Disagreement enjoyer. Unpopular opinion. Usually take an opposing stance. I believe I'm right so you don't have to.
Joined October 2009
- Tweets16.4K
- Following338
- Followers151
- Likes18.7K
Pinned Tweet
Z-Image Turbo as the photo generator
Flux 2 Klein as the style transfer
Totally random prompt.
Control by concept slider loras on ZIT and the self train style transfer loras blend on Klein to get this specific style consistently. 80-120 sec for the real-stylize pair on RTX3060.
RYOUGI Ilakya retweeted
ของดี เไ วันนี้ — IconMCP
iconmcp.com/
เว็บแจกไอคอนเวกเตอร์ ที่ใช้ เไ ค่อยๆวาดทั้งวันทั้งคืน (อันไหนที่วาดอยู่ น้องก็จะวาดให้เห็นเลย เป็นแอนิเมชัน 555) ตั้งเป้าว่าจะแจก 1 แสนแบบ
ไอคอนสามารถปรับแต่งขนาด ความหนา สี มุมมน ฯลฯ ต่างๆ ได้หมด และฟรี (MIT licence)
RYOUGI Ilakya retweeted
UniRig
Automatic skeleton and skinning for any 3D model, from SIGGRAPH 25
UniRig predicts a skeleton and skinning weights for a 3D mesh so it is ready to animate.
It was built by Tsinghua University and Tripo.
One model handles humans, animals and unusual shapes. You run skeleton prediction, then skinning, then merge the result back into your file.
The authors now also point to SkinTokens, a successor that does both steps in one pass.
github.com/VAST-AI-Research/…
RYOUGI Ilakya retweeted
また、ローカルPCで動く謎の技術が公開されてる…
写真一枚からガウシアン・スプラット（Gaussian Splats）の3Dモデルにしてくれるんだけど、結構いい感じに生成できる模様。
github.com/speridlabs/idol-s…
🤖 Made with AI
RYOUGI Ilakya retweeted
Qwen3.5-4B-GLB-3D-Coder
Renderable 3D assets from a 4B model.
Following its earlier 3D Coder project, CCORE has released a new Qwen3.5 fine tune focused on generating 3D assets for game development.
The model turns text prompts into standalone HTML and JavaScript scenes using Three.js.
Each scene includes a GLB export button, allowing generated assets to be imported into Godot.
Examples include a mana potion, goblin, desert merchant and detailed environment props.
The developer reports most generations work on the first attempt, especially with English prompts.
Q4_K_M and Q8_0 GGUF checkpoints are available.
Full model weights and the training dataset are planned for release after training finishes.
huggingface.co/ccore/Qwen3.5…
RYOUGI Ilakya retweeted
もしかして Tripo Canvas 凄い？
Making a character with the new "Tripo Canvas"
I'm not sure if I should use more nodes
I've tried out the new Tripo Canvas, which is still in beta, and it was very easy to work with nodes, just like in other 3D programs. The Tripo Canvas workflow lets you work with Tripo much faster than the conventional way—creating multiple references at the same time, separating parts of the model, and creating different views for the body and head, as well as accessories.
Although I found all of this extremely useful, I think the most useful feature is the AI Lab node, which lets you interact with an AI agent to separate meshes, close gaps, eliminate distortions, merge meshes, and set up or build a scene using various 3D models or images you provide (it also gives you all the results in a Blender file so you can continue working on them in Blender if you like).
I’ll keep testing the tool to see what else I can create🐋 #blender
RYOUGI Ilakya retweeted
Finally! Rigged hair cards from a single image.
HairDeck (Tsinghua + CMU + CUHK) takes one character image and builds a stylized haircut as separate 3D hair cards, with bone chains already attached. The hair sways when the head turns.
Exports to GLB, or straight onto a VRM avatar.
Code and weights are open under Apache 2.0. You get geometry and UVs, no textures yet. Local run wants Linux and CUDA.
github.com/hyz317/HairDeck
