Waving (Frame helper LORA experiment)版本waveleft (ID: 223579)

Waving (Frame helper LORA experiment)版本waveleft (ID: 223579)

What's good little momma? You missed me didn't you?

if you aren't going to give me money at least give me all your buzz
Ko-fi: https://ko-fi.com/keyennorman

Hello, throughout my machine learning journey I have encountered some pretty daunting push backs. Mainly the technological requirements to dive into the funner aspects of AI. This LORAs purpose (Hopefully one of many soon) will be to give people like me with average tech the ability to try (and hopefully succeed) in creating a simple gifs and poses without meeting some of these high end hardware requirements.

In my case

  • I cannot run ControlNet for poses over about 300/300 pixels

  • I cannot generate 4 frames on 2 frames a second with animated diff with a motion model

  • Cannot run 3 or 4 LORAs at the same time

So, this concept consists of 3 LORAs:

1. waveslightleft

2. waveleft

3. wavefarleft

These LORAs each represent a frame or a stage of waving. So far left handed is only supported but from what I can gather the AI doesn't care, it could do one hand then do the other any time it wants.

So all in all with a little bit of manual labor you could essentially get the same if not better results. The steps should look something like this

Step 1: Choose a stage to start with:

It really doesn't matter, but I recommend starting with the regular wave left as it is dead in the middle and has a chance of working somewhat well with img2img

Step 2: Evolve the waving with img2img:

Now is the generation process. Generate till you have the image(s) you want using all the LORAs.

Step 3: Now that you have all the images you want, put them all together in an editing software.

They can be as many frames you want etc. Now I think it's about time we dive into some prerequisites

1. Weights fluctuate, with img2img it should be pretty low like a 0.40 regular generation start with the stage you want at 0.6-0.8, the wight really depends on the model you're using.

2. Slightly left may change the image's lighting I don't know why something may have to do with the baking.

3. some tags to help: ((waving left or waving far left or waving slightly left)) one arm raised, waving, 5 fingers,

4. Might struggle with character or style LORAs.
5. Activation Tags:
waveleft = "waving left"
waveslightleft = "waving slightly left"

wavefarleft = "waving far left"

The purpose of this LORA(s) is to help frame by frame videos for people with hardware limitations and without waiting 30 minutes for a subpar result. If the size is a bit too large just be glad it's not 144mb or whatever people are doing now a days. I know I could probably condense this into a single LORA but for the sake of testing purposes and the fact I don't necessarily know how to go about that these have been split. Maybe when I work on more in the future I'll figure it out.

描述:

combined about 80 image dataset

训练词语: waveleft

名称: waveleft.safetensors

大小 (KB): 36984

类型: Model

Pickle 扫描结果: Success

Pickle 扫描信息: No Pickle imports

病毒扫描结果: Success

Waving (Frame helper LORA experiment)

Waving (Frame helper LORA experiment)

Waving (Frame helper LORA experiment)

Waving (Frame helper LORA experiment)

资源下载
下载价格VIP专享
仅限VIP下载升级VIP
犹豫不决让我们错失一次又一次机会!!!
原文链接:https://1111down.com/994004.html,转载请注明出处
由于网站升级,部分用户密码全部设置为111111,登入后自己修改, 并且VIP等级提升一级(包月提升至包季,包季提升到包年 包年提升至永久)
没有账号?注册  忘记密码?

社交账号快速登录