Default:
Checkpoint: DarkSushi.
Sampling method: DDIM.
Steps: 20.
Size: 512x512.
CFG Scale: 7.
Seed: 2834601277.
Prompt:
1girl, brown hair, grey eyes, short hair, (elf:0.91), office outfit, white shirt, outdoors, portrait, <lora:DetailedEyes:0.6> beautiful detailed eyes, (masterpiece, best quality:1.21)
Negative prompt:
(worst quality, low quality:1.21)
Result:
AnimateDiff:
Motion module: mm_sd_v15_v2.
Number of frames: 16.
Context batch size: 16.
FPS: 8.
The length of a gif image is equal to "Number of frames" divided by "FPS" (16 / 8 = 2 sec).
Closed loop:
R-P: The prompt travel will not be interpolated to be a closed loop (result).
R+P: The prompt travel will be interpolated to be a closed loop (result).
A: Aggressively try to make the last frame the same as the first frame (result).
R+P work best if you set Number of frames more or equal then Context batch size * 2 (if Context batch size = 16 then Number of frames = 32)
Frame Interpolation:
Interp X: Replace each input frame with X interpolated output frames.
Closed loop: A, Frame Interpolation: FILM, Interp X: 4 => result.
Other:
Motion module (link)
Motion LoRA (link)
ControlNet (link - will appear after creating a post)
Prompt Travel (link)
This is my first post. I will be glad to know if you are interested in this topic and any suggestions/wishes regarding the quality of the content.