In short, no.
I think it's still a long way to go to finally leave SD 1.5. There are at least two reasons people stay in SD 1.5-based models.
GPU Power
Not everyone has a super-fancy GPU to run an SDXL-based model locally. Even I, a model trainer, use RTX 2060 with 12GB of VRAM to train my SD 1.5 models and use Civitai's LoRA trainer for SDXL. SDXL is still very heavy to handle locally.
Overall generation
SDXL-based models can generate higher resolution without a hires fix or an upscaler for sure. But for most of the Anime SDXL models that are currently available, I think most of them haven't reached the ideal generations.
Take a look at these:
AingDiffusion XL is slightly more neutral than the other two.
Animagine XL generation is looking more like sketch/illustration than anime.
AAM XL is looking more like 2.5D or semi-realistic. Need some extra negative prompts to make it less 2.5D.
Then, take a look at these:
These three SD 1.5 models have something in common. Anime (Japanese Cartoon/Animation) looks by default. ElixirProject and AingDiffusion looked like the current ones, while ERA looked like retro ones.
SDXL models couldn't reach this kind of style at the moment, so I think we still need time to fine-tune SDXL to generate anime-style images.
Why training an SD 1.5-based model is still worth it?
My friend said, "There is nothing to do anymore with SD 1.5". I completely disagree with this statement. With new schedulers, tools, and images that didn't exist yet when the model was first released, there is more thing to do with SD 1.5-based models. I recently just moved the scheduler I use from AdamW8bit to Prodigy, and of course, Prodigy works better and more efficiently than AdamW8bit and the model also produces better pictures. As long as supporting tools for training SD 1.5 are still updated and supported, the SD 1.5-based model will likely still be worth it to train.
And that's what I'm currently doing.