Research map / Reinforcement learning / RL from feedback
FlowLess: Controlling Abstract Image Generation
Amir Hertz, Noah Snavely
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Preserve residuals instead of Free-use residuals on abstraction embeddings (ablation) | FlowLess · Abstraction-set generation (1,000 images) | set alignment vs text alignment: set alignment 0.475 -> 0.495; text alignment 0.196 -> 0.172 | Mixed |
| increase coverage score from 0.1 to 1.0 (ablation) | FlowLess · Abstraction-set generation (1,000 images) | text vs set alignment | Mixed |
| train on SVG data only instead of mixed SVG + raster data (ablation) | FlowLess · Abstraction-set generation (1,000 images) | image quality / alignment | Hurt |
| train on rasterized data only instead of mixed data (ablation) | FlowLess · Abstraction-set generation (1,000 images) | alignment at high coverage score | Hurt |
| use mixed SVG (100K) and rasterized (50K) dataset (ablation) | FlowLess · Abstraction-set generation (1,000 images) | abstraction and image quality | Improved |
| use FlowLess Preserve instead of Flux.2-dev + MMPE | FlowLess · Abstraction-set generation (1,000 images) | set alignment: 0.467 -> 0.495 | Improved |
| use FlowLess Free-use instead of Flux.2-dev | FlowLess · Abstraction-set generation (1,000 images) | text alignment: 0.239 -> 0.196 | Hurt |
| add inter-relation detail close-ups to abstraction set (ablation) | FlowLess · Control by details (qualitative) | composition control | Improved |
Closest papers by meaning
- Flow-GRPO: Training Flow Matching Models via Online RL (NeurIPS 2025 · similarity 0.90)
- Flow-DPPO: Divergence Proximal Policy Optimization for Flow Matching Models (NeurIPS 2026 · similarity 0.88)
- David and Goliath: Small One-step Model Beats Large Diffusion with Score Post-training (ICML 2025 · similarity 0.88)
- Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation (NeurIPS 2025 · similarity 0.88)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · similarity 0.88)
- PRECISE: SDE-Consistent Stochastic Sampling for RL Post-Training of Flow-Matching Models (NeurIPS 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
