👑 AngrA RealFlex close to real👑

@Quiron Thanks !

@Quiron you can add a pruned version to this model if you want

@Quiron This is an exceptional checkpoint, I hope you could compress this. Anyway, thanks in advance.

I need this

As long as the 16fp version isn't listed here, you can do it yourself in A1111, use the extension https://github.com/Akegarasu/sd-webui-model-converter
(and optionally https://github.com/arenasys/stable-diffusion-webui-model-toolkit to see a model's internal stats)
In the extension load the full version in,
Settings:
name: angraFlex_pruned, precision: fp16, prune: ema-only, format: safetensors, extra options: enabled, unet: convert, encoder: convert, vae: convert, others: delete
It will take take a bit of RAM/SWAP and about 10 seconds to convert, but when it's done, you can just delete the full model, and use the converted result. :)

@LongJohnSingleDuct unfortunately this didn't work for me - using Akegarasu and the settings you advised. It just produces a garbled image. I tried on some other large models I have and worked fine with those though.

Edit: Tried the other one just using the basic tab and it looks to have worked.

@jonk999 I think I was running a version with a bug at the time of writing. Getting images of solid colour sounds like running the EMA (training data surplus) version, so in that case it should've been prune: no-ema rather than ema-only. But if you managed to have it work (and still shrink), that's all that matters! :D

@Quiron So, no chance to get the smaller 2gb pruned-fp16 version ? It would be a good service to the many users with hw limitations, and 2gb is already the standard size for the vast majority of the published models (sometimes together with the full bigger version).

Didn't want to make a very prominent complaint on this, but I also wasn't able to figure out pruning, which is a shame because this model looks decent. There has just been an update which looks great, but I am definitely not collecting these file sizes in 1.5 these days. XL models blow all of these out of the water, but I can't do anything in the background on my machine while making batches, so 1.5 is still nice to mess around with. The fp16 versions use less memory and processing, not to mention storage, so a model like this is a non-starter for me. When I looked at the gens, I wondered why this wasn't in my collection, then I saw the size, and this inevitable comment thread. Too bad.