Concept - walking to viewer with matching background zoom
Its not perfect but it does what i wanted mostly
Steps 150 low 175 high 320x240 trained think it works on other res
FPS output recommended 10-15fps depending on how fast you want it and easier frame interpolation to 30
Tested only 32-64 frames, not sure if it does others
Looped Uniform works good, closed loop might mess things up not sure.
2 stride for 32 frames, 4 overlap works best
Strength 0.5 - 1 Just tested 0.6 and it helped remove the damn skip
Example Prompt
1girl, (full body, full shot:1.2), little red riding hood walking down a forest path, walking towards viewer
This does a fairly good job keeping the background scrolling with the walk, the only issue I have is a small skip that shows up sometimes to interrupt an otherwise very smooth walk.
Looping this works well, closed loop can cause the skip to show up more strong.
Lowering strength to 0.8-0.9 can help remove that skip but can cause other problems.
IP adaptor/ lora works if you dont go overboard on strength.
I think it can do a walk away from viewer as well but not much testing there.
Listens to prompts fairly well, but I don't do extreme prompts. Tagger tags worked mostly as well as long as you keep an eye on the tags.
Description
FAQ
Comments (3)
Hi. Hope someone can help. I have added extension in A111. Downloaded motion modules and put them in WebUI\stable-diffusion-webui\extensions\sd-webui-animatediff\model. I have clicked AnimateDiff drop down, loaded a motion module and enabled AnimateDiff - even on very low frame # and FPS - all I am getting is a PNG image. What am I missing - big thanks to anyone who can help.
think im doing something wrong, the motion wont work, keep geting a normal img =/, puted it in to the right map
it works fine with the author's settings, which checkpoint model did you use? A realistic one is connected to my base model and gives a blurry result. BTW "closed Loop" is messing up the result, i get it closed already without this setting.