[v3.0] QR Code Experiments - Abstract Animation [Breakdown] (Patreon)
Downloads
Content
A New Folder "Experiments" Has been added to the google drive where I will be putting all experiments files with Render Videos.
To give more clarity, Raw, refined and Face Fix steps files are organized to give you a flow of the process.
The v3 workflows used which are tweaked a little according to my needs.
Link : https://drive.google.com/drive/folders/1HoZxKUX7WAg7ObqP00R4oIv48sXCEryQ
__________________________________________________________________________
I have done 2 Experiments with ORcode monster Controlnet, and came with the following Observations :
Normal vs QRcode
- Images produced with QRcode are brighter, more appealing.
- Loras blend better with the Subject, See feet and chest area, Loras have affected those areas more than without QRcode
QR Code Theory
QR Code focuses on White Areas as Fore Ground and Black Areas as Background. So Using this theory and putting the Auto mask generated from here as the input images for the Qrcode Controlnet gave the above results.
QRCodeMonster Controlnet Model Link : https://civitai.com/models/111006/qr-code-monster
BREAKDOWN [Without LCM] :
1) Without LCM Render - https://youtube.com/shorts/kC0WeYx9vXA
2) With LCM Render - https://youtube.com/shorts/zsgmvIKXh6w
The Above Render is achieved without using the LCM. It is observed not using LCM here has produced more better Effects with Loras and Prompts.
- Epic Realism was used for Raw Animation, then Animerge model was used for 2x Refiner passes to get minute Details in the image.
- Then 2.1 Face Fix file
All Workflows can be downloaded from the Experiments folder.
Models, Prompts and Loras
*Important - Negative Prompts were kept to minimum and not using embeddings, they pushed the generations from being abstract...
Below are the models and Loras used for the LCM and No LCM workflows
Models List:
1) Epic Realism: civitai.com/models/25694?modelVersionId=189522
2) Animerge: AniMerge - V2.6 | Stable Diffusion Checkpoint | Civitai
3) HelloFlatColor2d: https://civitai.com/models/113447?modelVersionId=176342
Lora List:
1) Bubbles07: シャボン玉背景 - Soap bubble background Lora - v1.0 | Stable Diffusion LoRA | Civitai
2) Concept Pyromancer: https://civitai.com/models/131189?modelVersionId=144163
3) JD Lightning FX: 乾罡五神雷决 炼雷 Dry gang five gods thunder and thunder - v1.0 | Stable Diffusion LoRA | Civitai
4) Neon Outlines: Neon Outlines - v1.0 | Stable Diffusion LoRA | Civitai5) Rage Anime: Rage Unleashed - v1.0 | Stable Diffusion LoRA | Civitai
6) Water VFX: 💦WaterVFX - Create more consistent water - WIP - V1 | Stable Diffusion LoRA | Civitai
7) Water FX 3: 御水 v3 - v3.0 | Stable Diffusion LoRA | Civitai
8) World of Water: World of water - World of Water v1.0 | Stable Diffusion LoRA | Civitai
9) Diamond FX: https://civitai.com/models/100388/five-movements-ice
10) Style Swirl Magic: Style Swirl Magic - Style Swirl Magic | Stable Diffusion Embedding | Civitai
(PS: I renamed some of them due to other languages)
QRCODE CONTROLNET
AutoMask images are used for QrCode controlnet model with 1.5 weight.
also same AutoMask images are used as Masks:
Background:
Green Screen images of Fire has been used as background images to get a upward flowing water motion in the Segment 1 of the Without LCM Render ...
As an experiment to control water like motion ...like waves going upwards I used fire ... It will anyways be converted to line art so it didn't mater whatever we use here.
I'll Try to control background motion in my next Experiment :D
_________________________________________________________________________
Conclusion
It seems without LCM file performs better in producing minute elements details. So I've added 2 more files in the v3 folder which are without LCM.
1) Main Drive Link : https://drive.google.com/drive/folders/1HoZxKUX7WAg7ObqP00R4oIv48sXCEryQ
2) Experiments Folder : https://drive.google.com/drive/folders/1TP7NdslT1-TfxbkeeDP72xE-fJyKjePH
_________________________
My Discord Server : https://discord.gg/z9rgJyfPWJ