Desperate for a definitve answer on Lumen ghosting issue

Heya,

I’m pretty sure the walls are correct (backfaces on the inside, over 10cm) but to check i replaced with the stock cubes and still getting the same issue (still showing as black in visualizer). Any settings I can mess about with that might help?

It’s not the end of the world, i was considering baking the lighting anyway but just wanted to give Lumen a go. So if you think it wont get much better than this then no worries!

If you are using the stock cubes, you need to make sure generate distance fields are enabled in the static mesh settings.

1 Like

Thank you everyone for this thread - I was getting ghosting, too (tho not quite as much as the OP - oof!) - and this has been very helpful.

I don’t have much to offer – and this switch is not a cure-all, given the amount of ghosting you’re seeing here – but wondering:

In your sequencer, in pulldown next to the fps setting, is your “Lock to Display Rate at Runtime” off? (If it was on, that could also contribute to ghosting effect.)

Thanks again everybody - this was super helpful & seems to have solved my own ghosting issue.

2 Likes

Im getting this ghosting in Lumen, Im rendering for production. I moved to the max the values of Lumen in Postprocess Volume and still getting this weird ghosting

Unfortunately, that level of ghosting may just be essentially unavoidable. Lumen needs temporal accumulation in order to resolve well. That being said, you can use the ‘lumen scene lighting update speed’ and the ‘final gather lighting update speed’ in the PPV to make it flush lighting faster, but it runs the risk of both making the scene more expensive (not a problem for production renders) and noisier.

Also check the lumen scene debug views and make sure the lumen scene roughly matches the actual scene.

Yep. That’s Lumen temporal accumulation, which is 10 frames by default, in the highest setting. So yes, it will generate ghosting. This is Lumen, a dynamic lighting system created for slow motion or low action games (wait, isn’t it?).

PS: you can change those temporal frames accumulation with Cvars (but you will need to compensate the noise, flickering, etc, too)

Lumen definitely presents some tradeoffs in its’ current form. I think the fact that we have real-time GI and reflections that’s this robust is kind of a miracle, but the documentation also isn’t where it should be for solving the problems users actually have.

As we’ve seen in the matrix demo and fortnite, lumen can robustly support action games with a lot of destruction, just maybe not action+outdoors+emissive lighting+wild amounts of skinned geometry on a low-end machine. I’m not arguing with your observation, I just think it’s interesting to see what you can and cannot get away with in lumen.

1 Like

Of course @jblackwell !

I agree, so great to have a realtime GI, but also Cyberpunk has realtime ‘path tracing’, just to mention an example or ‘state of the art’ of realtime GI.

Regarding Lumen, it give us some incredible advancements but also some steps backward against Ray tracing, which could has been also updated with some of the old missing things, like reflecting RT GI too, for example. Of course RT could be more expensive than Lumen, but in the same proportion more realistic/precise too.

IMO, Lumen offers a quite balanced quality/performance, but it’s difficult to rise the quality (speed, ghosting, reflections) if you don’t care about the perf.

I have seen lot of people having troubles with Lumen just with action+outdoors/indoors, even without considering your worse scenario of +emissive+wild geo.

1 Like

I hear what you’re saying, and I absolutely share the enthusiasm for the cutting edge technology. I get the feeling that Nvidia and Epic might be in talks atm on porting some of that technology more directly to Unreal Engine, but that’s just speculation.

That being said, as I’ve gotten to talk to with a few devs working on a recently-released AAA project, I’ve come to really understand the bottlenecks this generation is going to be facing. Lumen may be the best solution we have for real-time lighting this generation, and other GI systems across the industry (E.G. GIBS), share similar limitations.

For shipping large products, anything major one releases this generation has to scale down, minimally, to the Xbox Series S, which is only a 4 TFLOP machine with seriously compromised bandwidth. Lumen can manage it at 60FPs, but just barely and with a lot of cuts. If we can’t hit that machine as developers with lumen, then real-time path-tracing isn’t going to see the adoption it needs to really take the industry by storm.

Not to mention, I have a pretty console-comparable GPU (2070s) and a stronger CPU to my knowledge, and Portal RTX and Cyberpunk Overdrive are basically no-sells for me. I can’t even hit 720p 30 on the former, and the latter requires ultra performance DLSS to even be playable. The only thing real-time PT provides over lumen (currently) is high-quality GI in reflections, and even then Epic is seriously closing the gap.

Add to that the fact that lumen RT relies on what’s called inline ray-tracing, which is basically a very raster-y way of processing ray data that makes multiple recursive bounces (the cornerstone of path-tracing) harder to do from my understanding.

On your note of Epic’s old RT implementation having some strengths over lumen, the old RT system also shared quite a few limitations. It can’t handle occlusion in any real way, and light leaking is pretty prevalent. It can’t support as many instances performant, and multiple bounces explode with noise due to their limitations. Emissive objects couldn’t propagate lighting either, and lumen’s ability to do large area lights essentially for free is one of its’ big selling points.

I’m not trying to shut you down or bring up any level of contentiousness, I’m just relaying what I’ve learned from a few devs who’ve had to target the broader console space rather than just indie or PC-exclusive. Sorry for the long post.

Perhaps the biggest killer may just be Moore’s law: if we can’t build better chips to handle the tasks cheaper, then the market that can afford to use this tech will get smaller and smaller, and therefore less and less financially viable to develop for.

1 Like

Making a render in sequencer, using mostly Spatial samples I get more ghosting, but if I use temporal and increase the samples to 32, the ghost mostly goes away