Sign In

Stable Diffusion/Animatediff/ADetailer /People consistency : Generate video with the same actors

17

Objective

Generate many video with the same actors, maintaining the face consistency.

Result

Check out the result : https://civitai.com/images/5442376

Install all the necessary

You need

-- To install the necessary tools there are many articles online. --

As actor I use the greats models made by @tower13Studios (https://civitai.com/user/Tower13Studios)

To install the Girls in Stable Diffusion I use the Civitai plugin (
https://github.com/sunnyark/civitai-shortcut.git)

1) Click on the civitai tab

2) paste the url of the girl

3) click Download

4) Click again download .

Installation will be very fast (only few Kb) and will work without immediately.

If you do it manually, You can install the files under the folder StableDiffusion\stable-diffusion-webui\embeddings

Write the prompt

Positive prompt:

2 girls (Art Photography, Award Winning) DV_Kira_Vicious (dark blonde Hair :1.4) (on the bank of a mountain river with bikini silk)(skinny body, fit body, slim, abs:1.3) (kissing:1.5) DV_Ingrid_vicious <lora:people_count_slider_v1:2>

Negative prompt:

WATERMARK (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, Nrealfixer, nfixer, nartfixer

Here all the settings of the picture below.

2 girls (Art Photography, Award Winning) DV_Kira_Vicious (dark blonde Hair :1.4) (on the bank of a mountain river with bikini silk)(skinny body, fit body, slim, abs:1.3) (kissing:1.5) DV_Ingrid_vicious <lora:people_count_slider_v1:2>
Negative prompt: WATERMARK (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, Nrealfixer, nfixer, nartfixer
Steps: 35, Sampler: Euler a, CFG scale: 4.5, Seed: 401434481, Face restoration: GFPGAN, Size: 768x512, Model hash: e73d775ff9, Model: theTrualityEngine_trualityENGINEPRO, ENSD: 31337, Token merging ratio: 0.1, RNG: CPU, Lora hashes: "people_count_slider_v1: 4ef0b7737563", TI hashes: "DV_Kira_Vicious: 87bba1b0acb7, DV_Ingrid_Vicious: 5fc1955a9911", Eta: 0.5, Version: v1.7.0

You can notice that the faces are not correct.

To improve the faces Enable "ADDetailer"

In the 1st tab set the name of Kira DV_Kira_Vicious

Set that the system will check 2 faces and replace them. (He will replace both faces with Kira face)

Now in the second tab write DV_Ingrid_Vicious and set 1. He will change only 1 face with Ingrid.

In this case one face will remain with Kira and one will be modified with Ingrid.

Here is the result:

Now you can generate the "random" videos with animateDiff.

Set the seed to -1

Enable the "AnimateDiff" , set the motion module that you prefer: I use normally mm_sd_v15_v2.ckpt or mm_sd_v15_v3.safetensors.

Set the MP4 . I leave to 0 the number of frames, it will generate videos of 2 second with 8 frames per second.

Right clic to "Generate Forever"

Then you will have to wait and get many videos (I left my computer for many hours during then night) and I got many videos.

Then I generate the video with KdenLive.

For the text I wrote an idea and ChatGPT wrote the text.

I changed it a little bit to make it shorter.

For the voices I have used OpenAI API, I have a easy tool (I will make it free soon with a better interface, of course you need to set your Open API code to use it).

for the music I use:

Music by AlexGrohl from Pixabay

I hope you enjoy

11

Comments