Research map / Image generation / Editing and style transfer
Training-Free Multi-Concept Image Editing
Niki Maria Foteinopoulou, Ignas Budvytis, Stephan Liwicki
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 |
|---|---|---|---|
| combine ordered timesteps, regularisation and negative prompts over DDS (ablation) | CDS · InstructPix2Pix | CLIPScore: 0.305 -> 0.308 | Improved |
| add only strict ordered timesteps to DDS (ablation) | CDS · InstructPix2Pix | LPIPS: 0.264 -> 0.461 | Hurt |
| add only regularisation to DDS (ablation) | CDS · InstructPix2Pix | CLIPScore: 0.305 -> 0.291 | Hurt |
| add only regularisation to DDS (ablation) | CDS · InstructPix2Pix | LPIPS: 0.264 -> 0.081 | Improved |
| use CDS objective instead of PDS | CDS · InstructPix2Pix | CLIPScore: 0.298 -> 0.308 | Improved |
| use dynamic concept weighting instead of Merge | CDS · ComposLoRA Reality | LPIPS: 0.610 -> 0.474 | Improved |
| use dynamic concept weighting instead of Switch | CDS · ComposLoRA Reality | LPIPS: 0.473 -> 0.474 | No clear effect |
| use CDS instead of Composite on CLIPScore | CDS · ComposLoRA Reality | CLIPScore: 0.365 -> 0.359 | Hurt |
| use CDS instead of Switch | CDS · ComposLoRA Anime | LPIPS: 0.345 -> 0.314 | Improved |
| use smaller patch size (2x2 vs 16x16) in concept weighting (ablation) | CDS · ComposLoRA subset | CLIPScore: 0.332 -> 0.351 | Improved |
| lower temperature tau from 1 to 0.002 (ablation) | CDS · ComposLoRA subset | CLIPScore: 0.344 -> 0.351 | Improved |
| increase regularisation strength eta from 0.01 to 10 (ablation) | CDS · InstructPix2Pix | LPIPS: 0.407 -> 0.089 (CLIP 0.316 -> 0.299) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- InstructPix2Pix — InstructPix2Pix: Learning To Follow Image Editing Instructions (CVPR 2023) · better on 2 of 2 shared results
- DiffusionClip — DiffusionCLIP: Text-Guided Diffusion Models for Robust Image Manipulation (CVPR 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Concept Lancet: Image Editing with Compositional Representation Transplant (CVPR 2025 · similarity 0.90)
- LoRAShop: Training-Free Multi-Concept Image Generation and Editing with Rectified Flow Transformers (NeurIPS 2025 · similarity 0.88)
- PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor (CVPR 2024 · similarity 0.88)
- An Item Is Worth a Prompt: Versatile Image Editing with Disentangled Control (AAAI 2025 · similarity 0.88)
- Object-Aware Inversion and Reassembly for Image Editing (ICLR 2024 · similarity 0.88)
- Pattern Analogies: Learning to Perform Programmatic Image Edits by Analogy (CVPR 2025 · similarity 0.87)
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.
