EraseAnything++: Enabling Concept Erasure in Rectified Flow Transformers Leveraging Multi-Object Optimization

Zhaoxin Fan; Nanxiang Jiang; Daiheng Gao; Shiji Zhou; Wenjun Wu

cs.CVMar 1, 2026

EraseAnything++: Enabling Concept Erasure in Rectified Flow Transformers Leveraging Multi-Object Optimization

Authors: Zhaoxin Fan, Nanxiang Jiang, Daiheng Gao, Shiji Zhou, Wenjun Wu

Organizations: Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, School of Artificial Intelligence, Beihang University. · University of Science and Technology of China.

Abstract

Removing undesired concepts from large-scale text-to-image (T2I) and text-to-video (T2V) diffusion models while preserving overall generative quality remains a major challenge, particularly as modern models such as Stable Diffusion v3, Flux, and OpenSora employ flow-matching and transformer-based architectures and extend to long-horizon video generation. Existing concept erasure methods, designed for earlier T2I/T2V models, often fail to generalize to these paradigms. To address this issue, we propose EraseAnything++, a unified framework for concept erasure in both image and video diffusion models with flow-matching objectives. Central to our approach is formulating concept erasure as a constrained multi-objective optimization problem that explicitly balances concept removal with preservation of generative utility. To solve the resulting conflicting objectives, we introduce an efficient utility-preserving unlearning strategy based on implicit gradient surgery. Furthermore, by integrating LoRA-based parameter tuning with attention-level regularization, our method anchors erasure on key visual representations and propagates it consistently across spatial and temporal dimensions. In the video setting, we further enhance consistency through an anchor-and-propagate mechanism that initializes erasure on reference frames and enforces it throughout subsequent transformer layers, thereby mitigating temporal drift. Extensive experiments on both image and video benchmarks demonstrate that EraseAnything++ substantially outperforms prior methods in erasure effectiveness, generative fidelity, and temporal consistency, establishing a new state of the art for concept erasure in next-generation diffusion models.

Figures & tables

Fig. 1: Illustration of Difference between EraseAnything++ and Existing Methods . In this paper, we introduce EraseAnything++, an advanced concept erasure technique for Flow-based T2I/T2V images. First row : Classical concept-erasing methods—ESD [ 10 ] , CA [ 11 ] , and EAP [ 12 ] —have been transplanted into Flux.1 [dev] and are tested with the input “ $\mathtt{nudity}$ ” ( blue bars indicate author-added sensory harmony). Second row : Visualizing EraseAnything++’s impact—pre and post-concept removal. Original outputs ( yellow box) are displayed in the upper right.

Fig. 2: Geometric interpretation of the proposed gradient surgery. (Left) When the erasure gradient $\nabla\mathcal{L}_{e}$ lies within the safe region (non-conflicting or satisfying the tolerance), the update direction $\boldsymbol{d}_{t}^{*}$ remains unchanged. (Right) When a significant conflict arises, $\boldsymbol{d}_{t}^{*}$ is obtained by projecting $\nabla\mathcal{L}_{e}$ to satisfy the preservation constraint. The tolerance $\varepsilon_{t}$ creates a “tolerance cone” (dashed yellow line) that relaxes strict orthogonality, preventing optimization deadlocks while ensuring controlled utility preservation.

Method	Top-3 closest synonyms
Claude 3.5	”naked”, ”undressed”, ”unclothed”
GPT-4o	”bare”, ”naked”, ”unclothed”
Kimi	”naked”, ”unclothed”, ”bare”
T5 feature	’lean’, ’deer’, ’girl’

TABLE I: Comparison of synonym discovery for nude .

Fig. 3: Correlations between text and attention maps in Flux. Despite the lack of explicit cross-attention layers, a linear relationship persists between text embeddings and the unified attention map.

Fig. 4: Limitations of Attention Map erasure. Zeroing out attention columns ( $\mathbf{W_{attn}}[:,:,idx_{i}]=0$ ) works for simple prompts but fails against adversarial attacks like keyword obfuscation ( e.g. , ”soccerrs”).

Fig. 5: Single concept erasure . We compare our model with mainstream concept erasing methods. The Attention Map column shows the simple token localization method from Section IV that erases target concept effectively, yet its vulnerable to the minor change of tokens, making it difficult to widely adopt in practical applications.

Fig. 6: User Study results . EraseAnything++ yields the best overall performance, balancing erasure effectiveness with the preservation of image quality and prompt adherence.

	Detected Nudity (Quantity)									MS-COCO 10K
Method	Armpits	Belly	Buttocks	Feet	Breasts (F)	Genitalia (F)	Breasts (M)	Genitalia (M)	Total $\downarrow$	FID $\downarrow$	CLIP $\uparrow$
Flux.1 [dev]	147	168	29	62	152	9	28	10	605	21.32	30.87
CA (Model-based) [ 11 ]	86	103	26	38	57	8	20	6	344	22.66	29.05
CA (Noise-based) [ 11 ]	91	122	37	40	61	11	22	6	390	23.07	28.73
ESD-x [ 10 ]	107	131	22	69	138	7	24	8	506	23.08	28.44
ESD-u [ 10 ]	81	76	5	48	69	8	20	5	312	22.48	28.07

TABLE II: Quantitative comparison on nudity erasure. (Left) Count of explicit body parts detected by NudeNet on 4,703 I2P prompts. (Right) Zero-shot generation quality on MS-COCO 10K. The performance of the original Flux.1 [dev] is presented for reference.

Method	Acc e $\downarrow$	Acc ir $\uparrow$	H a $\uparrow$	FID $\downarrow$	CLIP $\uparrow$
CA (Model-based) [ 11 ]	29.26	28.54	-0.72	22.63	29.13
FMN [ 67 ]	29.63	28.90	-0.73	24.32	27.66
ESD-x [ 10 ]	20.89	21.21	0.32	22.89	28.12
ESD-u [ 10 ]	19.66	19.55	-0.11	22.06	27.94
UCE [ 20 ]	21.31	25.70	4.39	30.22	24.08
EAP [ 12 ]	21.10	26.34	5.24	22.84	29.25

TABLE III: Quantitative Results on Artistic Style Erasure. We evaluate performance on the 200-artist dataset [ 44 ] . We report the CLIP alignment score for erased styles (

\textsc{Acc}_{e}

) and preserved styles (

\textsc{Acc}_{ir}

). The overall score is defined as

H_{a}=\textsc{Acc}_{ir}-\textsc{Acc}_{e}

(Higher is better). General image quality is measured by FID and CLIP on MS-COCO.

	Entity				Abstraction				Relationship
Method	Acc e $\downarrow$	Acc ir $\uparrow$	H a $\uparrow$	Acc g $\downarrow$	Acc e $\downarrow$	Acc ir $\uparrow$	H a $\uparrow$	Acc g $\downarrow$	Acc e $\downarrow$	Acc ir $\uparrow$	H a $\uparrow$	Acc g $\downarrow$
CA (Model-based) [ 11 ]	14.8	24.2	9.4	27.3	25.8	23.1	-2.7	30.5	22.7	23.6	0.9	23.1
FMN [ 67 ]	14.9	23.7	8.8	27.6	25.4	22.5	-2.9	31.3	23.4	23.5	0.1	23.4
ESD-x [ 10 ]	14.5	25.3	10.8	26.1	26.1	23.4	-2.7	28.4	22.3	23.9	1.6	22.9
ESD-u [ 10 ]	12.8	24.7	11.9	25.8	22.8	21.3	-1.5	28.1	20.5	22.2	1.7	22.0
UCE [ 20 ]	12.4	19.6	7.2	27.5	20.4	18.0	-2.4	30.2	18.9	17.6	-1.3	23.8

TABLE IV: Evaluation of Erasing the specific category: Entity ( e.g. soccer), Abstraction ( e.g. artistic style) and Relationship ( e.g. kiss) are presented. CLIP classification accuracies are reported for each erased category in three sets: the erased category itself (Acc e , efficacy), the remaining unaffected categories (Acc ir , specificity) and synonyms of the erased class (Acc g , generality). All presented values are denoted in percentage (%).

Method	Unlearn DiffAtk [ 68 ]	Ring-A-Bell [ 69 ]	Ring-A-Bell Union [ 69 ]	ReFlux [ 70 ]
CA [ 11 ]	85.32	27.50	44.03	87.16
FMN [ 67 ]	92.66	32.11	46.78	-
ESD-x [ 10 ]	76.14	11.01	24.77	8 6.24
ESD-u [ 10 ]	7 0.64	1 8.34	2 9.36	89.91
UCE [ 20 ]	77.98	22.93	2 9.36	-
EAP [ 12 ]	98.16	35.77	45.87	99.08

TABLE V: Robustness Evaluation. Comparison of erasure performance under various prompt attacks (lower is better).

	Nudity Rate (%) $\downarrow$
Method	Gen [ 21 ]	Ring-A-Bell [ 69 ]	Object Class $\uparrow$	Subject Consistency $\uparrow$
Open-Sora	58.56	28.63	9 0.52	95.80
SAFREE [ 71 ]	32.12	13.17	48.48	94.92
NegPrompt [ 72 ]	36.00	11.75	91.94	93.45
VideoErasure [ 22 ]	21.45	10.13	80.62	93.77
T2VUnlearning [ 21 ]	1 9.73	6.97	87.00	94.70

TABLE VI: Quantitative Results on Video Nudity Erasure. Evaluated on Open-Sora. Nudity Rate (Lower is better) measures erasure success. Object Class and Subject Consistency (Higher is better) measure video quality and coherence.

Fig. 7: Multi concept erasure . Multiple concepts can be removed simultaneously by a normalized linear weight composition of their LoRA updates.

Fig. 8: Qualitative comparison of nudity erasure on Open-Sora. Given the explicit prompt “ A beautiful girl takes off her top to reveal her bare breasts ”, the original model generates NSFW content (masked for safety). In contrast, EraseAnything++ successfully sanitizes the video by removing the nudity while preserving the main subject and maintaining temporal consistency across frames.

Method	ESR-1 $\uparrow$	ESR-5 $\uparrow$	PSR-1 $\uparrow$	PSR-5 $\uparrow$
Open-Sora	21.62	5.09	78.38	94.91
NegPrompt [ 72 ]	48.59	19.79	65.37	88.62
SAFREE [ 71 ]	61.65	36.41	53.46	79.17
T2VUnlearning [ 21 ]	92.38	77.09	54.03	82.14
EraseAnything++	94.15	79.20	7 6.45	9 3.10

TABLE VII: Results of ImageNet Object Erasure on Open-Sora. We compare the Erasure Success Rate (ESR) and Preservation Success Rate (PSR). Higher values indicate better performance for both metrics.

Fig. 9: Ablation of different loss configs. Our full method achieves the most effective equilibrium between concept erasure and semantic preservation.

Config	Acc e $\downarrow$	Acc ir $\uparrow$	H a $\uparrow$
$\mathcal{L}_{esd}$ + $\mathcal{L}_{attn}$	19.2	23.5	4.3
$\mathcal{L}_{esd}$ + $\mathcal{L}_{lora}$	26.8	28.8	2.0
$\mathcal{L}_{esd}$ + $\mathcal{L}_{rsc}$	21.5	27.1	5.6
$\mathcal{L}_{attn}$ + $\mathcal{L}_{rsc}$	23.1	26.5	3.4
$\mathcal{L}_{attn}$ + $\mathcal{L}_{lora}$ + $\mathcal{L}_{rsc}$	21.8	27.9	6.1
Full (EraseAnything++)	20.1	28.2	8.1

TABLE VIII: Ablation on Loss Components. We report CLIP score of Erasure Accuracy ( Acc

{}_{e}\downarrow

), Preservation Accuracy ( Acc

{}_{ir}\uparrow

), and their difference score (

H_{a}=\textsc{Acc}_{ir}-\textsc{Acc}_{e}\uparrow

). While biased configurations may excel in a single metric, our full method achieves the best overall balance (

H_{a}

Fig. 10: Ablation of the Spatio-Temporal strategy. (Top) w/o Anchor: The target concept is not suppressed at initialization, leading to immediate leakage. (Middle) w/o Propagation: Although erased initially, the concept re-emerges in later frames due to temporal drift. (Bottom) Full Strategy: Our method ensures consistent erasure across the entire video duration.

Strategy	Acc e $\downarrow$	Acc ir $\uparrow$	H a $\uparrow$
Linear Sum ( $\lambda=0.1$ )	18.5	22.5	4.0
Linear Sum ( $\lambda=10.0$ )	28.6	28.9	0.3
PCGrad [ 51 ]	24.1	28.5	4.4
MGDA [ 50 ]	25.8	28.7	2.9
Ours	20.1	28.2	8.1

TABLE IX: Ablation on Optimization Strategy. Evaluated on CelebA (same as Table VIII ). General MOO solvers (PCGrad [ 51 ] , MGDA [ 50 ] ) tend to be conservative, prioritizing preservation at the cost of weak erasure. Our asymmetric strategy yields the best balance.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Category	# Number	Prompt template	Concepts
Entity	10	‘A photo of [ Entity ]’	‘Fruit’, ‘Ball’, ‘Car’, ‘Airplane’, ‘Tower’, ‘Building’, ‘Celebrity’, ‘Shoes’, ‘Cat’, ‘Dog’
Abstraction	10	‘A scene featuring [ Abstraction ]’	‘Explosion’, ‘Green’, ‘Yellow’, ‘Time’, ‘Two’, ‘Three’, ‘Shadow’, ‘Smoke’, ‘Dust’, ‘Environmental Simulation’
Relationship	10	‘A [ Relationship ] B’	‘Shake Hand’, ‘Kiss’, ‘Hug’, ‘In’, ‘On’, ‘Back to Back’, ‘Jump’, ‘Burrow’, ‘Hold’, ‘Amidst’

Appendix

TABLE X: Complete list of concepts of Entity, Abstraction, and Relationship.

Category	# Number	Celebrity
Erasure Group	50	‘ Adele ’, ‘ Albert Camus ’, ‘ Angelina Jolie ’, ‘ Arnold Schwarzenegger ’, ‘ Audrey Hepburn ’, ‘ Barack Obama ’, ‘ Beyoncé ’, ‘ Brad Pitt ’, ‘ Bruce Lee ’, ‘ Chris Evans ’, ‘ Christiano Ronaldo ’, ‘ David Beckham ’, ‘ Dr Dre ’, ‘ Drake ’, ‘ Elizabeth Taylor ’, ‘ Eminem ’, ‘ Elon Musk ’, ‘ Emma Watson ’, ‘ Frida Kahlo ’, ‘ Hugh Jackman ’, ‘ Hillary Clinton ’, ‘ Isaac Newton ’, ‘ Jay-Z ’, ‘ Justin Bieber ’, ‘ John Lennon ’, ‘ Keanu Reeves ’, ‘ Leonardo Dicaprio ’, ‘ Mariah Carey ’, ‘ Madonna ’, ‘ Marlon Brando ’, ‘ Mahatma Gandhi ’, ‘ Mark Zuckerberg ’, ‘ Michael Jordan ’, ‘ Muhammad Ali ’, ‘ Nancy Pelosi ’,‘ Neil Armstrong ’, ‘ Nelson Mandela ’, ‘ Oprah Winfrey ’, ‘ Rihanna ’, ‘ Roger Federer ’, ‘ Robert De Niro ’, ‘ Ryan Gosling ’, ‘ Scarlett Johansson ’, ‘ Stan Lee ’, ‘ Tiger Woods ’, ‘ Timothee Chalamet ’, ‘ Taylor Swift ’, ‘ Tom Hardy ’, ‘ William Shakespeare ’, ‘ Zac Efron ’
Retention Group	50	‘ Angela Merkel ’, ‘ Albert Einstein ’, ‘ Al Pacino ’, ‘ Batman ’, ‘ Babe Ruth Jr ’, ‘ Ben Affleck ’, ‘ Bette Midler ’, ‘ Benedict Cumberbatch ’, ‘ Bruce Willis ’, ‘ Bruno Mars ’, ‘ Donald Trump ’, ‘ Doraemon ’, ‘ Denzel Washington ’, ‘ Ed Sheeran ’, ‘ Emmanuel Macron ’, ‘ Elvis Presley ’, ‘ Gal Gadot ’, ‘ George Clooney ’, ‘ Goku ’,‘ Jake Gyllenhaal ’, ‘ Johnny Depp ’, ‘ Karl Marx ’, ‘ Kanye West ’, ‘ Kim Jong Un ’, ‘ Kim Kardashian ’, ‘ Kung Fu Panda ’, ‘ Lionel Messi ’, ‘ Lady Gaga ’, ‘ Martin Luther King Jr. ’, ‘ Matthew McConaughey ’, ‘ Morgan Freeman ’, ‘ Monkey D. Luffy ’, ‘ Michael Jackson ’, ‘ Michael Fassbender ’, ‘ Marilyn Monroe ’, ‘ Naruto Uzumaki ’, ‘ Nicolas Cage ’, ‘ Nikola Tesla ’, ‘ Optimus Prime ’, ‘ Robert Downey Jr. ’, ‘ Saitama ’, ‘ Serena Williams ’, ‘ Snow White ’, ‘ Superman ’, ‘ The Hulk ’, ‘ Tom Cruise ’, ‘ Vladimir Putin ’, ‘ Warren Buffett ’, ‘ Will Smith ’, ‘ Wonderwoman ’

Appendix

TABLE XI: Complete list of celebrities used in ablation study.

Fig. 11: User Study Interface on Imaging Quality, Prompt Adherence and Output Diversity .

Fig. 12: User Study Interface on Erasing Cleanliness and Irrelevant Preservation.

Concept Erasure

Concept erasure removes specified knowledge or attributes from models while preserving their ability to perform unrelated tasks.

See all 57 papers

FADE: Frame-Aware Diffusion-Transformer-based Multi-Concept Erasure for Video Unlearning
Oct 2, 2026Yuchen Li, Kaiyuan Deng, Chaoran Feng +2Diffusion Model Unlearning Machine Unlearning
Peking University · University of Arizona
EraseSAE: Surgical Concept Erasure in Text-to-Video Diffusion Models via Sparse Autoencoders
Sep 3, 2026Xinghao Wang, Dong Li, Wei Yu +5Video Diffusion Models Sparse Autoencoders
University of Science and Technology of China, Hefei, China · HiDream.ai Inc. · Anhui Province Key Laboratory of Digital Security, China
Erasing Thousands of Concepts: Towards Scalable and Practical Concept Erasure for Text-to-Image Diffusion Models
Apr 12, 2026Hoigi Seo, Byung Hyun Lee, Jaehyun Cho +2Safe T2I Generation Diffusion Model Unlearning
Dept. of Electrical and Computer Engineering, Seoul National University, Republic of Korea · IPAI, Seoul National University, Republic of Korea · INMC, Seoul National University, Republic of Korea

Oct 2, 2026J/K move · Enter open · S save

FADE: Frame-Aware Diffusion-Transformer-based Multi-Concept Erasure for Video Unlearning

Yuchen Li, Kaiyuan Deng, Chaoran Feng, Zhenyu Tang+1

Peking University · University of Arizona

Diffusion Model Unlearning Machine Unlearning T2V Generation

Open paper AskPDF arXiv

Text-to-video (T2V) diffusion models can reproduce copyrighted, violent, or explicit content, which motivates concept erasure: removing designated concepts from a pretrained model while preserving its behavior on everything else. Existing T2V erasure methods leave two problems open. Their frame-agnostic suppression can leave isolated frames in which an erased concept resurfaces, a frame-reactivation gap that clip-level averages obscure; and they are usually evaluated with one target concept or category at a time. We propose Frame-Aware Diffusion Erasure (FADE), a multi-concept video unlearning framework. FADE first applies a joint closed-form key/value edit that suppresses all target concepts, then trains per-concept frame-aware low-rank adapters whose strength is gated by the frame index and the denoising timestep to remove residual per-frame leakage. Each adapter is trained with the other targets' prompts as hard negatives, which keeps the concept-specific components of different adapters well separated, and a similarity-based soft router combines the adapters according to the prompt. With 16 concepts (objects, artistic styles, and nudity) erased from a single Wan2.1-T2V-1.3B backbone, FADE reduces the residual accuracy on the object benchmark to 4.9%, against 15.5% for the strongest of eight baselines, while keeping the VBench average within 0.9% of the unedited model. The ranking is unchanged under a VLM judge and a blinded human study, and the advantage over the strongest baseline carries over to prompts that combine several erased concepts, to 30 simultaneously erased celebrity identities, and to Wan2.1-T2V-14B, CogVideoX-2B, and HunyuanVideo-1.5.

Abstract

Concept Erasure

Explore similar work

FADE: Frame-Aware Diffusion-Transformer-based Multi-Concept Erasure for Video Unlearning

EraseSAE: Surgical Concept Erasure in Text-to-Video Diffusion Models via Sparse Autoencoders

Erasing Thousands of Concepts: Towards Scalable and Practical Concept Erasure for Text-to-Image Diffusion Models