Research map / Image generation / Editing and style transfer
DiTailed: Ensuring Visual Object Consistency in Text-Image-to-Image Flow Matching Models
Francesco Taioli, Daniel Coelho, Iaroslav Melekhov, Roberto Alcover-Couso, Jose Saiz, Virginia Arguedas, Artur Bekasov
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 |
|---|---|---|---|
| add FlowMirror auxiliary loss to fine-tuning | Qwen-Image-Edit-2509 · ABO-Edit validation | FID: 26.48 -> 25.91 | Improved |
| add FlowMirror auxiliary loss to fine-tuning | Qwen-Image-Edit-2509 · ABO-Edit validation | PSNR: 19.64 -> 19.79 | Improved |
| add FlowMirror auxiliary loss to fine-tuning | Qwen-Image-Edit-2509 · ABO-Edit validation | CLIP / DINOv2: 97.21 -> 97.17 (CLIP), 93.98 -> 93.94 (DINOv2) | No clear effect |
| add FlowMirror auxiliary loss to fine-tuning | FLUX.1 · ABO-Edit validation | FID: 31.68 -> 31.26 | Improved |
| add FlowMirror auxiliary loss to fine-tuning | FLUX.1 · ABO-Edit validation | DINOv2: 90.99 -> 91.25 | Improved |
| fine-tune on ABO-Edit versus zero-shot | Qwen-Image-Edit-2509 · ABO-Edit validation | FID: 57.66 -> 26.48 | Improved |
| fine-tune on ABO-Edit for orientation control | Qwen-Image-Edit-2509 · Azimuth error, left view | azimuth error (deg): 39.1 -> 2.4 | Improved |
| add VAE channel-wise mean/std matching of conditioning embedding (ablation) | FlowMirror · ABO-Edit validation | FID: 25.18 -> 26.15 | Hurt |
| normalize aux loss by number of scales (ablation) | FlowMirror · ABO-Edit validation | FID: 26.52 -> 26.15 | Improved |
| reduce lambda_aux from 0.1 to 0.01 (ablation) | FlowMirror · ABO-Edit validation | FID: 25.93 -> 25.55 | Improved |
| add high-frequency term to aux loss (ablation) | FlowMirror · ABO-Edit validation | FID: 27.07 -> 25.55 | Improved |
| increase lambda_HF from 0.2 to 1.0 (ablation) | FlowMirror · ABO-Edit validation | FID | Hurt |
Closest papers by meaning
- SplitFlow: Flow Decomposition for Inversion-Free Text-to-Image Editing (NeurIPS 2025 · similarity 0.88)
- FluxSpace: Disentangled Semantic Editing in Rectified Flow Models (CVPR 2025 · similarity 0.88)
- Safe Few-Step Generation via Velocity Editing (NeurIPS 2026 · similarity 0.87)
- Improving Editability in Image Generation with Layer-wise Memory (CVPR 2025 · similarity 0.87)
- Preserve or Modify? Context-Aware Evaluation for Balancing Preservation and Modification in Text-Guided Image Editing (CVPR 2025 · similarity 0.87)
- Contextual Flow Matching: Adaptive Step Selection in Flow Models for Efficient Visual Generation (NeurIPS 2026 · similarity 0.86)
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.
