chambersu
patreon
Assets and tips of Sept 2025
🕑 Added 2025-10-07 20:06:49 +0000 UTCIf you have any questions or would like to discuss anything further, please don't hesitate to:
Leave a comment
Please email me at chambersu1996@gmail.com
Join my Discord channel: https://discord.gg/XAfS57Rrj6
I'll respond as quickly as possible.
Thank you very much to all my supporters on Patreon. I am sure I can't insist without your help.~
If you read from somewhere else, consider supporting me on Patreon if possible.~
patreon.com/chambersu
If still suffering from the bill like me or other plights, please relax and feel free to enjoy the contents, which will encourage me to post more and better too~
[Outline]
Design reference/3d sketch improvement:
(1) LMArena text-to-image by different AI for various design references.
(2) ComfyUI and LoRAs text-to-image or image-to-image for the consistent style/theme design references.
(3) Nano Banana for multi-view designs from concept designs,
and details tweaked by text (work on Photoshop Beta now~).
(4) Hunyuan 3D V3.0 is free and good enough for 3d sketch from concept art or multi-view designs.Assets Update:
(1) Modern Anime character Mia rigged update and 2 extra skins.
(2) Photoreal character test Lee Ju-Eun update.
(3) Chibi character sketch Ju-Eun Dolls with different suits and hairstyles.
(4) Medieval Vmod character Revna and Svend update.
[Content]
I wrote the tips down here based on my actual work during modeling/developing,
attached the example/projects/files,
which is not the best way to create, but should be useful enough for some practices/common work.~
AI design reference/3d sketch improvement:
(1) LMArena text-to-image by different AI for various design references.
LMArenaImgSample.zip
I used the LMArena text-to-image feature more and more frequently during these days,
which is suitable for some quick cases that are free and open about different art styles.
LMArena will randomly use different AI to output the results, which is reasonable to get more fresh, various, and unpredictable concept references at the very early stage of development.
And we can stay sensitive about the latest AI here instead of relying on or getting addicted to one of them.
It is cool for creative reference work, and free to use.
We can use LLM/AI chatbots like ChatGPT to read the game document/setting to output some image prompts quickly.
Try tags from images by MLLM (Multimodal Large Language Models including visual features) or other tag tools like Joy Caption, or to use image-to-image directly, which should work well too.~
We can still use some smart/traditional tools to composite or paint further these materials if needed.
(2) ComfyUI and LoRAs text-to-image or image-to-image for the consistent style/theme design references.
ComfyUIImgSample.zip
Because we can use Nano Banana to output (t-pose) multi-view designs from single-view concept designs now,
For ComfyUI, we can consider bypassing the CtrlNet and multi-view LoRAs,
directly use suitable checkpoints and style LoRAs for concept designs reference,
which will work better for prompt responses and details in most cases,
and able to output a series of designs based on the same style CKPT/LoRAs.I used Nanchaku Flux Dev and some anime/game-style LoRAs here,
The sample images include the workflow data,
Download and drag them to the ComfyUI will load their workflows directly.But I thought and tested, found that human/character faces are hard to describe with words/prompts,
(I can't make AI/people know the face features inside my head by words, even some faces we are both familiar with.)
That's one of the important reasons why I try further about photoreal human (for Ju-Eun) and style shifting work,
Those are my plan/solution to describe a character's face,
by collecting and using suitable faces (as parameters) and blending them stylishly for special character settings.
(3) Nano Banana for multi-view designs from concept designs,
and details tweaked by text (work on Photoshop Beta now~).
NanoSample.zip
Prompt:"Retain the subject and color of the image and transform the character design into T Pose on the front view. White Background."
"Retain the subject and color of the image and generate the side view of the subject. Don't change the pose. White Background."
"Retain the subject and color of the image and generate the back view of the subject. Don't change the pose. White Background."Also, lots of slight tweaks can be tried by words/prompts on PS directly now, nice tool
(4) Hunyuan 3D V3.0 is free and good enough for 3d sketch from concept art or multi-view designs.
link: Hunyuan 3D
HunyuanAndRodinSample.zip
Separate the geometry and texture generation,
and try smart poly topology between them,
Then the output results based on these settings are much easier to edit as sketches in my opinion.~
I am not satisfied with the AI 3D body and head/face structure now,
so I use/replace other base meshes instead, which I have mentioned some of them before and below, like anime/vmod/avatar/Advanced...
will talk further on the asset parts, also an important part to improve my artwork.
(Rodin works better on the head modeling as mentioned before,
which can deal with the head,
then merge to the body like my Blender project sample here,
but hard to run out of all the credits monthly for me,
will test the latest Rodin Gen-2 further later.)
For AI 3D, I mostly need the rough 3d structure to edit further.
For important characters, we should rebuild lots of things later,
but the sketch models with camera-space-based texture projection look nice already,
(if the camera won't get closer, like in lightweight isometric games, MOBAs/ TRPGs/ Roguelike...)~
Assets Update:
(1) Mia Rigged and 2 extra skins.
They are early assets for Vertical Slicing on Unity.
MiaRig_FanAndTer_v03.zip
The first character I tried the topology learned from Universal Human - Advanced Rig,
Then I found some disadvantages behind the nice look and performance of it,
which require at least subdivision level 1 to make the shape look right,
but less control for the bodyshape in this way,
and make less ‘low-poly’ lol for real-time rendering/gaming.
And also, the edge ring/loop is hard to read/edit,
too complex independence between meshes(like pose space deformations),
bones (for the control rig connecting to the deformed bones),
and shaders (like dynamic wrinkle/micro normal maps),
which are necessary for better results (aiming at films maybe), I understand, but not suitable for all cases, we should learn and figure them out first.
Also, some potential improvements here,
If we need better models, we can try to make them watertight (to avoid most mesh clipping issues when animating). Choices between watertight and modular assets depend on requirements,
and pose/camera space deformation(shape keys) for better dynamic looks,
and hand-painting textures for better details.
The physics solution for cloth and hair simulation will matter too, too obvious to ignore.
(which seems to work for other assets here if looking for something product-ready, be patient and enjoy until the final results~)(2) Photoreal character test Lee Ju-Eun update.
L33_AdvRig_v02_t01_Post2.zip
Remove the base color map of her face to avoid abuse,
and remove the micro displacement/depth/normal map of her face to save space.will polish the suit step by step, which will work well, I guess,
But the problem is the floating hair and hair card shader, which need more research.~
(3) Chibi Doll sketches with different suits and hairstyles.
Plan to improve the chibi base mesh and test something printable and shading figures during these tests.
L33Doll_v01_t03_Post.zip
(4) Medieval Vmod character Revna and Svend bodyshape and rig update,
which will be more standard to retarget for other animations.
SvendV11RevnaV14.zip
The old asset used the Avatar(vmod/anime) topology (like the old character assets, but customised a few for the requirement, may roll back a little bit like UVs, material slots, and object containers hierarchy for myself later).
which I plan to improve by knowledge from the other assets,
their topology around the shoulder, butt, and chest can be better to draw their shape later,
and the hip part connected to the root of the thighs can add some edge loop for better dynamic deformation,
I still love them for they are strictly using easily readable and editable edge loops everywhere
but a lack of good-looking static shapes and dynamic deformations before I learn and practice further,
but now I can merge the better features, needed advantages together,
and avoid the weakness in the future, to get a better base mesh.Also, free to improve their design and textures with AI now.
Replace the simple customised rig with the Auto Rig Pro humanoid rig,
which has a better skinning engine for auto weighting,
control bones for animating,
and useful export presets for different game engines.
...
Some other messy things and ideas will be posted when ready.~
If you have any questions or would like to discuss anything further, please don't hesitate to:
Leave a comment
Please email me at chambersu1996@gmail.com
Join my Discord channel: https://discord.gg/XAfS57Rrj6
I'll respond as quickly as possible.
Thank you very much to all my supporters on Patreon. I am sure I can't insist without your help.~
If you read from somewhere else, consider supporting me on Patreon if possible.~
patreon.com/chambersu
If still suffering from the bill like me or other plights, please relax and feel free to enjoy the contents, which will encourage me to post more and better too~
