# zai-org/ImageReward

[NeurIPS 2023] ImageReward: Learning and Evaluating Human Preferences for Text-to-image Generation

Repository: https://github.com/zai-org/ImageReward
Canonical: https://ross.abutalabs.com/products/imagereward
Language: Python
License: Apache-2.0
License Family: permissive
Topics: diffusion-models, generative-model, rlhf, human-preferences
Last push: 2025-10-29T13:40:51+00:00

## Health v2 (maintenance only)
Score: 52/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 49, release rhythm 35, longevity 89
- inputs: {"age_days": 1250, "days_push": 308, "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 1702, forks 92 (observed 2026-08-28T04:05:24.440162+00:00)

## What it is
ImageReward is the first general-purpose human preference reward model for text-to-image generation, trained on 137k expert comparison pairs. It provides a Python package for scoring generated images against prompts and ReFL fine-tuning of diffusion models using reward feedback.

## Use cases
- score text-to-image outputs by human preference
- rank candidate images from stable diffusion prompts
- fine-tune a diffusion model with reward feedback learning
- evaluate text-to-image models against human preferences
- replace CLIP-based scoring for image quality assessment
- benchmark generative image models

## When to choose
- you need human-preference-aligned scoring of text-to-image outputs
- you want to RLHF-fine-tune stable diffusion with a reward model
- CLIP or aesthetic scores are insufficient for evaluating generations

## When to avoid
- you need video or multi-dimensional reward evaluation (use VisionReward instead)
- you have no GPU or cannot run large vision-language models
- you need general image classification rather than preference scoring

## Facets
- artifact type: library
- maturity: stable
- function: machine-learning, deep-learning, benchmarking
- domain: machine-learning, image-processing, artificial-intelligence
- platform: python
- tags: text-to-image, reward-model, rlhf, diffusion-models, human-preference, image-scoring, gpu

## Member repositories
- zai-org/ImageReward (main) score 52

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:24.440162+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 2026-08-30T03:37:41.495631+00:00, confidence not recorded.
  - readme: https://github.com/zai-org/ImageReward (fetched 2026-08-28T04:05:24.440162+00:00, sha 8d890db9c134)
- Data as of 2026-08-30T08:39:29.467469+00:00.