Meta tags:
description=HY-WU: An Extensible Functional Neural Memory Framework and An Instantiation in Text-Guided Image Editing;
keywords=HY-WU, Text-Guided Image Editing, Functional Neural Memory, Tencent Hunyuan;

Headings (most frequently used words):

and, an, functional, neural, memory, evaluation, human, outfit, high, transfer, hy, wu, part, extensible, framework, instantiation, in, text, guided, image, editing, key, features, showcases, acknowledgments, bibtex, scalable, for, large, models, strong, preference, cross, domain, clothing, fusion, creative, cosplay, character, migration, fidelity, face, identity, seamless, virtual, try, on, quality, texture, synthesis, gsb,

Text of the page (most frequently used words):
and (12), the (10), source (5), #neural (5), memory (5), for (5), with (5), models (5), remains (5), part (4), functional (4), framework (4), image (4), tencent (4), human (4), extensible (3), instantiation (3), text (3), guided (3), editing (3), team (3), wang (3), evaluation (3), nano (3), banana (3), scale (3), model (3), high (3), large (3), instance (3), website (2), from (2), code (2), under (2), creative (2), arxiv (2), bibtex (2), hunyuan (2), substantially (2), open (2), closed (2), commercial (2), systems (2), while (2), modest (2), that (2), level (2), conditional (2), even (2), outfit (2), transfer (2), preference (2), strong (2), scalable (2), conditioned (2), adapter (2), without (2), adapted, following, this, licensed, commons, attribution, sharealike, international, license, article, wu2026hy, title, author, mengxuan, xuanlei, zhao, ziqiao, ruicheng, feng, atlas, qinglin, kai, journal, preprint, 2603, 07236, year, 2026, thank, their, support, project, acknowledgments, other, outperforms, leading, competitive, top, tier, pro, achieve, slightly, higher, overall, scores, respectively, margin, given, these, are, likely, trained, larger, backbones, proprietary, data, performance, gap, suggests, our, operator, adaptation, effective, more, constrained, gsb, quality, texture, synthesis, seamless, virtual, try, fidelity, face, identity, cosplay, character, migration, cross, domain, clothing, fusion, showcases, achieves, win, rates, against, exceeds, baselines, close, latest, series, practical, foundation, 80b, parameters, structured, parameter, tokenization, method, naturally, compatible, architectures, introduces, lightweight, generates, per, request, finetuning, enabling, personalization, preserving, base, general, capability, key, features, propose, fly, generation, low, rank, lora, updates, synthesizes, weights, hybrid, instruction, representations, injects, them, into, frozen, backbone, during, forward, pass, producing, specific, operators, test, time, optimization, prompt, handbook, twitter, report, huggingface, demo,

Text of the page (random words):
hy wu part i an extensible functional neural memory framework and an instantiation in text guided image editing hy wu part i an extensible functional neural memory framework and an instantiation in text guided image editing tencent hy team demo huggingface code report twitter prompt handbook bibtex we propose hy wu a scalable framework for on the fly conditional generation of low rank lora updates hy wu synthesizes instance conditioned adapter weights from hybrid image instruction representations and injects them into a frozen backbone during the forward pass producing instance specific operators without test time optimization key features functional neural memory hy wu introduces a lightweight neural memory for ai it generates conditioned model adapter per request without finetuning enabling instance level personalization while preserving the base model s general capability scalable for large models hy wu remains practical for large foundation models even at 80b parameters with structured parameter tokenization the method is naturally compatible with large scale architectures strong human preference hy wu achieves high human preference win rates against open source models exceeds strong closed source baselines and remains close to the latest nano banana series showcases cross domain clothing fusion creative cosplay and character outfit migration high fidelity face identity transfer seamless outfit transfer and virtual try on high quality texture synthesis evaluation gsb human evaluation hy wu substantially outperforms leading open source models and remains competitive with top tier closed source commercial systems while nano banana 2 and nano banana pro achieve slightly higher overall scores 52 4 and 53 8 respectively the margin remains modest given that these commercial systems are likely trained with substantially larger scale backbones and proprietary data the modest performance gap suggests that our operator level conditional adaptation remains effective even under more constrained model scale human evaluation with other models acknowledgments we thank the tencent hunyuan team for their support hy wu is part of the tencent hunyuan project bibtex article wu2026hy wu title hy wu part i an extensible functional neural memory framework and an instantiation in text guided image editing author tencent hy team mengxuan wu xuanlei zhao ziqiao wang ruicheng feng atlas wang qinglin lu and kai wang journal arxiv preprint arxiv 2603 07236 year 2026 this website is licensed under a creative commons attribution sharealike 4 0 international license website adapted from the following source code