Newer Guide/Workflow Available https://civitai.com/articles/2379
Using AnimateDiff makes things much simpler to do conversions with a fewer drawbacks. The major one is that currently you can only make 16 frames at a time and it is not easy to guide AnimateDiff to make a certain start frame. My workflow stitches these together. It also tries to guide the next generation using overlaping frames from the previous one. I expect that in the next while we will have further improvments on this.
How to use:
1/Split your video into frames and reduce to the FPS desired (I like going for a rate of about 12 FPS)
2/Run the step 1 Workflow ONCE - all you need to change is put in where the original frames are and the dimensions of the output that you wish to have. (for 12 gb VRAM Max is about 720p resolution). [If for some reasons you want to run somthing that is less that 16 frames long all you need is this part of the workflow]
3/Run the step 2 Workflow as many times as you need- you need to input the location of the original frames and the dimensions as before. You also need to go to the comfy output and find the blendframes folder and input the location of that in here too. This will take 12 frame blocks and run and then combine them so you will hit run (or batch run) how many times you need to run all your frames. You need 4 frames at the end of the last batch of 12 so you will need to add these if you do not have enough frames (and delete them at the end if you wish). If you accidentally hit prompt too many times it will just give an error and not run when you hit the max. Wait for this to finish.
(ie. for something that is 124 frames long you will run step 1 once and then run step 2 9 times - if you only had 119 frames you would copy the last frame 5 times in order to ensure you had 124 frames if you wanted it all rendered otherwise it will stop after the 112th frame)
4/The last 4 frames end up in the blend frames folder you can choose to put them back into the output folder
5/You have you completed conversion - put the frames back together however you choose.
Things to change:
I have preset parameters but feel free to change what you want. The model and denoise strength on the KSampler make a lot of difference. You can add/remove control nets or change the strength of them. You can add IP adapter. Also consider changing model you use for animatediff - it makes some difference too.