# MUSE-AI-Lab/Concept-Brittleness

> Research code for interpreting and fixing object-dependent concept brittleness in text-to-image diffusion models: a sparse autoencoder is trained per denoising step to locate missing concept dimensions and correct failures at inference without retraining.

- Magnitude: 4.2 out of 10 — Early signal
- Stars: 262 total · +0 stars measured 2026-10-31, 00:27–10:10 UTC, ≈ 33 by evening
- Star trust: star growth looks organic
- Category: Generative media · Language: Python · Created: 2026-10-28 · Last push: 2026-10-29
- GitHub: https://github.com/MUSE-AI-Lab/Concept-Brittleness · Page: https://gitnova.dev/en/r/MUSE-AI-Lab/Concept-Brittleness

## Useful for

- Diagnose why a fixed style prompt fails on specific objects across seeds
- Repair failed style or attribute generations at inference without retraining
- Compare concept brittleness across FLUX, SD3.5, SDXL, PixArt and SD1.5

## Why it’s here

- Star-counter measurements on 2026-10-31 (UTC), 00:27–10:10: 262 → 262 stars (+0). This is the change over that interval.
- Estimated end-of-day forecast: about +33 stars, using observed gains and the previous day.
- The repository is 3 days old and already has 262 stars. With less than two weeks of history, there's no usual pace to compare the spike against yet.
- Top new repositories this week: #81.

## Star trust

Star growth looks organic.

Star-trust labels are heuristics based on the repository’s behavior, not a check of every stargazer.

## Numbers

- Forks: 0
- Issues and pull requests: 6
- Watchers: 0
- Average over the last week: 74 per day
- Usual pace: too little history (under two weeks)
- Stars in the last hour (measured): 0

## Stars per day, last 4 days (oldest → newest, today is partial)

2026-10-28 … 2026-10-31: 5, 156, 101, 0

## Magnitude by day, last 2 days

10-30 5.1, 10-31 4.6

## Spotted in now

- Top new repositories this week: #81

## More in this category

1. **paipiange/nvzhuang** — 8.0 · Early signal · Generative media · Go · +307 stars measured 2026-10-31, 00:27–10:09 UTC, ≈ 794 by evening
   A static catalog of 1407 AI-generated women's clothing images with bilingual search and optional virtual try-on via external image generation models.
   Full card: https://gitnova.dev/en/r/paipiange/nvzhuang.md
2. **anhao/ai-pose-gallery-zh** — 6.5 · Peaking · Generative media · TypeScript · +387 stars measured 2026-10-31, 00:27–10:09 UTC, ≈ 838 by evening
   Chinese localization of an AI pose reference gallery: 1439 records and 2878 images with a ready-to-deploy dependency-free static site. Useful for artists and 3D modelers picking poses.
   Full card: https://gitnova.dev/en/r/anhao/ai-pose-gallery-zh.md
3. **sleepy-zone/ai-ref-library-cn** — 6.0 · Early signal · Generative media · JavaScript · +53 stars measured 2026-10-31, 00:27–10:09 UTC, ≈ 150 by evening
   A Chinese-language search interface for three sets of AI drawing reference materials (faces, poses, clothing) by author agi/gia, with translated tags, bilingual search and one-click prompt copying.
   Full card: https://gitnova.dev/en/r/sleepy-zone/ai-ref-library-cn.md
4. **EntranceStarfish/VoiceStudioV2** — 5.8 · Early signal · Generative media · +1 star measured 2026-10-31, 00:35–10:09 UTC, ≈ 43 by evening
   Local desktop voice studio for real-time voice changing, text-to-speech, voice cloning and dubbing, with an OpenAI-compatible API and fully offline operation.
   Full card: https://gitnova.dev/en/r/EntranceStarfish/VoiceStudioV2.md
5. **D3voz/openjutsu** — 5.8 · Early signal · Generative media · TypeScript · +64 stars measured 2026-10-31, 00:27–10:09 UTC, ≈ 173 by evening
   A local web app for swapping characters and objects in videos: drop a video and a character photo, and the main person in the frame is replaced by your character while camera, cuts and soundtrack stay intact. Runs on your own GPU via…
   Full card: https://gitnova.dev/en/r/D3voz/openjutsu.md

---

Magnitude (0–10) measures how fast and how unusually interest in a repository is growing right now. It is not a quality score.
Days are UTC. “So far today” is a fact; “expected by the end of the day” is a forecast.
Summaries and use cases are written by an LLM (DeepSeek V4.1 Flash) from the README and may be inaccurate: verify specific claims (benchmarks, speed, hardware) in the repository itself.
Data as of 2026-10-31 10:19 UTC, updated every 30 minutes.