@epinikion agree that, many people at the crossroads and hesitate, so many factors, especially the cost of time and hardware

@epinikion Guess what, there's apparently a 1.6 now, so... the peak just got higher!

@Aptronym Yes. SDXL is still in its infancy so what epi might be saying could be SD 1.6? I still use 1.5 but I refuse to update since i like the results i get already. Plus everything just works on 1.5. A man once said: If it broke don't fix it, but i do want to try SDXL/new samplers lol

@epinikion I've seen no need to move over to SDXL when 1.5 can get this good with a quarter of the size. I understand this model may be at the peak, but I meant your era of any other work/model you may have in the back burner or thinking about in general (concepts maybe? hmmm)

Of course, don't burn out for internet strangers, but we do enjoy your work and look forward to it if there's more to come in any other form aside this model :)

@epinikion It is available on the Stability AI API page so I would expect the actual model to come out soon.
https://platform.stability.ai/pricing

It's probably reference to unicorn dream from classic bladerunner

@SharpArrow Not sure if you're making a joke 😄 but I'm pretty sure it's a reference to the book and/or movie "The Last Unicorn" which are about the last unicorn in the world. Implying that this may be the last iteration of this SD1.5 model. 🥲

@epzer or it implies he's gae

Same question

vae-ft-mse-840000-ema-pruned

@nrofis Thanks

You can try the Negfeet negative embedding https://civitai.com/models/90707/negfeet-improve-feet-quality but the success rate is quite low from a quick test run.

wtf dude lol

@Asbarka13 hey man, whatever floats your boat.

😁

I get decent results combining my FAPP lora https://civitai.com/models/129350?modelVersionId=146638 with Tgirls_V4. Set both to 0.3-0.5

Agree. Join the club - it's a bit confusing :) Version numbers are clearer I think. They all pretty much do the same thing very well, it seems. Happy to be corrected (politely).

I'm using LastUnicorn now and it's great. Someone commented it does hands and feet better - haven't got the time to properly test all the model variants so not 100% sure. Brilliant models but a bit more explanation would really help :)

it sais right there when you click ''about this version'' under the model info.

And what the difference between X and Y ?

@sevilleianproduc2250 Hands are really good indeed! Eyes are hit or miss though

You can use Photoshop Generative AI to fix hands. Use the lasso tool, select a hand. In the "Generate" field, type "fix hand.".... voila.

*I might add, it may take a few tries to get it right.

@EryxAtys thanks mate, do you have more little tipps like these?

@EryxAtys True, I'll do that. thanks.

Is it important to have many sampling steps when using refiner?

@goodly666521 No actually, I believe if you are using it as a refiner for XL just use 4 steps for 1.5. And use standard Refiner for SDXL.

Humm. I gave up on SDXL because it just doesn't do NSFW realistic like 1.5 does. Maybe I'll give it a try again. Can you provide more details? what base model would be good to use? I have some good results of SFW images using Model: nightvisionXLPhotorealisticPortrait as a base SDXL with Refiner: realisticStockPhoto_v10, Refiner switch at: 0.1, Steps: 40, Sampler: DPM++ 2M Karras, CFG scale: 5, ADetailer model: mediapipe_face_full, ...but still cannot get a decent nipple to for the love of god!

@djwashout If you want anything special like nipples or weird looks like dissolving people or ink styles, you'll wanna go with a LoRA, which are awesome I've come to find... just recently in fact. As far as refiner, I should have been more specific that I didn't know about at the time. I believe it may be a Fooocus exclusive feature where you'll drop the 1.5 refiner to .4, and it'll really help. I'm not quite sure if it's a help in A1111 or Comfy, as I don't really use refiners in those, considering I use them for more experimental purposes. Testing the new stuff, for example. For the record, I'm awful with Comfy. Best of luck in getting your quality images.

I normally use 5 with this model.

Hi, Im new in SD, but I also faced a lot of issues with quality. If you hape problems with faces then you shoud enable After Detailer (ADetailer section), it is "must have", especiaaly for full body shots. Also pay attentinon to Clip Skip setting, it is 1 by default but i see that usually it shoud be 2. It could be found it global settings in Automatic 1111. Also people use upscalers before posting something to public, nobody posts 512px blurry images. Upscalers may have signnificant impact and change the image look, it depends on upscaler settings and type. Just google this things, there are some short youtube tutorials for every thing I mentioned.