AnE v5 is out,
Featuring a semi-realistic art style, the newest edition in my Anything and Everything series is here!
Keeping with the style of my previous versions, this one has been remixed from the bottom up - removing some subpar models and adding some new ones! Just don't ask what, because I didn't write it down. I tried to include a wide variety of example pics, many of them do use a LoRA for their gen.
Description
FAQ
Comments (22)
you know you can put all the versions on the same page, right?
Do you think that'd be better overall? I was typically separating them if I though they were significantly different enough.
I mean, you've given them the same name; that alone says a lot. If you intend them all to be used as opposed to higher versions being upgrades, then they should really have more distinct names, otherwise a lot of people will just assume that higher is more recent and also better, and ignore everything else. In that case, instead of numbering, give each one a distinct subtitle denoting its style and maintain separate internal versioning for each one.
@nashidaran That's certainly a fair point. I'll keep that in mind for future uploads. Thanks for the suggestion!
What is AnE v3 - (WS.20-Kenshi) - (AD.30-Pastel-Kenshi)? A different model then the one used?
((Anything and Everything version 3 + [email protected] Weighted Sum)[email protected] Add Difference)
That's v4. It was the working name before I finalized and uploaded it.
Do I need an specific VAE? I'm not getting nohting as your examples :(
I don't think the specific vae matters too much, though the one I'm using is 'vae-ft-mse-840000-ema-pruned' (not sure where to get it). Otherwise, make sure your webui is up to date. If you can, upload one of your gens from my examples with the exif data, and I'll take a peek and see if I notice anything. Also make sure you've got any embeds or LoRA that I'd used. I know at least many of the gens had some negative embeds, some with one or two having LoRAs
@actually_anonymous Here's an example: https://imgur.com/a/eejnC3K
@SmartNeto The image doesn't have quite the full exif, but the first thing I'm noticing is that that's a square image, which none of the ones I'd uploaded are. To try and recreate my images, did you just copy and paste the prompts as civitai shows them, or did you use pnginfo on the webui with the image to get all the info? You'll want to do the latter to get a more accurate recreation.
@SmartNeto Playing around a bit with my settings, I was able to make something a bit more like your image using an old copy of my config from months ago, so I may have changed something in my settings - but comparing the 2 configs, I can't tell what. You can try my settings with this (https://files.catbox.moe/d5810o.json), just rename it to config.json and put it in your webui directory (backup your old version first). Be warned there's a good change a fair few setting will need to be changed for your machine, because I have them setup specifically for mine.
I am unable to recreate your results, for example the second one where you used to Lora.
I git pulled, got your config, installed your used upscaler and even got the VAE (without it all images weren't that great and were too unsaturated/gray like).
Here's my result: https://imgur.com/a/cxd38gu
I'm a bit of a beginner to stable diffusion, is there anything I might've done wrong?
@natalliastopka90351737 Hmm, one thing I can think of is that a different GPU can actually cause differing results. Second, command line args in the start .bat can cause differences (I have 'set COMMANDLINE_ARGS=--xformers --no-half-vae --deepdanbooru --api'). 3rd, the upscaler that I'm using is '4x anime sharp' which doesn't come naturally with the webui (it can be found here: https://upscale.wiki/wiki/Model_Database). Would you be able to upload your image to https://catbox.moe/ , that way I can see the exif data for myself and see if there's any difference (or just make a new comment with the image, metadata attached)?
@actually_anonymous I did your steps, here's the image https://files.catbox.moe/k7b27b.jpg
@natalliastopka90351737 I'll be honest, I'm at a loss. GPU differences can definitely contribute, but not to this degree. Can you recreate images from other models, or is it just mine that turn out poorly? Regardless, sorry I couldn't be of more help.
@actually_anonymous I actually don't know why is this other guy asking in this comment, I just changed the image size and it worked for me
@actually_anonymous how to install "the upscaler '4x anime sharp'"
can you install custom upscale "4x-AnimeSharp" to webui ? what are the steps or any guide link ?
The custom Upscale is usually in the ESRGAN or GFPGAN.
what is the width and height normal used for this model ? 700x700?
Start with 512x768 with Upscale by 1.25 or divide by 2 with upscale by 2 times.