V2 - Different training set more of a 3/4 view
320x320 16 frames training 350 steps - Helper lora available, not always needed with some models
Fully loopable closed loop
Comes out at IMO 15 FPS
IP adaptor worked both masked and not, but non-masked cant be too strong to disrupt the horse.
Find a model that draws horses well to start
I used this for examples
https://civitai.com/models/8124/a-zovya-rpg-artist-tools
Helper lora uploaded if needed
Trained on 320x320 16 frame clip
Low strength 200 steps
High strength 300 steps
Using the test workflow
if you use low strength, you will get a closer up view of the horse/rider
if you use high strength, you will get a full shot of the horse/rider
Strengths can be 0.6-1.0
If you do not use the lora when upscaling, it will slow the run to a walk.
This can be looped by using looped uniform context options,
You can run 2 looped uniform context options in series, first is closed loop, then at 0.5 50% it changes to non-closed loop for a looped video
All examples used ip adaptor for upscale, and sometimes style transfer for base.
Models used for testing
Base
kawaiirealisticanime r5
gmcWhiteanime v3
Upscale
Photon
aniverse v2 pruned
Prompt used for examples
1girl, (wide angle, side view:1.2), a girl wearing a cowboy hat and blue jeans is riding a horse through a desert, white shirt, bare midriff, horse riding
1girl, (wide angle, side view:1.2), a female knight wearing (gold armor:1.2) riding a (large white unicorn:1.2) adorned with gold through a field of grass, horse riding
描述:
训练词语:
名称: animateDiffMotionLora_v10LowerStrength.safetensors
大小 (KB): 132075
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success