# Fugtemypt123/VIGA

VIGA: Vision-as-Inverse-Graphics Agent

Repository: https://github.com/Fugtemypt123/VIGA
Canonical: https://ross.abutalabs.com/products/viga
Language: Python
License: MIT
License Family: permissive
Last push: 2026-05-06T07:32:15+00:00

## Health v2 (maintenance only)
Score: 55/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 81, release rhythm 35, longevity 30
- inputs: {"age_days": 433, "days_push": 119, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1275, forks 124 (observed 2026-08-28T04:04:12.911619+00:00)

## What it is
VIGA is an analysis-by-synthesis code agent that reconstructs 3D scenes and slide layouts from images by generating and executing Blender Python or PowerPoint programs. A single self-reflective agent alternates between a Generator role that writes scene code and a Verifier role that compares renders against the target image, iterating until they match without any finetuning.

## Use cases
- reconstruct a 3D scene from a single image as editable Blender code
- generate Blender Python programs whose renders match a target image
- automatically synthesize PowerPoint slide layouts from reference images
- build 4D dynamic scenes with physics from images
- evaluate multimodal agents on 3D graphics editing benchmarks like BlenderBench and BlenderGym
- iteratively refine generated scene code until rendered output matches a reference

## When to choose
- You want image-to-3D reconstruction that outputs editable, programmatic Blender scenes rather than raw meshes
- You need a self-correcting generate-render-verify agent loop that works without finetuning
- You want to reproduce or extend research on BlenderBench, BlenderGym, or SlideBench, or run custom static/dynamic scene modes

## When to avoid
- You need fast or real-time reconstruction - the write-run-compare-revise loop is slow and GPU/CUDA-intensive
- You want production photogrammetry or CAD tools rather than a research-grade LLM agent
- Your environment cannot call large multimodal LLMs due to cost, latency, or privacy constraints

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, computer-vision, graphics
- domain: artificial-intelligence, computer-vision, graphics, large-language-models
- platform: python
- tags: inverse-graphics, blender, 3d-reconstruction, scene-synthesis, analysis-by-synthesis, multimodal-llm, agentic-coding, render-and-verify, slide-generation, 4d-scenes, no-finetuning, ai-agents, linux, gpu

## Member repositories
- Fugtemypt123/VIGA (main) score 55

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:12.911619+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T05:02:59.735943+00:00, confidence not recorded.
  - readme: https://github.com/Fugtemypt123/VIGA (fetched 2026-08-28T04:04:12.911619+00:00, sha 4872e91b9c0d)
- Data as of 2026-08-30T08:39:29.467469+00:00.