I am by no means an expert and would love any feedback or additional thoughts. I have spent a bit of time tinkering with Animate Diff workflows specifically for morphs and transformation GIF's and animations.

Shoutout to a1lazydog for the excellent workflow that set me down this path. (Animatediff Workflow: Openpose Keyframing in ComfyUI - 3 keyframes v1.1 | Stable Diffusion Workflows | Civitai)

I don't use all nodes at all times. Latent upscale really helps smooth motion. Distant faces get warped even with upscale and ReActor seems to make the faces look nice. I think there is some refining to do here!

I am using ControlNet Tile for images. I will use Photoshop to make backgrounds similar between them. I have found that low ControlNet weights give the Txt2Vid room to breathe but often play with the strength weights.