Someone wanted me to do this, so here's a LoRA trained on SRW J and W's character portraits.
You must generate at 1024x1024 and use something like sd-webui-pixelart to downsample by 21. (48x48 pixels). Reduce palette to 64 colors or less.
J's portraits are actually 44x44 and W's 48x46, so you'll need to crop a few pixels to make it fully authentic.
Trigger words: srwjwport, solo, portrait, close-up, looking at viewer
You'll probably need high strength (1.5) for the LoRA to actually take effect. Character LoRAs will definitely fight against it.