Dennis-JwWeng

Dennis-JwWeng / PartFlow

Public

PartFlow: two-stage image-conditioned 3D editing (inference code)

32
0
89% credibility
Found May 27, 2026 at 32 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

PartFlow is an AI-powered 3D editing tool that transforms existing 3D models based on a target image. Instead of creating 3D objects from scratch, you show it what you have and what you want, and it edits your model accordingly. It's designed for designers, artists, and developers who need to modify 3D assets quickly without manual sculpting or rebuilding. The system works in two stages: first adjusting the structure, then refining the appearance, outputting a ready-to-use 3D file.

How It Works

1
💡 You have a 3D model you want to change

Maybe it's a chair, a vase, or any 3D object. You want to modify it somehow without starting from scratch.

2
🎨 You imagine the edited version

You create or find a picture showing how you want the modified object to look. It could be a different color, material, or style.

3
🖼️ You prepare your inputs

You put your original 3D model and the target picture in a folder, following simple naming instructions in the documentation.

4
⚡ You run the editor

With one command, you launch the editing process. The system reads your model and target image, then transforms your 3D object to match your vision.

5
🔄 Your model gets edited in two stages

First, the overall structure is adjusted. Then, the detailed appearance is refined. Everything happens automatically in the background.

🎉 You receive your edited 3D model

The system saves a complete 3D file you can open in any 3D software. Your vision is now a real, editable 3D object you can use anywhere.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 32 to 32 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is PartFlow?

PartFlow is a feedforward 3D editing tool that transforms existing 3D assets based on a target image. You feed it a source 3D model and a reference image showing what you want, and it outputs an edited textured mesh. Built in Python on top of Microsoft's TRELLIS 3D generation framework, it uses a two-stage flow matching approach: first predicting the edited voxel structure, then refining the structured latent representation. The pipeline outputs ready-to-use .glb files with mesh and gaussian splatting representations baked in.

Why is it gaining traction?

The main hook is speed and simplicity. Traditional 3D editing requires per-asset optimization taking minutes or hours; PartFlow does it in one forward pass. It also eliminates the need for 3D masks at inference time, which is a major friction point in competing approaches. The image-conditioned editing is particularly compelling for workflows where you have a reference concept but want to preserve the original asset's structure.

Who should use this?

Game developers who need quick variations of existing assets. 3D artists prototyping edits before committing to full production pipelines. Researchers in 3D content creation evaluating feedforward editing baselines. This is not for beginners: the setup requires CUDA extensions, TRELLIS dependencies, and pre-encoded input formats.

Verdict

PartFlow delivers on its core promise of fast, mask-free 3D editing, but the 0.8999999761581421% credibility score and modest star count reflect a research codebase in early stages. Documentation is adequate but setup complexity will trip up anyone expecting a plug-and-play experience. Worth evaluating for specialized editing workflows, but expect to invest time in environment configuration before seeing results.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.