Pulse voltage will kill led. Led is not slow switch, you cant simple pwm it with bigger voltage like dc motor. Current will raise to enormous amount and led will die in 1/1000 sec.
then what if we add a cap to get rid of the spike?
You need to limit output voltage (direct drive), current (linear) or power (small buck). Otherwise your power supply can meet such point at led V/R curve when current will exceed usual limits.
PWM already sucks enough with our über sophisticated brains and eyes, let alone messing with it for night shots or filming in a world of ubiquitous photography and filming devices. Sucks like there's no tomorrow, imho. For simple low profile designs, maybe you could try switching to better linear drivers like the AMC7136, plus a simple big resistor in series for moon mode. Found something related, btw: Seeking AMC7136 - for a single sided driver design
Cheers ^:)
Hey, that’s cool Barkuti.
A 3 Ampere XQ7136, which can be regulated with a shunt (sense resistor) (if i understand correctly).
Very interesting.
edit:
It seems they’re made of unobtanium…
I know pwm is suck, but not that bad, with high frequency pwm, i bet 90–95 people can’t distinguish
With 32 bit cpu, i think we can push the PWM frequency upto 1Mhz
I know you guy don’t like the Blinking led, but, there are many things blinking around me, and you, we just accept it, and ignore it( your screen which you are using to read that line is blinking, too)
I think all linear, direct drive, pwm drivers must die.
It is like steam engine in 21th century.
What do you suggest that we use in their place? Got a design we can implement? Fully tested? Efficient? Powerful? Easily programmed to any UI we want?
That’s not how you compare
Wheel was invented for a very long time
Did we get rid of it. No, we just make it better
Meteor drivers do exist for over 3 years, components are not very modern, schemotechnics and theory is well-known for 20 or 30 years. But some “driver developers” (fet is not driver imo) were still sure that boost circuits will never work (until klarus realized G20).
In most simple version buck driver can be separated from mcu board, pwm management any constant modes (some fet developers are still sure that this can influence on output ripples, hah), thermal management can be realized with thermistor in parallel with current-setting R.
One board design with high-current curcuit require time that is enough to make 100 or 200 A6-style designs. Testing and bags searching will take extra hundreds of hours. Enormous work for amateurs. Dale provides idea that engineers are nothing, modders are everything. This doesnt work for buck or boost drivers.
My hope is that with the new low Vf emitters, and the new 21700 cells, we start to see lights based on those cells that have larger driver sizes. If we could use 22mm+ drivers, all of a sudden higher current buck and boost drivers become more of a possibility.
I do a lot of long exposure night photography and PWM has never been an issue for me. For video, or non-long exposure photography then it may be an issue.
I know several such drivers , buck and boost.
They are more complicated and require more knowledge. So people simply use old good “steam engine” instead of it.Fortunately the most of new powerful led are 6v-12v.So the only way is buck boost drivers.
Please list the drivers which exactly meet all of my criteria above and are not linear, direct drive, or PWM drivers. I’m very interested.
1)[[ GXB20 Driver – Homemade Constant Current Programmable XHP50 Single-Cell Boost Driver! ]] - Flashlight Modding and DIY Parts - BudgetLightForum.com
2)Драйвер для питания светодиодов CREE от 1 LiPo - V7M | Мастерская: МОНАРХ и AVSel
3)Indigo 5.0, или "не совсем форумный" драйвер? | Мастерская: INFERION
4)MiniEagle - Импульсный драйвер для фонарей с силовой кнопкой | Мастерская: Tamagotchi и Rime
For example.
Took a peek at the last one, at least for as far as browsing a few russian/spanish transgoogled pages. Buck-boost engine, 2.75-4.35V input range, lads speaking of triple parallel nichias. I guess it's geared for 1S emitter output voltage ranges (actual range?), Tamagotchi commented some sort of negative answer after someone asked about driving UV emitters.
Nice.
I believe you should move forward, step by step or whatever. Share and improve your designs. Once you get some good buck, boost and buck-boost designs freely available everything would go much smoother and with way less complaining. 
Cheers ^:)
P.S.: #1 link fix: [[ GXB20 Driver – Homemade Constant Current Programmable XHP50 Single-Cell Boost Driver! ]]
What did you smoke? There is no buck-boost.
Those appear to be great drivers, yes. But literally none of those drivers meets all the simple criteria I listed above. There are dozens of such drivers that are linear, direct drive, and/or PWM drive. Can you come up with even one driver that meets my criteria? I’m not trying to be rude, but I firmly believe in put up or shut up. You said linear, direct drive and PWM drivers are like steam engines and need to die. I’m just asking that you prove it! Come up with ONE direct replacement!
DavidEF, maybe AEDe sounded a tad radical. I believe what he/she wanted to mean is that he/she is irked for the fact there's a lot (cough…) of research in certain aspects of drivers/flashlights (UI, lighted tailcaps, built-in charging…) and so little in others like actual regulation, efficiency, proper battery voltage range utilization and protection. I sort of have to agree. 
Don't expect flashlight manufacturers to magically get involved in stuff if there are no good chances for profit, you may already know that, right? However, if you/we were to develop some nice buck and/or boost designs we could share its design with a manufacturer AND make a somewhat decent deal for an initial group funding/buy, research and development improvements could “hit the shelves”. Don't you did this with Banggood in the past, for example? 
Cheers ^:)