Supposedly A1111 weighting is "wrong"

[WASasquatch]

According to comfyanonymous A1111 weighting is "broken" and "wrong" and has implemented InvokeAI type weighting, but through every examples shown he doesn't get how even low weighting causes latent artifacts, and hard contrasting, and it doesn't respect DDIM samplers by retaining the whole image outside what you are weighting. Every level of weighting substantially changes the entire image, unlike in A1111 where what you are weighting is substantially changing while rest stays the same more or less until approaching high levels.

It seems to me the ComfyUI's weighting is broken. It's not deterministic with DDIM samplers, making it essentially useless for editing your original diffusions by seed, and also effectively useless for frame editing, even with finding latent noise of the frames.

I notice in the A1111 code it talks about how it might not be the best way to do the weighting, but prevents heavy anomalies from showing in the image. It seems this is the source of all the issues. But when it comes to a art driven software, isn't ensuring fidelity and quality more important like what A1111 did, then what they think is better on a strictly math side? It seems obvious normalizing the prompt is preserving the integrity of it over what's being weighted.

Have you used ComfyUI or InvokeAI? Have you used it for processes animation frames? How do you get by something as simple as a 1.01 increment in weighting starting to impact the rest of the image outside of your target?