chambersu
patreon
2025 September Preview
🕑 Added 2025-09-20 12:04:29 +0000 UTCwill post some interesting tests of Sept, but too busy to finish soon,
I list some key points in advance here, and will detail with samples asap:
“If you have any questions or would like to discuss anything further, please don't hesitate to:
Leave a comment
Email me at chambersu1996@gmail.com
Join my Discord channel: https://discord.gg/XAfS57Rrj6
I'll respond as quickly as possible.”
1. Nano Banana can turn a single-view design into a multi-view design,
Then input it to 3D AI, which has improved a lot, too
(1)I used comfyUI to create single-view designs(I need the stable styles/theme from personal LoRAs, we can try whatever we need actually)
(2)I prefer to use Nano Banana here:(Photoshop users should be able to use soon)
https://lmarena.ai/
Prompt for multi-view designs:
Convert to T pose first with the single-view designs input,
"Retain the subject and color of the image and transform the character design into T Pose on the front view. White Background."
Then the side view with T pose input,
"Retain the subject and color of the image and generate the side view of the subject. Don't change the pose. White Background."
Finally, the back view with Tpose and side view input,
"Retain the subject and color of the image and transform the back view of the character design into T Pose. White Background."
(also powerful in other ways, we can find a lot online~)
When AI can totally focus on single-view designs first, the feelings and elements will be much better than multi-view designs in one shot.
(3) Huanyuan 3.0 for multi-view sketch models,
https://3d.hunyuan.tencent.com/
Then use the camera space projection mentioned before to tweak the shape and bake the texture (costs about an hour or two),
separating the head/hair and body/suit will be much better.
2. Combining GPT and Suno can create some songs and sounds for games/films.
https://suno.com/playlist/20261a20-0a1f-4fb8-bee9-150763c3a298
I prefer to use Deepseek in my mother language.
https://www.deepseek.com/
I think there are some key points among the whole prompts,
like providing a Game Design Document and a personal favorite playlist as reference,
and
"
(1)Provide an overall description of the song,
(2)Describe the key performance points of each musical section along the timeline,
(3)Write English lyrics,
(4)Suggest what type of vocal tone would be suitable for the singer.
(5)Offer recommendations on how to utilize details such as chords, melody, instrumentation, and synthesizers,
(6)Formulate a prompt that meets the standard format for input into Suno AI.
- The prompt can include lyric suggestions (within 5,000 characters, using brackets [] to indicate performance effects, with more detailed line-by-line annotations to guide dynamic performance nuances) and style suggestions (within 1,000 characters, which may not suffice for describing all details comprehensively).
- The prompt must not include specific song titles or artist names as references; instead, only abstract descriptions should be used to approximate the desired style.
"
But I am not sure. I recommend trying it by yourself~
Suno is something easier to use, just need credit to roll for times, improve song by song(Generate/Remix),
Pick the best song, then improve word by word if needed,
or further tweaking on DAW/by hand/vocal from its Stem results if you are good at.. ~
3. Some assets/plans can be shown and posted later~
(1)The basic rig of Mia with 2 extra skins(suits), advanced topology, Auto Rig Pro (body rig), and Face It (facial rig), works not bad for retargeting or mocap~
They are used for Unity Engine.
(2)(WIP) hairstyle/body/suit of Ju-Eun,
no matter chibi, anime, or ball-joint doll-like, they seem easier to create better than the photorealistic, while real people are more charming, kind of challenging in my spare time.
Try to get over and cover this style as a 3d artist by an artwork that I will finally be satisfied with.
Also tiny doll test based on the old base mesh "Chibi" (5 years ago lol), soon. for fun.~
and maybe printable tests with clear/separate parts and pure/flat color design to print easier by a 3d printer and airbrush,
(All rigid body things like BJD and mech should work the same if possible.
(3)other secret bio/mech project...~
I tried some AI video, but I think text, images/textures/3D, and sound can help me a lot.
While the videos, like offline rendering, can't, and the results are still slow/expensive to create, not ideal now~ will post when ready.
