feature idea: FILM interpolation + cross attention prompt editing

[sbaier1]

Idea:

• Two prompts as input, second prompt is slight edit of the first (e.g. person changing pose, etc.)
• Generate first prompt txt2img
• Generate second prompt using img2imgalt / prompt editing using first image as input (not sure if this can have generally valid defaults)
• run FILM interpolation between the two
• run low denoising strength img2img on the interpolation frames to refine them(maybe? if necessary / improves the result)

optionally: use attention weight to interpolate between the prompts, generate more than 2 frames, interpolate between those. not sure, experimentation needed