The really short recipe is: base image => image edit for key frame => flf2vid => manual post process.
Now in details:
To get a realistic base image I used zImage turbo (macro shot of a detailed red eye): no sharingan, just a natural eye with red iris.
For key frame I used the base image and image-edit worflow/model to create different eyes (sharingan forms). For that I used flux2 klein 9b fp8: cfg4 and step 4 to 6 work best (sometimes higher step work too ~20). The reference images used for the edit are just a red circle with clear black marks/pattern for the sharingan.
Now all different sharingan frames need animation to stich them. for that I used wan2.1 custom model with FirsLastFrame2video using first and last frame or sometime only with the first frame (without the last frame the eye move/wonder in another way that is natural too). I created clip between 9 to 21 frame with wan and then interpolated with rife vfi to get near 30 FPS (avoid model sampling/shift to lower camera movement).
For the post processing, It's important to check color of the frame generated and correct if need ("color match" kjnode with mgvd methode and 0.8 strength, reference on last clip). Also it's good to trim/delete a few frame at the begining/end of clip to keep the final video more fluid (not perfect).
Improvement: should have saved all keyframe, all flf2vid clip, and then visualise the final video. a part of the video is glitchy but I don't have the key frame/clip and I dont want to restart from scratch so...
Some clips in this post: https://civitai.com/posts/26047139
Some images in this post: https://civitai.com/posts/26046935
final result: https://civitai.com/posts/26047773