{"slug":"image-editing","label":"Image Editing","item_count":4,"day_count":4,"source_count":2,"first_seen":"2026-10-23T13:46:53+00:00","last_updated":"2026-11-04T16:23:24+00:00","generated_at":"2026-11-06T00:13:28.162297+00:00","sources":["arxiv_cs_ai","arxiv_cs_lg"],"days":[{"date":"2026-10-23","items":[{"title":"EditJev: System One Routing for Image Editing Models","url":"http://arxiv.org/abs/2610.32635v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Instruction-based image editing models exhibit pronounced sample-specific performance variation: different models can produce substantially different results for the same source image and editing instruction, and the...","sid":"d1fe1050c29bd1ca","published":"2026-10-23T13:46:53+00:00","editor_note":"Frames model routing as the answer to sample-specific editing quality variation."}]},{"date":"2026-10-27","items":[{"title":"MoI-Bench: A Benchmark for Image Generation and Editing on Mobile Devices","url":"http://arxiv.org/abs/2610.37599v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Developing on-device image models requires considering practical user needs, task types and device resource constraints. However, existing benchmarks provide limited guidance on identifying capability gaps between on-...","why_it_matters":"Matches feed focus: evaluation.","sid":"f1594d4deb389b81","published":"2026-10-27T17:31:40+00:00","editor_note":"Adds a benchmark for on-device image generation and editing under resource limits."}]},{"date":"2026-10-30","items":[{"title":"RetryRoute: Retry-Aware Model Routing for Agentic Image Editing Services","url":"http://arxiv.org/abs/2610.41460v1","source":"arxiv_cs_lg","type":"paper","summary_1line":"Web agents increasingly edit images by calling open editing models in a critique-and-retry loop, and these models differ in cost by more than an order of magnitude. After every rejected edit, a service must decide whe...","why_it_matters":"Matches feed focus: agentic.","sid":"71f1f68adb68c378","published":"2026-10-30T09:12:21+00:00","editor_note":"Adds retry-aware routing to cut cost in critique-and-retry editing loops."}]},{"date":"2026-11-04","items":[{"title":"TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing","url":"http://arxiv.org/abs/2611.06741v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Strong single-turn editing performance does not guarantee reliable multi-turn interaction: as edited outputs are repeatedly reused, small unintended changes can accumulate into severe visual drift. This long-term drif...","sid":"ab758b1d2464847a","published":"2026-11-04T16:23:24+00:00","editor_note":"Shifts focus to cumulative visual drift across repeated edits."}]}],"editorial":{"tldr":"Papers began treating image editing as a serving problem: EditJev routed each request across editing models because per-sample quality varies widely. MoI-Bench then benchmarked on-device editing, and RetryRoute added cost-aware routing for critique-and-retry loops.","stale":false,"whats_new":"TailSteer (Nov 4) targets multi-turn editing, where small unintended changes accumulate into visual drift even when single-turn quality is strong.","why_it_matters":"Editing models differ in cost by over an order of magnitude and in per-sample quality, so routing, retry policy and multi-turn drift become reliability and cost levers for any service that edits images.","take_for_builders":"If you run an edit-and-retry loop, log per-model reject rates and cost per accepted edit now, and cap multi-turn chains or re-anchor to the original image until drift fixes are independently verified.","beats":[{"kicker":"ROUTING","tone":"launch","headline":"Per-sample variation motivates routing across editing models","sids":["d1fe1050c29bd1ca"]},{"kicker":"DEVICES","tone":"rising","headline":"MoI-Bench measures image generation and editing on mobile hardware","sids":["f1594d4deb389b81"]},{"kicker":"COST","tone":"turn","headline":"RetryRoute makes routing retry-aware for agentic editing services","sids":["71f1f68adb68c378"]},{"kicker":"NOW","tone":"now","headline":"TailSteer attacks drift in multi-turn editing","sids":["ab758b1d2464847a"]}],"open_questions":["Do any of these routing or steering methods get reproduced on independent editing benchmarks?","Will MoI-Bench results show on-device models closing the gap with server-side editors?","Does TailSteer's drift reduction hold across editing models beyond those it tested?"],"generated_at":"2026-11-06T00:13:00+00:00"}}