vastly benefit their content creation pipeline in terms of visual quality and content efficiency, that says bad things about what will eventually unfold once something such as materials is released onto the grid.


    1. Yes Geenz and even a friend mesh clothing developer who is not arrogant but learned her skills similar to how i did – reading tutorials, blogs, forums, and lots of experimentation… she creates mesh clothing based on the quality of the surface attained by how many polygons can make the wrinkles and nice smooth corners and edges. That an open slightly curved area of the model has hundreds of polygons where only a few needed was not worth removing.

      There is no rez penalties or visible lag on attachment penalties that her clothing creates to force her to develop low polycount models. It works good enough without trying to learn model efficiency techniques like bump/specularity baking on the model now and future materials systems in the future.

      But when so-called expert/professional mesh modellers told you they know better and do not believe that materials systems provide a much more lag friendly and beautiful model… that is pretty sad.

      I would love to work with models with unlimited polygon counts and not have to denormalize my mesh models for the virtual world of SL. Its a major pain and you always have to worry about the model going wonky if you reduce your LODs too low. But at least for my mesh models that ARE PENALIZED for beign rezzed in SL, my models look good and are only 1 – 6 prim LI.

      Sadly, the clothing mesh makes have no incentive to learn model efficiency techniques nor are they incented to develop effecient models for clothing.

      That is why I suggested a “formal course” on mesh creation for SL developers where their certification and reduced LI and upload charges for being certified would be some incentive.


  5. Hi Inara, not sure where you got the claim that Avastars sliders go from 1-100. It’s kind of funny as it’s trivial to refute for anyone actually with Avastar :). There _is_ a round off problem that Darien pointed out to me but this turned out to really be an SL bug. See here for details:

    We will change Avastars slider’s to be continuous rather than integer values which will mean that we will support the actual shape that SL generates. Unfortunately this may not be the shape reported SL’s own sliders due to the bug.

    If you have any other cases where a shape or rigged mesh doesn’t match please let me know. We are keen on fixing any bugs we find.


    1. Hey, Magus

      Many thanks for dropping by and correcting the info. It came up at discussion at the OpenDev meeting in what appeared to have been a first-hand conversation between Darien and Gaia, so I took it as verbatim. More than happy to add a correction :).


Comments are closed.