AI
AutoRef: Harness Optimization for Agentic Multi-Reference Image Generation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 12:19:59 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 12:18:24 UTC.
Briefings
AutoRef is an automated coding agent that optimizes harnesses for multi-reference image generation by iteratively rewriting code while keeping underlying models frozen. The system improves open-weight FLUX.2 4B models on the MultiBan benchmark, raising held-out four-reference scores from 5.72 to 7.37. This performance gain matches or exceeds proprietary models like Nano Banana Pro and GPT-Image-1.5. The method distinguishes between harness optimization tasks and candidate selection tasks, using a beam search from top-ranked harnesses to refine results.
Key points
- AutoRef uses an automated coding agent to iteratively rewrite harness code for multi-reference image generation while keeping the underlying image generation models frozen.
- The system improves open-weight FLUX.2 4B models on the MultiBan benchmark, raising held-out four-reference scores from 5.72 to 7.37, matching or exceeding proprietary models like Nano Banana Pro and GPT-Image-1.5.
Context and limitations
Performance improvements are measured on held-out four-reference tasks of the MultiBan benchmark; results may vary with different generator, reference count, benchmark, evaluator, or reasoning model configurations.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 12:19:59 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeieiwzseoagus5iqehjbyjlnzpbesdb5v6ymgps4dua4ppaqlssnta