Fortunately and thankfully yes.

There's lots of use for SD1.5 models still:

- spam out a bunch of pics to choose a good one for img2img w another model

- ControlNet your stuff in a fraction of the time

- create poses for ControlNet to use with other models

- some niche interests are still served better by SD1.5; for example lots of homoerotic content is still arguably better made with some of the SD1.5 models

Honestly - I'm still using SD1.5 for the speed a lot of the time myself. Plus - n this is nuts - I've noticed many LoRA I've made on Pony or SDXL work better on SD1.5 when the dataset is retrained for SD1.5 n I honestly don't know why that is 🤷

Sure - you give up the ability to do a lot with tags in Pony etc but, especially as part of a workflow, SD1.5 is still useful af

@digitalghosts Agreed 100%. I was thinking about the 30+ reasons I prefer SD1.5 over SDXL, or even Flux, but outside of speed (I have a 4090), you pretty much summed me up. And no, it's not nuts, Most of my LoRAs look and perform much better on SD1.5 as well. Even on Civ, most of the SD1.5 LoRAs produced have better likeness than their SDXL and Flux counterparts, even when trained by the same authors. I've tried and trained on them all and I'm still perfectly happy with SD1.5. I don't see that changing anytime soon.

SD1.5 is the only model my computer can handle

SD 1.5 is where I can do ANYTHING. There is no limit to the base model, over the years the models have been honed to near perfection. Here I can mix almost anything with anything, there are many Lora, there really are no limits.

The Flux and SDXL are beautiful and rich in detail, if they are more resource intensive, the base models are NSFW stupid, and it will take several more months to bring them up to normal. I like Pony too, but there realistic models are not stable, so there is no good base model to train Loras on, because they are not consistent, meaning either Pony capability is reduced or realistic quality is.

So SD 1.5 has everything, you can do anything.

SD1.5 models are able to run on 8gig Vram cards with decent channel speeds (limitations in combination of batch sizes, image size or steps to any number of combinations, but still able to be done) and while potentially slower and intensive, they are able to be run limitedly via CPU. AKA not everyone has the money to be going out and buying Nvidia 3080/4070+ or AMD 6-7 XT series (and higher level there in). Plus most cards that have Vram over eight gigs tend to bottleneck in performance when running at 1080p. Providing less raw power due to in part being made for higher native resolutions. This performance bottleneck runs across the board, not just in games or visual programs. And you'd be surprised at how many people still use 1080p as their system/monitor resolution.

@Bit_Shifter 100%

SD 1.5 only used for porn nowadays ... but SDXL (pony) is far more better for it ...

@Meravokas Question is for what you are using 1.5 model ... I assume 99.9999% for porn only (where is pony sdxl far better)

:)

@mirek190 is good by default, that's true, but there is no realistic Pony model yet that can be trained to be a good Lora. All models are unstable. Some lack this, some lack that, and Lora trained on one Pony model is not always good on another Pony model. For SD 1.5, I have everything, and indeed, I prefer the knowledge of the Pony, but I'm still waiting for a stable realistic model with the knowledge of the Pony.

@mykeehu I use pony for composition generation and a mix I made from jugernot babes dreamshaper and a few others for the high res fix. As long as you put SD-1.5 loras that show everything in the generation pic, into the high res prompt you get pony style composition with sd 1.5 realism.

@saftle I can't wait for it to be released, and I hope it can be used to train Lora models, because then it would replace SD 1.5.

Yes, but some of us forget to release any of their workings and look at their pile of unreleased on huggingface and screem XD