New Version.

On the LoRA information (link further down), they mentioned this would work with SD3.5 Large and not need a LoRA so I figured I'd try the starting prompt with Flux, and it worked better than I hoped for. So no LoRAs needed, but you can use other LoRAs depending on what look you're going for. Also means only one model needs to be loaded which is always a bonus.

As well as changing the first generation stage from SD3.5M to Flux, I've added DeepSeek to build the prompt, it's already instructed to use the starting prompt in the System instructions, so you only need to type what sort of theme you want in the Prompt box: i.e. glacial blue ice, snowflakes, diamonds, and DeepSeek will make the prompt. I've also included PAG Guidance node, DetailDaemon and Multiply Sigmas (Stateless) as their own separate groups to the generation section and same again for DetaiDaemon and Multiply Sigmas for the Uspcaled Refiner section. There's a Fast Group Bypasser for these specific groups, so you can turn then off and on easily. I wouldn't recommend running PAG, Detail Daemon and Multiply Sigmas all together as I found the quality decreased in the final image.

The DeepSeek prompter node doesn't appear in ComfyUI manager for me and I had to Git clone it into my Custom Nodes folder from here:  https://github.com/comfyuiblog/deepseek_prompt_generator_comfyui

It's an Ollama generate node that automatically filters out the thinking part of the DeepSeek ouput.

---------------------------------------------------------------------------------------------------------------------

I play D&D on Roll20, so was looking at ways to generate borders for characters and came across this SD3.5M LoRA:  https://civitai.com/models/948412/vtt-token-rings-35

I liked the results but felt they could be better.

So after some work came up with this.

SD3.5M for the initial generation, it's then upscaled to 2000 x 2000 pixels and a 0.5 Denoise with Flux to refine it. Final step is REMBG 2.0 to remove the background and make it transparent. Just pop your prompt in and hit generate.

I've left in one of the prompts, started it the same as the examples on the LoRA page "a circular tokenframe on a white background with an empty white centre" I suggest not changing this and only editing the prompt after that part.