Type | |
Stats | 324 1,832 581 |
Reviews | (53) |
Published | Feb 23, 2025 |
Base Model | |
Hash | AutoV2 9DBC44C260 |
Pony merged ver:
These LoRAs are designed for my ComfyUI workflows only,
Please take a look at the details on my Patreon Chamber | creating Game, Art | Patreon,
Download the generated images above, then drag them to ComfyUI to load the workflow behind them (the images generated by ComfyUI will attach their workflow data inside the image files).~
You can also try to use them in another way, but I can't promise their efficacy in other cases/projects.~ (I will provide support to Patreon followers in 24 hours if needed)
If you are interested in or want to download more...
more my personal LoRAs, merged checkpoints for game design
The raw generated image with generation info,
LoRA training preset files,
raw pictures, data training set, and tag preparation,
prompt structures
and how to model these images into 3d models,
...
please take a look at my Patreon Chamber | 3d Asset, Dev Tool, Game | Patreon,
I will provide support to followers there, and answer whatever I know in 24 hours~
Patreon Subscriptions can help me update further and create more~
There are detailed posts and assets there, and I am willing to help if needed~
Personally, I prefer to use the separated turn LoRAs(BG3/Odin-MMORPG/Anime Mod/MUJI...), which are more predictable/controllable for different styles,
and seems IL/F1 works better than Pony for me,
IL/F1/XL BG3 Turn, Multi-View, Turnaround, Model Sheet, Character Design - IL | Illustrious LoRA | Civitai
IL Anime Model Turn, Multi-View, Turnaround, Model Sheet, Character Design - v1.0 | Illustrious LoRA | Civitai
But a merged one can simplify the turn LoRAs arrays as one LoRA, and then we get simplified workflows that are easier to understand/use. Pony may be better for other projects in some ways,
also simplified all unnecessary nodes like the complex Detail Daemon Samplers(which are used to decrease details to clean anime designs),
expanded and organized the nodes for easier workflow previewing ~
###
(WebUI History)
1. for better results, a suitable CKPT is important, I prefer CKPTs based SDXL instead of Pony, I will upload XL ver today, I think AbyssOrange XL Else is a good start, then test, merge, or train for different cases/projects,
2. the prompts behind (VIEWS:0) are important,
3. use at least 2 Turn LoRA, the main one weight around 0.7, and the secondary one weight around 0.15 is one of good strategy, test the weight and element for better results please,
4. the resolution should be 1536*1024 then upscale, because of the train data, personal sample setting: Euler A, 20 steps, R-ESRGAN 4x+ Anime6b, denoise strength 0.3, upscale by 1.25, CFG 7, clip skip 2, sdxl VAE,
5. turn on after detailer,
6. turn on CtrtNet OpenPose, 0.6, pixel perfect, pick one of the samples above as input,
Actually, I am a 3d artist/game developer, multi-use LoRAs are necessary as a bridge to both 3d mesh and texture(like more stable special static pose and seamless details), and they work fine for my commission and project(the setting for products are more complex but not only about AIGC/SD which may not suitable post on CivitAI), I just want to inspire some guys with my workflow and look for potential chances,
merge them can make them one to use but separating them helps me tweak styles more easily,
Welcome to customize them by yourself or just leave them here to use the better LoRAs for different projects~
Please take a look at my Patreon Chamber | 3d Asset, Dev Tool, Game | Patreon,
Subscriptions can help me update further and create more~
There are detailed posts and assets, and I am willing to help my Patreon followers in 24 hours if needed~