I've noticed this too in my image generation with qwen. Something is definitely going on, I can't get anything to render correctly and there is almost all noise and this is using just a default Qwen IMage Edit template. Something broke.

Make your own workflow, or try using wan through Pinokio

@Niboo your workflow is embedded in the video (I used the first one) and it exactly the standard one supplied by ComfyUI in its template, with the only change being adding your lora.
I also tried the default one without loras and it's the same: I get a video, but it only follows the prompt, the image is completely ignored... and this happens on 2 different computers, one with a RTX 3070 and one that's a Ryzen AI 370... I tried redownloading VAEs, removing the alpha from the image, running the VAE on CPU, removing strange node packs I had installed, forcing the VAE to be FP16 and probably a few more things... still I get videos totally unrelated with the input images :-/

... Ok, I'm a total idiot: I selected a t2v wan model... 🤪🤪🤪

@TooLittleVRAM easily done!

@Niboo I use the same prompt than in the preview video, i even added a lot of more detailed, it never worked.
Notice: in my first video scene, there are two girls, full body visible, as one is kneeling doing a blowjob to the other one.
I wanted to use your lora for my second scene, the kneeled girl should stand up, turns around, and the futanari inserts her penis into her ass.
It never worked., even with strength at 1.5 for H. but at 1.5, this is the only test where the girl starts to turns around, but not really. And at the end of each tests, the futanari inserts her penis in the pussy instead of the ass, because the girl who was kneeled, don't turn around.

So, i can say with my multiple tests, this don't work when the main character who is kneeled, the full bod is visible.
I don"t tried with a first video scene where only the upper body of the kneeled girl is visible. As for my adult scenario video, the full body of both girls (girl and futa) must be visible in the first scene, because i use another Lora for this scene.

@Niboo this is the first thing i tried, hehehe!!!
Making the girls stands up and turn around with or without any lora.
But, if she can stand up without any problem each time. I never succeed making her turn around so she shows her back to the futanari who is standing next to her.

@Pat3dx What checkpoint were you using? Some are liable to bias with actions/features over others. Consider WAN Video 2.2 Remix instead of others like Smooth Mix and see if that works.