Supporting partial image sampling like in img2img

[reesekneeland]

In the original stable diffusion and other diffusion models, there is a feature of the image sampler to give an original base image (used as the latent Z vector), and a strength parameter (between 0 and 1), and a new prompt, which then instructs the model to add noise to the Z vector in accordance with the strength parameter, and denoise it conditioned on the clip of the given prompt.

I was hoping this model would have similar functionality but with the ability to perform that last denoising step conditioned on both the image clip AND the text clip, so you would give it a base image, which would be converted to Z and img_clip, a prompt, and a strength parameter, and it would add noise to Z, and denoise it conditioned on both img_clip and prompt_clip.

Is a feature like this possible given the current noise scheduling system?