animats

Thinking about it, I may have seen this, on rare occasions, in Firestorm. My home login location, in a second floor office with windows, overlooks three adjacent regions, and sometimes one of those regions comes up blank, then appears after a delay. If your login location doesn't overlook other regions, you might never see this. Comments?

> RegionHandshakeReply is complete. Yes, it is, and object updates are pouring in. They go on a holding queue waiting for the EstablishAgentCommunication and the capability fetch for the asset capabilities. Once those are in, the region goes live and all the held object updates are processed, at which point the region snaps into visibility. > But Dore's seed capability invocation by the viewer is delayed. I'd normally blame the viewer for the delay. No, as soon as the viewer gets EstablishAgentCommunication, it makes the HTTPS request for the initial region capabilities. I have logging of that; I just didn't send an entire log file. How does the viewer "invoke the seed capability"? RegionHandshakeReply is done, and we know the simulator got it, because object updates started coming in. > but it must be driven by IL activity forwarded from main (Morris). That is where I'd look. "IL activity"? What is that? There are definitely plenty of AgentUpdate messages being generated. The avatar can be moved around within Morris, and another viewer in Firestorm looking at the same spot sees it moving. Moving the avatar around, or keeping it still, does not seem to affect the 1 minute delay. Is there something else that the viewer must send to the main agent simulator? I've tried making the event poller time out and retry every 15 seconds. That has no effect on this issue. I'll get several empty timed-out polls, then an EstablishAgentCommunication message. Usually I use a 90 second timer and let the simulator time out. Works the same either way. So it's not a consequence of the poller timeout. Maybe. The 1 minute delay (± 2 secs) before EstablishAgentCommunication shows up is very consistent in all these scenarios. More excerpts from the same log summarized previously: 06:06:53 [WARN] EventEstablishAgentCommunication, unparsed: Map({"seed-capability": String("CENSORED"), "agent-id": UUID(CENSORED), "sim-ip-and-port": String("54.202.5.63:13003")}) 06:06:53 [WARN] Establish agent communication event: [MsgServerEvent { region_handle: (255232,256256), event: EstablishAgentCommunication(EventEstablishAgentCommunication { socket_addr: 54.202.5.63:13003, agent_id: CENSORED, seed_capability: "CENSORED" }) }] 06:06:53 [WARN] Establishing agent communication to (255232,256512) 06:06:53 [WARN] Region (255232,256512) state change from Connected to SeedCapabilityReceived: Seed capability from establish agent communication event 06:06:53 [WARN] Fetching capabilities for region (255232,256512) ... 06:06:53 [WARN] Fetched capabilities: RegionCapabilities(RegionCapabilities { start_time: Instant { tv_sec: 2291337, tv_nsec: 167534157 }, region_handle: (255232,256512), region_capabilities: {"GetMesh2": "CENSORED", "GetMesh": "CENSORED", "RenderMaterials": "CENSORED", "GetTexture": "http://asset-cdn.glb.aditi.lindenlab.com", "EventQueueGet": "CENSORED"} }) 06:06:53 [WARN] Region (255232,256512) state change from SeedCapabilityReceived to CapabilitiesReceived: Capabilities received. 06:06:53 [WARN] Region [Ahern] (255232,256512) state change from CapabilitiesReceived to Live: Live This is from the same Sharpview log as previously posted. There's been a 1 minute delay between RegionHandshake/RegionHandshakeReply for Ahern before EstablishAgentCommunication was received at the viewer end. That message was rerouted from the viewer's main agent region (Morris) to the neighbor region in the viewer (Ahern). That causes the viewer to request region capabilities "GetMesh2", etc, and those come back within the same second. Ahern then goes to Live state and appears on screen. So the delay is before EstablishAgentCommunication is received, not in fetching capabilities. It's quite possible that Sharpview needs to send something more to wake up the simulator, but I don't know what.

Because SL does not have a full hierarchy yet. You only get one level of child prims. Fixing this is an accepted JIRA. Philip Rosedale once said that not having a full hierarchy, like every other 3D program, was his biggest design mistake. You can sit on animesh. That's how SL horses work. Yes, if you wore a sex HUD and others could sit on you, you could have sex anywhere. Unclear if that would be a good thing.

What's the best current info for someone who wants to make basic mesh clothing? Are there free Blender models for basic shirts, dresses, pants, etc. that can be modified? Or do you have to get Marvelous Designer and do it from scratch?

Is there a legit use case for having permissions persist for items in inventory? If this really is an irrevocable situation, this is a bug. If someone can make that happen with a simple, reproducible, example, it would help to write it up as a JIRA. Make a scripted prim that, when sat on, makes an avatar do something once every few seconds when nearby. Demonstrate that none of the standard ways to stop it work. Make a video. Put a link to the video and the text of the script in the JIRA. Consider marking the JIRA as security-related; that's a checkbox. Then come to Server User Group (Every Tuesday, noon, Denby) and bring it up with the developers. Do the JIRA first. Otherwise you'll be told to file a JIRA.

The problem is real, but it's more puzzling. I set the poll timeout on my side to 10 seconds. That gets me a timeout and re-poll every 10 seconds. But it still takes a full minute to get the EstablishAgentCommunication events. Puzzled. Somewhere server side, there is a 1 minute timer, apparently. I'm doing something wrong, or a least differently, here.

Here's some logging info that provides a bit more insight. This is Wednesday evening on Aditi, and times are UTC. Selected log entries from Sharpview. Logged into Morris, standing near the Morris/Ahern/Bonafacio/Dore corner, a place I visit far too often. Here we're seeing the four regions being found by the viewer. They all get found. Morris gets found at login. The other three regions show as Discovered, which means EnableSimulator came in. Each of those regions quickly gets a RegionHandshake, and object updates start pouring in. (They go on an internal hold queue, though, until the region is fully live.) Ahern goes live quickly; 2 seconds after the RegionHandshake, the EstablishAgentCommunication event comes in via the event poller. The region goes live and appears on screen. But then, for a full minute, Bonafacio and Dore don't show up. They're stuck waiting for their EstablishAgentCommunication events. Note the times above. One minute later, the event poller in Sharpview reports an unexpected EOF on the poll request. Sharpview's poller has a 90 second timeout, so this is presumably a server-side timeout. So Sharpview does another poll, and now the EstablishAgentCommunication events for Dore and Bonafacio regions come in, and those regions quickly appear on screen properly. So long polling isn't working properly here. With a long poll in progress, an event from the server did not cause an outstanding long poll to complete and quickly return the new event. The poll had to time out first, and the next poll retrieved the event. Sharpview's polling is very simple - make HTTP request, wait 90 seconds. No libcurl retries. Uses the simplest Rust HTTPS crate: "ureq". I have more logging info, and can make this happen on request. I've tried this three times. Moving the avatar around Morris near the corner, to force more agent updates, does not change this behavior. One minute delay, every time. (I wonder if the libcurl retries are what makes this work for other viewers. A retry by the viewer would get the polling process un-stuck. That should not be necessary, but may be why it works in the mainstream viewers.)

Current tier pricing. Possible "purchase what you need" tier pricing.

There's Kawaii City. Look in search. This is a very strange place. Kowloon Walled City meets the Russian Mafia, with ruins of Tokyo subway stations. It's a stack of skyboxes connected by elevators and experience teleports. I spent an hour there and I'm still lost. More Blade Runner than kawaii as you get further from the entrance. Many clubs and bars, nobody there. They claim to have rentals, but I never found them. Japanese-themed areas include Hiroba Town in Davros and the stretch along Electra/Burns/Neumogen. Look around; you may find rentals.

Hm. I wonder if particles could be made to stop at reflection probe objects. Those sort of indicate "outside world ends here". Something to think about after full PBR deployment.

SmartBots has a good female bodyguard. Follows you around, even through teleports. Looks good. Not too expensive.

There's an initial implementation for the Other Simulator. Haven't tried it. It was at one time associated with an NFT scheme of some kind, back when NFTs were a thing. It's one of those things where you look up the subject and get interesting theory papers, but not success stories.

There's some work in that area over on the Other Simulator, involving "perceptual hashes", a way to determine if one mesh is very much like another. SL Marketplace could use something like that.

Can you run with 256m draw distance? That's about as good as it gets for now.

More dakka at last! The pew-pew crowd likes it. That will get more gamers into SL. Now this can get an upgrade to a higher and more realistic rate of fire.

Akamai server map. Each of those locations has a big cache. Viewers usually run out of JPEG 2000 decoding compute capacity or local network bandwidth before they hit the limits of the asset server network. Long avatar rezzing delays, especially after teleports, are a bug. LL has been working on that, and there have been test runs after Tuesday's Server User Group meetings where everyone present teleported to a test region running new simulator code. There, nobody stayed in pink/white cloud mode for more than 2-3 seconds. Things not appearing at all is also a bug. Those are called "interest list" problems. Monty Linden is working on some fixes to the networking level which may help there. These are not network traffic problems. If you watch network traffic, these bugs show as low traffic levels, not high ones. LL seems to be working hard on this. The motivator seems to have been a new emphasis on the WelcomeHub and new user retention from Linden Lab management. The WelcomeHub has enough traffic, especially when Motown has an event, to bring out most of those bugs. If new users encounter serious bugs in their first hour, SL looks broken and loses business. After years of waiting for fixes in this area, there is progress.

Lighting is about to get more serious. as PBR finally kicks in. There are some gimmicks, though, so every space unlit doesn't go pitch-black by default. That's the way PBR is supposed to work, but from last week's creator user group, it seems like that won't turn on until you add at least one reflection probe. Please, though, light your public spaces. I'm in a newly built region right now, and about half the stores have lighting, and half are dark. If you have a place open to the public, visit it with your viewer set to midnight and shadows on.

Anything that drapes is hard. You need physics with collisions during display, which needs much compute power. This is why long dresses don't work well in SL. Few games have full clothing physics, and even then, usually only on the hero character. There's an intermediate step, though. When you dress an avatar, cloth physics should run, layer by layer from the skin out, and get all the clothing properly layered. No need for alphas. I've seen a clothing design system where this happens. When you change the hair, it fluffs out and then falls into place. That only happens during changes; after that, it's just rigged mesh. (Rendering rigged mesh is an amazingly simple thing mathematically, and GPUs do it. It's the setup that's hard. So anything that can be represented as standard rigged mesh is fast.) Bakes on Mesh, The Next Generation.

Oh, if it's next to your parcel, put in a support ticket and say why you want it. LL may let you have it at $1/m^2. It's up to LL to do that, but if you have a nice looking build and aren't renting out your land, you may well get it. I got one parcel in Vallone that way. It's the one that has the big fountain and the stone lions. (I sometimes put tech demos around the fountain. Right now, one of the stone lions is a flat impostor. Can you tell without walking into the fountain? There's also a bench which, if you stand near it too long, will say "Please take a seat" and seat you using an experience. I've wanted those at "safe hubs", with maybe a 4 hour timer, to get the lost bots out of the way.)

There are quite a few systems for that now. None yet fully compatible with Second Life. That's a conversion problem, not an AI problem. Ready Player Me. They are not as photorealistic as SL, and that's by intent. They implemented full photorealistic avatar generation and backed off, because of the uncanny valley problem. You can get something out of Ready Player Me that you can import into Blender and work with. It won't have an SL skeleton or UVs. The system is primarily intended for games designed to use its avatars. Their business model is that they will someday sell you clothing and accessories for your avatar. Fotor. Can make a 3D avatar from pictures, but mostly from head pictures, not full body. Can't get logged in there; all my anti-hostile-code protections (Linux/Firefox/Privacy Badger) are breaking their login system. A bit suspicious. Real illusion. A pro tool, costs a few hundred dollars. Used for film and game production. Has anyone tried this? There are many more, so many that "10 best" spam sites have lists. I'm not recommending anything specific. Creators, please try some and report back. My art skills are good enough to make buildings, but not humans. There are also new tools for turning un-rigged models into rigged models. See AccuRig. This looks interesting, and it's free. You start with a model in A-pose (standing, arms at sides slightly out from body) or T-pose, and AccuRig rigs the model semi-automatically. Turning its output into an SL skeleton will take some programming, probably. This whole area has become much more automated. As SL moves more towards glTF models, it will become much easier to use standard tools to make SL content. The 3D industry seems to be going to glTF as a universal import/export format. Caution. Read the terms of service on these systems. Some of them restrict what you can do with what you get out.

We might get something else instead - AI tools to do the job. There are AI tools that can make an avatar from a few still pictures. And there are AI tools that can make good still pictures from a text prompt. "Victorian factory girl with dark dress and white apron." Generated by AnimeGenius in "realistic" mode, using GPT-4. Test for a possible NPC for a project in New Babbage. Have any creators been down this road yet?

Larger regions seem to be the big thing that is off the table. They work fine technically in the Other Simulator. This is an LL business decision, I suspect. People would want giant regions for their little house and yard. Most large regions on the grids that support them are mostly empty, and they make the world seem sparse. Personally, I'd like to have them even if the only thing they were used for was Linden-owned open space regions. Large areas of open water or bare ground with a road or two to fill in the gaps between revenue regions, perhaps. Extra water around the edges of the continents, especially Sansara. Clean up rough edges, such as the southeastern empty space of Zindra and the northern edge of Bellessaria, where you can see the edge of the world not going down to ground level. (There's been progress there. The Moles put in some end of road barriers and shrubbery, so you usually don't see those spots by accident.)

Multi-user Blender access is coming. Multi-user Maya access is working now. NVidia Omniverse Connector is a complicated system which allows collaborative editing across multiple programs. There's a special version of Blender which will connect to an NVidia Omniverse server, and when you change something, you can export to the server and everybody else in your session sees it. Some group of people who work together in Blender or Maya should try this and report how it works. This could be a big win for large group efforts, such as mole builds and Fantasy Faire. This is all based on the glTF standard, which SL is now starting to use for materials, and which LL says SL will use for meshes at some point. Once SL makes it to glTF, it might be possible to put an Omniverse Connector into an SL viewer, for real-time import/export. It would be semi in world editing - you see it in world, and when you edit it in Blender it changes in world, but you have to commit changes and pay the upload fee for others in world to see it. So there's a path that leads to what Arielle is asking for, but other things have to be finished first. This is unlikely to be a typical user thing. More like a tool for serious creators with enough screen space and compute power for both Blender/Maya and SL at the same time. The Omniverse thing is for production teams, not end users. There's a free version and a paid version. It may be too industrial-strength for SL.

Maybe protections against objects that can debit should be stronger. Require both an experience permission and the yellow popup. LL takes abuse by experiences seriously, because they give objects considerable power over avatars and have griefing potential.

I have the happy path for attaching to neighbor regions working in Sharpview. The long poll seems to work well with a 90 second timeout viewer side. As Henri points out, not using libcurl seems to help. (Sharpview is all Rust code, so the lower levels are totally different from C++ land.) I get that the seed capability is asynchronous. I realize that's part of the security system. I haven't dealt with teleports yet. I'm currently looking at how objects cross region boundaries. I spend a lot of time watching the New Babbage trams making their rounds. Miscellaneous questions: What indicates a region restart to the viewer? I accept EstablishAgentCommunication any time after sending UseCircuitCode. Sometimes it takes a while to show up, as in tens of seconds, and yes, it does seem to be driven by camera updates from the main region. Thanks for confirming that. As I said above, I thought I needed to do something to kickstart that. Logging in and just sitting there results in no EstablishAgentCommunication events from the simulator. I probably need to send some agent updates even if the agent isn't doing anything at first. I notice that if I get an EnableSimulator but don't bring up the region, I keep getting EnableSimulator messages. Is it a problem for the simulators if I don't bring up a region viewer side the simulators would like to have go live? I'm considering a setup where I have a draw distance long enough to fully cover 4 regions around the main agent, but only the closest 4 regions are brought u in the viewer. Some kind of region impostor will be shown beyond the nearby regions. Is this going to create any problems simulator-side? At logout, is CloseCircuit sufficient to close out child regions, or is some other handshaking required? I really appreciate the answers posted above. Us TPV people have to try to figure this stuff out from logs and the viewer code, and that works well only for the happy path.