ChronoEditっていう、NVIDIAの画像生成AI。オープン。
いわゆるnanobananaみたいな画像編集が得意なヤツ。
お試しで色々やってみた。
元画像がImagen4で、それを「夜景」「水着」「走る」と変換。
nanobananaほどの自由度はまだちょっとわからないけど、一応今回のはちゃんと命令通り。

#AIgirl
#AIイラスト
#AIart

論文読んでこれがなんであるかわかった。ChronoEditって、ようは動画生成AI。例えば「車を反転させて」っていう命令を受けたとき、車が反転する動画を作る。そして最後のフレームだけを使う。こうすればただの画像編集のような意味不明な物理現象をする絵が作られないよ、ってこと。
逆に言えば、動画になりにくい命令を与えてもあんまり上手くいかないんじゃないかな。僕の画像の場合、水着に変換とか。走るもあんまり良くないね。実験が必要だ。

Apache-2.0 LoRA, sub-100MB, drops into Diffusers and rides on ChronoEdit-14B — frictionless install for anyone already on that stack. The model card is the problem: no sample outputs, no VRAM numbers, and a 2026 date that looks like a stub. "Artistic flair" is a vibe, not a claim, until...

14B on Apache 2.0 with a diffusers-compatible checkpoint is a real draw for commercial image-editing pipelines, Ascend angle aside. But 22 downloads, a near-empty model card, and zero benchmarks make this a wait-and-see — hard to recommend over Qwen-Image-Edit until someone publishes...

Apache-2.0 with LoRA fine-tunes baked in is the part that matters here — a 14B upscaler you can actually ship on Ascend without a licence fight is rare. The model card dodges the obvious questions though: no benchmarks against Real-ESRGAN or SeedVR, no training data, and the "Chrono" branding...

Tried a ChronoEdit, it got picked up as Qwen. ChronoEdit is based on Wan

Tried a Chroma, it got picked up as....... zimage, which is interesting :v I expected Flux

Wu, Ren, Shen, Cao, He, Lu, Gao, Xie, Lan, Alvarez, Gao, Fidler, Wang, Ling: ChronoEdit: Towards Temporal Reasoning for Image Editing and World Simulation https://arxiv.org/abs/2510.04290 https://arxiv.org/pdf/2510.04290 https://arxiv.org/html/2510.04290

[2025-10-20] 📚 Updates in #DiffM

(1) ChronoEdit: Towards Temporal Reasoning for Image Editing and World Simulation
(2) SPICE: A Synergistic, Precise, Iterative, and Customizable Image Editing Workflow

🔍 More at researchtrend.ai/communities/DiffM

[2025-10-20] 📚 Updates in #VGen

(1) Genie Envisioner: A Unified World Foundation Platform for Robotic Manipulation
(2) ChronoEdit: Towards Temporal Reasoning for Image Editing and World Simulation

🔍 More at researchtrend.ai/communities/VGen

ChronoEdit frames edits as a short video with temporal reasoning to enforce physical consistency, and will open‑source 14‑billion‑ and 2‑billion‑parameter models. https://getnews.me/chronoedit-introduces-temporal-reasoning-for-physically-consistent-image-editing/ #chronoedit #temporalreasoning

research.nvidia.com/labs/toronto...
ChronoEdit: image editing (Nvidia research)
Nano banana like but open source.

Jay Zhangjie Wu, Xuanchi Ren, Tianchang Shen, Tianshi Cao, Kai He, Yifan Lu, Ruiyuan Gao, Enze Xie, Shiyi Lan, Jose M. Alvarez, Jun Gao, Sanja Fidler, Zian Wang, Huan Ling
ChronoEdit: Towards Temporal Reasoning for Image Editing and World Simulation
https://arxiv.org/abs/2510.04290

@rohanpaul_ai https://x.com/rohanpaul_ai/status/1990482337977807119 #x-rohanpaul_ai 🎥

🖼️ @nvidia just released ChronoEdit-14B-Diffusers-Paint-Brush-Lora on @huggingface

ChronoEdit treats image editing as a tiny video generation task so edits follow consistent physics and keep object...