Skip to content
← Models
ClosedGoogle Research/10 trustMetrics-checked

StyleDrop

Learns any visual style from a single reference image; 86% of raters preferred it over DreamBooth.

Visit official page →

About

StyleDrop, from Google Research in mid-2023, solves a problem text prompts can't: describing a visual style precisely. Feed it one reference image and it fine-tunes a tiny adapter (under 1% of the model's parameters) on Google's Muse text-to-image transformer, using iterative training with human feedback to avoid overfitting to that single image. The result generates new images that hold the reference's colors, textures and shading while following new text prompts.

Google reported a style-consistency CLIP score of 0.694 versus 0.556 for the base model and 0.644 for DreamBooth on Imagen, and 86% of human raters preferred StyleDrop's style fidelity. It never shipped as open weights — it was adapted into Vertex AI as a custom style model — so today it's a technique to know rather than a model to download. Google's own writeup flags the obvious risk: copying a living artist's style without consent.

Specs

Context window

Not applicable — built on the Muse text-to-image transformer

Pricing

No public release; related capability offered via Vertex AI (paid)

Free quota

None

API access

No open release — Google adapted it into Vertex AI custom style models (provider-claimed); no public API for the research version

Weights

Closed — API only

Released

Not disclosed

Benchmarks

  • Provider-reported (research.google blog): style-consistency CLIP score 0.694 (HF version) vs 0.556 base Muse, 0.644 DreamBooth on Imagen; text consistency 0.322
  • Provider-reported human study: 86% of raters preferred StyleDrop on Muse over DreamBooth on Imagen for style consistency
  • No independent third-party benchmarks

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

style-transfertext-to-imageimage-generationdesign-toolsbrand-assets

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (TIER metrics, metrics-v1). Evidence: arXiv preprint (2306.00983) with quantitative CLIP-score evals and a human preference study; official Google Research blog post corroborates the numbers. Strengths: influential method, honest reporting of limitations including style-copying risks. Docked for: (1) built on the internal Muse model with no public weights — not reproducible by outsiders; (2) all metrics are provider-reported. License: not applicable — no public release.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
6.0/10
Maintenance · recency, releases, issue load
6.0/10
Openness · license + docs
2.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · ArXiv 2306.00983; style-consistency CLIP 0.694 vs 0.556 base, 0.644 DreamBooth; 86% human raters preferred StyleDrop over DreamBooth (provider-reported)
  • · Google Research blog corroborates; honest limitations incl. style-copying risks
  • · Built on internal Muse model — no public weights, not reproducible; all metrics provider-reported

Sources