Can We Perform Online RL for Image Editing without Editing Rewards?

• URL: http://arxiv.org/abs/2608.22780v1
• Date: Mon, 24 Aug 2026 04:08:52 GMT
• Title: Can We Perform Online RL for Image Editing without Editing Rewards?
• Abstract summary: We argue that the standard image editing dimensions have potential to be mapped to the text-to-image (T2I) reward space.<n>We introduce Lever-Edit, a framework that learns a reward-aligned captioner for counterfactual target descriptions.<n>Experiments show competitive editing alignment and source preservation against editing-reward-based fine-tuning, while outperforming intuitive transfer baselines.
• Score: 56.482951092646026
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Reinforcement learning (RL) enables direct preference optimization for image editing through editing-specific rewards, which remain less developed due to costly triplet supervision and complex task-dependent calibration. In contrast, text-to-image (T2I) generation benefits from a mature and diverse reward ecosystem spanning semantic alignment, aesthetics, realism, glyph shape, and other visual preferences. Extending this ecosystem to image editing would substantially broaden the range of visual preferences accessible to RL-based optimization, prompting the central question: \emph{Can We Perform Image Editing RL without Editing Rewards?} In this paper, we argue that the standard image editing dimensions have potential to be mapped to the T2I reward space: image quality can transfer directly, prompt following can be aligned through a description of the desired visual state, and reference consistency admits a coarse semantic conversion by encoding the source content to preserve. However, editing instructions specify relative changes, whereas T2I rewards require self-contained target descriptions; moreover, semantically valid captions from generic vision-language models may be incompatible with the frozen reward. Hence, we further introduce Lever-Edit, a two-stage framework that learns a reward-aligned captioner for counterfactual target descriptions, freezes it, and optimizes the editing policy solely with the transferred T2I reward. Experiments show competitive editing alignment and source preservation against editing-reward-based fine-tuning, while outperforming intuitive transfer baselines.

Related papers

• Model the Edit, Not the Image: Visual Autoregressive Editing from a Source-Centric Perspective [23.408232115772833]
  Next-scale visual autoregressive models ( VARs) have emerged as a powerful generative paradigm, producing high-quality images through efficient coarse-to-fine prediction.<n>We propose textbfEditMod, which compares source- and target-conditioned predictions under a shared autoregressive context.<n>Experiments show that EditMod achieves leading source-image fidelity while maintaining strong text alignment, and completes end-to-end editing of a 1K image in only 1.57 seconds on a single A100 GPU.
  arXiv  Detail & Related papers  (2026-08-10T03:05:09Z)
• ReEdit: Multimodal Exemplar-Based Image Editing with Diffusion Models [11.830273909934688]
  Modern Text-to-Image (T2I) Diffusion models have revolutionized image editing by enabling the generation of high-quality images. We propose ReEdit, a modular and efficient end-to-end framework that captures edits in both text and image modalities. Our results demonstrate that ReEdit consistently outperforms contemporary approaches both qualitatively and quantitatively.
  arXiv  Detail & Related papers  (2024-11-06T15:19:24Z)
• E4C: Enhance Editability for Text-Based Image Editing by Harnessing Efficient CLIP Guidance [13.535394339438428]
  Diffusion-based image editing is a composite process of preserving the source image content and generating new content or applying modifications. We propose a zero-shot image editing method, named textbfEnhance textbfEditability for text-based image textbfEditing via textbfCLIP guidance.
  arXiv  Detail & Related papers  (2024-03-15T09:26:48Z)
• DiffEditor: Boosting Accuracy and Flexibility on Diffusion-based Image Editing [66.43179841884098]
  Large-scale Text-to-Image (T2I) diffusion models have revolutionized image generation over the last few years. We propose DiffEditor to rectify two weaknesses in existing diffusion-based image editing. Our method can efficiently achieve state-of-the-art performance on various fine-grained image editing tasks.
  arXiv  Detail & Related papers  (2024-02-04T18:50:29Z)
• Optimisation-Based Multi-Modal Semantic Image Editing [58.496064583110694]
  We propose an inference-time editing optimisation to accommodate multiple editing instruction types. By allowing to adjust the influence of each loss function, we build a flexible editing solution that can be adjusted to user preferences. We evaluate our method using text, pose and scribble edit conditions, and highlight our ability to achieve complex edits.
  arXiv  Detail & Related papers  (2023-11-28T15:31:11Z)
• StyleDiffusion: Prompt-Embedding Inversion for Text-Based Editing [115.49488548588305]
  A significant research effort is focused on exploiting the amazing capacities of pretrained diffusion models for the editing of images.<n>They either finetune the model, or invert the image in the latent space of the pretrained model.<n>They suffer from two problems: Unsatisfying results for selected regions and unexpected changes in non-selected regions.
  arXiv  Detail & Related papers  (2023-03-28T00:16:45Z)
• Zero-shot Image-to-Image Translation [57.46189236379433]
  We propose pix2pix-zero, an image-to-image translation method that can preserve the original image without manual prompting. We propose cross-attention guidance, which aims to retain the cross-attention maps of the input image throughout the diffusion process. Our method does not need additional training for these edits and can directly use the existing text-to-image diffusion model.
  arXiv  Detail & Related papers  (2023-02-06T18:59:51Z)
• EditGAN: High-Precision Semantic Image Editing [120.49401527771067]
  EditGAN is a novel method for high quality, high precision semantic image editing. We show that EditGAN can manipulate images with an unprecedented level of detail and freedom. We can also easily combine multiple edits and perform plausible edits beyond EditGAN training data.
  arXiv  Detail & Related papers  (2021-11-04T22:36:33Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.