Actually I now looked through most of the pictures on a PC. The AGC was great on colours before 20 minutes to sunset - then it went haywire - some reddish, some yellowish, some vivid, some ... Would need lots of time to fix them. 9 out of 10 pictures are essentially rubbish.
Before the AGC was great.
The vivo stock camera fared worse - it looks great on a phone but once on a 4K screen 1/3 is somehow out of focus - e.g. focus on legs instead of upper body for portrait. Notice nearly all pics were in portrait modus on 1x or 2x camera AND against the light. Exposure was really bad for most pictures - and it's clear that the AI just didn't know what to do with them. 5 pics in a row underexposed, then 5 pics heavily overexposed.
Now here is the thing - 1 pic out of 10 put the Iphone 13 Pro pictures in the dust - but 7 out of 10 are too bad to fix them - and it's really not possible to tell on the phone screen.
The iphone 13 Pro pictures look boring - but they nearly never miss exposure, autofocus is spot on and colours are boring but didn't have any of those fallouts like Vivo did. Detail and sharpness are much lower but for Instagram plenty enough. Also all cams work very similar in colour - and if you reframe it will be again same colour and exposure. I am pretty sure Vivo just takes way more pictues and fuses them together - while the Iphone is much more coservative here. I guess the bracketing on the Vivo is much wider - which makes for better results if the AI can stitch them together - but makes stitching them together much harder.
Gcam by default with 15-20 pics in HDR Enhanced of course also fore the AI/Fuse approach - but yeah gcam is tuned for Tensor and their own hardware. Put the same algo on other hardware and it will be much more hit and miss.
AGC had sometimes probelms with focus as well - but overall better. As conditions get difficult overall AGC clearly did better.
Essentially the Vivo can give you great shots in diffictult light - but it's really errratic as the conditions get difficult. You will need many shots - change modes, change camera - zoom somewhere else and back so the AI starts from scratch. If you just keep klicking away it will not restart and you can shoot 10 horrible pics in a row.
And yeah a lot of this is due to the preview having way too little dynamic range. On the iphone you pretty clearly see what you get. On the Vivo you have to watch the picture aftwards and it will be really different when you shoot against the sun/light source.
Gcam clearly had AI problems on colour as it got close to sunset - or better it used way too much color and then got the WB wrong on top. But exposure was fine and not sure if autofocus was better - because it does so much sharpening - but overall in the end you don't end up with the legs sharp and head unsharp already on the 1x lens.
In good light/easy shots don't bother with gcam. The vivo stock cam is good enough and much faster. Though I still have the problem of vivid vs Zeiss and want something in between.
When it hits, it hits, but far too often you end up with rubbish, really rubbish. But yeah the X90 Pro+ can sometimes (!) turn out shots like no other smartphone - but that takes many tries and you will only really be able to assess it later on the PC. On the phone it's hard to assess. At the same time it turns out a lot of junk while other phones will just turn out a lot of average pics.
A lot of this is of course due to sharpness - the sharper a pic the more apparent mistakes/problems get.
A big problem is of course - the best pcitures are taken around sunrise/sunset. If the sun stands high you don't get interesting pictures most of the time. I don't know why the gcam at some point goes horrible on colors on the X90 Pro+ otherwise it woudl be the best solution.
And yeah - I don't think this will be fixed anytime soon. This likely will take years - years of training the AI by vivo taking lot's of pictures and assessing them to get it right. It has the hardware to work great, it has a chipset fast enough to really get it done - but it clearly misses thousands of hours of optimization. Like google toook millions of pictures ot get skin tones right - it will take millions of pictures in difficult situations to ge the AI right on Vivo. And yeah this is mainly about portraits - not face portraits but full body (which is much harder). I think far too often it simply doesn't identify the face - and that's why the HDR clearly fails. If you take typical upper body portrait also in difficult light you have like a 1/2 chance. But if you take a picture with the body being 1/3 of the height of the picture - then very rarely the AI gets it right.
On Iphone the AI definitely gets the body - it just lacks the hardware to churn out a great shot and it goes conservative while the Vivo AI goes for the best shot but messes up ever so often, messes up badly. If Vivo would go more conservative I am sure it could get a better rate. AGC is much slower and takes much more pictures. I think Vivo on Auto takes very differing wide bracket, maybe too little pictures.
Note I took pics in Vivid - because in lower light Zeiss is too boring I find. That may have excarbated the problems. On Zeiss the approach is more conservative but for my taste Zeiss does not work well at all with shots where the body is 1/3 to 2/3 size in the picture. The gcam shows that the AI is much better for full body portrais with scenery - but it just shows it isn't made for the Vivo hardware. Because colors are the strong point of Pixel 7 bare none.
The vivo feels like a supertalent - but raw diaomd that just started training. In good light the differences are pretty small. in difficult light they are enourmous. BTW - my P30 Pro would not do well at all in those conditions - because it has very poor dynamic range and HDR. Those against the sun or similar shots simply would not be possible 3-4 years ago. Smartphones now just do way more AI and fusing pics together. That's why Huawei/Apple/Google win on dxomark - they just put in way more work than the others the last years into getting reliable shots. Vivo I feel wants to overtrump them but just misses years of training data. There's a reason that BBK uses the IMX766 on so many phones - they had time to finetune it. The sensors on the X90 Pro+ are all new to BBK - and that is apparent. For easy situations it's good enough, in difficult it just goes havoc and sometimes hits by chance (yeah because it's clear that very few shots work while most don't).
A lot of this surely can be overcome if you use the camera every day and remember when and where it worked - so the photographer is way more important here as long as results are so erratic. And sharpness, texture, details aren't everything - those the Vivo always gets right if the autofocus doesn't miss. However as I feel every update/upgrade changes quite a lot on the camera front - maybe after some weeks you will just not take those high risk, high reward shots anymore. Shots that you could take with a DSLR of 5-10 years ago - whith a simple conservate HDR and not what is happening right now - you can all take with the X90 Pro+ too and they will work. It's those shots that a good photographer would never have taken because they just would not work without crazy HDR - but that smartphones now make it possible to take.
e.g. face in the shadow, just take the picture and hope for the AI to fix it. Just click and let the AI figure a way to make a traditionally impoosible situation nice. That's where Apple, Iphone, Google (yeah especially google) found ways to still turn out great shots. No light on the face - no worry the AI just calculates it and the HDR puts in the contrast.