AI
FlowTool: Controlling Tool Parameter in Image Retouching via Flow Matching
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 21:04:27 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 09:18:24 UTC.
Briefings
FlowTool is a framework for image retouching that models high-quality tool parameters using conditional rectified flow, combining a vision-language model backbone with a Diffusion Transformer generator. Trained via two-stage supervised flow-matching curriculum and reward-based post-training, it outperforms specialized MLLM editing agents and proprietary MLLMs on benchmark tasks while maintaining competitive performance on reference-free evaluations. The system achieves at least 50x faster inference and requires nearly 2x less memory than baselines, proving that tool-based editing can be modeled as conditional generation over structured continuous parameters without autoregressive reasoning.
Key points
- FlowTool uses a Diffusion Transformer parameter generator to transform Gaussian noise into an editing plan conditioned on input images and user instructions.
- FlowTool achieves significantly stronger reference-based performance than specialized MLLM editing agents and proprietary MLLMs across multiple benchmarks.
Context and limitations
Results are limited to four benchmark datasets (MMArt-Bench, FlowTool-Eval, ArtEdit-Bench, and MIT-Adobe5K) and represent preprint status with uncertainty regarding generalizability.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 21:04:27 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeidgq37gfahr3mvbkpx6unfonsmb3chbqxppyw2etg7zn66ya6kjnm