Congratulations on learning ComfyUI. It's worth it.

There are several ways to constrain the annoying randomness. It sounds like you need to learn (if your laptop can take it) about ComfyUI's Controlnets (Canny etc) and perhaps also Liveportrait (change eye gaze direction).  

Poser can output good line-art that can be used with a Controlnet to constrain the AI generation, so that it more closely matches what you have in Poser. Liveportrait in Comfy means you no longer have to fuss about eye gaze direction in Poser.

For 2026... Poser is Python, and so is ComfyUI, and they're both very mature ecosystems in what they can do. They can theoretically talk to each other in a stable way. Such as auto rendering a set of named files, that then get picked up and auto-slotted into a ComfyUI workflow and run. With the result returned back to a panel inside the Poser UI.

What AI can't yet do just from a prompt is complex ensemble action scenes, with the sort of precision that requires. But, as you've discovered, it can also do relatively subtle Img2Img makeovers for a single dressed figure. It can also act more like a Photoshop filter, using what are called 'Edit' models such as Flux Kontext. These are dedicated to image editing and filtering rather than to image generation. 

---

Learn the Secrets of Poser 11 and Line-art Filters.