New masking & scheduling workflow issues (very slow, 100% computer resources usage and no results)

[ColonelJay]

Hi all,
I am currently trying the new masking & scheduling workflows: link

Those new features looks awesome but unfortunately, it's not working for me.
The issue I have is when the workflow has reached the KSampler node, all my computer and 4070 resources are used at 100%, only to get 30+ minutes later an almost blank picture (see attached pictures).
I had to take a picture of my screen because with everything at 100% usage, it is impossible to take a proper screenshot.
I ran the workflow with flux_dev-fp8.safetensors as model.

After a first try, and following @uberneko's recommendations - here : ComfyUI discussion - I tried to reduce some image generation parameters (downsized to 512x512) and adjust some sampler parameters (steps down to 20, CFG to 1 and denoise to 0.75) but with no differences.
Also, sometime, the workflow just stop with a red connexion lost message in the top right corner.

ComfyUI was updated yesterday.
My computer has 32 Gb of RAM and I have a 4070 with 12 Gb of VRAM.
With the same prompt and a basic txt2img workflow, I can get some decent results in about 30 seconds.

Thank you in advance for any hints about this issue.

comfyui.log

[uberneko]

Hello !
Actually I see you're trying the exact workflow from the link -
I did modify it as I want two LoRAs (two characters). So the one on the bottom I turned it into a "Create Hook LoRA" like the top one.

Actually, I tried other combinations today, and all were unsuccessful : the process is stopped, it just says "Line 11 - process stopped - python3 main.py". Ejected from ComfyUI...

Strangely the one I was successful with yesterday do still work... so the only thing I've done is change the prompt and the LoRAs...

Another thing I noticed is LoRAs created with FluxGym work better (less time).

(I did update ComfyUI this morning).