Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance

Recent controllable generation approaches such as FreeControl and Diffusion Self-Guidance bring fine-grained spatial and appearance control to text-to-image (T2I) diffusion models without training auxiliary modules. However, these methods optimize the latent embedding for each type of score function with longer diffusion steps, making the generation process time-consuming and limiting their flexibility and use. This work presents Ctrl-X, a simple framework for T2I diffusion controlling structure and appearance without additional training or guidance. Ctrl-X designs feed-forward structure control to enable the structure alignment with a structure image and semantic-aware appearance transfer to facilitate the appearance transfer from a user-input image. Extensive qualitative and quantitative experiments illustrate the superior performance of Ctrl-X on various condition inputs and model checkpoints. In particular, Ctrl-X supports novel structure and appearance control with arbitrary condition images of any modality, exhibits superior image quality and appearance transfer compared to existing works, and provides instant plug-and-play functionality to any T2I and text-to-video (T2V) diffusion model. See our project page for an overview of the results: https://genforce.github.io/ctrl-x

Paper

References (45)

Scroll for more · 33 remaining

Similar papers

Peer review

Reviewer YZfZ5/10 · confidence 3/52024-07-12

Summary

This paper presents a training-free and guidance-free method for controllable image/video generation with structure and appearance control. Specifically, Ctrl-X injects structural and appearance features directly into the noised samples via cross-attention. Compared with other baselines for structural and appearance control, the proposed method achieves good appearance alignment and structure presentation.

Strengths

The proposed method is a good complement to training-based and guidance-based controllable visual generation methods. Experiment results also shown the effectiveness of the proposed method.

Weaknesses

- It looks like the generated images/videos are a little bit painting style. I guess this is caused by injecting structure features directly without training/fine-tuning. - The training-based baselines are not very strong. Combining two existing models (e.g., ControlNet+IP-Adapter, T2I-Adapter+IP-Adapter) might not be a fair comparison, since they are not specifically designed for both appearance and structure control. I would suggest compare structural preservation ability with ControlNet (SD1/2/3, SDXL) and T2I-Adapter, and then compare appearance control ability with IP-Adapter.

Questions

My understanding is that for training-free methods, we are making a trade-off between control quality and training computation cost. There exists some efficient training-based methods for controllable generation [1,2]. A natural question is: compared with such training-based method, is the training-free property good enough for us to use Ctrl-X? [1] Ran, Lingmin, et al. "X-adapter: Adding universal compatibility of plugins for upgraded diffusion model." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024. [2] Lin, Han, et al. "Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse Controls to Any Diffusion Model." arXiv preprint arXiv:2404.09967 (2024).

Rating

5

Confidence

3

Soundness

2

Presentation

3

Contribution

2

Limitations

NA

Reviewer 6fgc5/10 · confidence 5/52024-07-12

Summary

This paper proposes a training-free framework (Ctrl-X)to control the structure and appearance when diffusion generation without any training. The method does not need much more inference time cost or GPU resource cost. The insight is that diffusion feature maps capture rich spatial structure and high-level appearance from early diffusion steps sufficient for structure and appearance control without guidance. The experiments demonstrates superior results compared to previous training-based and guidance-based baselines (e.g. ControlNet + IP-Adapter [4, 5] and FreeControl [2]) in terms of condition alignment, text-image alignment, and image quality.

Strengths

1. The method is novel and the motivation is clear. 2. The writing is good and easy to follow.

Weaknesses

The main concern to me is the performance of the proposed method. 1. The authors do not provide any User Study results. They only show the quantitative results evaluated by DINO Self-sim and DINO-CLS, which are not widely used. In supp, they report the results in CLIP score, LPIPS in Table 3. From the result shown in Table 3, the proposed method gets a bad performance compared to others. For example, Ctrl-X gets the worst Self-sim performance. 2. The qualitative comparison is also unsatisfactory. For example, in Figure 5, I observe that ControlNet+ IP-Adapter gets a better result than Ctrl-X, e.g, the 3rd rows.

Questions

See the weaknesses.

Rating

5

Confidence

5

Soundness

3

Presentation

3

Contribution

2

Limitations

Yes.

Reviewer DRHE6/10 · confidence 4/52024-07-12

Summary

This paper introduces a method for controllable generation using diffusion models. The approach is designed as a training free technique for 1) structure/layout controlled generation (like e.g. controlNet) and 2) appearance transfer. The approach leverages manipulation of attention mechanisms and information transfer from reference images to the generative image. Authors evaluate their methods in multiple settings, and compare to state of the art related work, showing that their approach can achieve similar performance to more expensive alternatives.

Strengths

The paper is, for the most part, well written and well grounded in the related literature. Relevant related works are well references, and works from which ideas are borrowed are acknowledged. The main benefit of the work is its fully training-free nature, which can offer more flexibility when generating images with different types of controls. The proposed methodology is relatively simple, leaving room for future improvements. Another benefit is the fact that the method can do both appearance and structure control, while related works often focus on a single one. Experiments show promising results, with a performance similar to FreeControl, but without inference time optimization steps. Some ablations experiments are provided, in an effort to analyse the different components of the model.

Weaknesses

The main limitation of the work is the limited methodological novelty. The tools employed are not very novel: the structure transfer simply uses the method proposed in [34], while attention map manipulation is very commonly used for structure control (e.g. for editing methods such as prompt-to-prompt). Another noticeable limitation, highlighted in Figure 9a, is the lack of flexibility with regards to using appearance OR structure control. Results show that appearance control is required when performing structure control, requiring to generate a separate appearance image. This can increase generation cost, and reduces control over the content of the image, as the appearance image is simply controlled by a prompt. While the training free nature of the approach allows to use different types of structure images, the approach seems limited to control types with exact edge definition. Higher-level constraints like pose or bounding boxes do not appear to be an option with this type of method. While the experiments on video generation are interesting, I would recommend that the authors focus more space on structure-only generation (as in the appendix) and expand the ablation and limitation experiments. For example, studying the impact on the quality of generated appearance image for structure-only generation would be beneficial.

Questions

-Can the methodology handle pose or bounding boxes types of controls? If not, are there modifications that can be done to achieve this? -Can the method handle images with more than one subject? All object centric experiments (structure + appearance) show generation with a single object, often at the center of the image. How does the approach perform with more complex images ? -In certain settings, conditions appear to be too strong and can affect image quality (e.g. figure 11, dog image). The benefit of guidance/inference is that one can control the influence of a structure image, offering more flexibility. Is there a way to adjust the influence of a structure image in this approach?

Rating

6

Confidence

4

Soundness

3

Presentation

3

Contribution

3

Limitations

One experiment in figure 9 shows one limitation of the proposed approach. However, as pointed out in above sections, there are several additional points that should be discussed (flexibility, single subject, loose controls, etc). For some of these, additional experiments could have allowed to understand the behaviour of the methodology more clearly. Broader impact is adequately discussed.

Reviewer tSTQ7/10 · confidence 4/52024-07-14

Summary

This article presents Ctrl-X, a simple method for T2I diffusion models to control structure and appearance without additional training or guidance. Specifically, it uses feature injection and spatially-aware normalization in the attention laters to align the given structure and appearance. Doing so, Ctrl-X achieves training-free and guidance-free generation in both image and even videos. The effectiveness of Ctrl-X is demonstrated through experimental results on the collected benchmark, underscoring the model's capability.

Strengths

Strengths 1. The technical elaboration of the proposed method is clear. 2. The motivation for the proposed method is straightforward. The insight of the paper is practical. Overall, I appreciate the high-level idea of this paper. 3. The evaluations conducted on the provided benchmarks provide evidence of the effectiveness of the proposed methods. However, there are some concerns regarding the experimental results, which will be further discussed in the weaknesses section.

Weaknesses

Weaknesses 1. Model extension I am curious whether this method can do multiple object generation. For example, given two object sketches/cannies in one picture, and two object appearances in another picture, can this framework automatically match the most suitable appearance and structure alignment for the two objects, thereby generating the picture? 2. Comparisons on latency Since the paper claims that other training-free models significantly increase computing time and require more GPU memory. It is suggested to add another table on comparing computing resources and inference latency.

Questions

Shown as above

Rating

7

Confidence

4

Soundness

4

Presentation

4

Contribution

4

Limitations

The paper has included this part in the conclusion section and checklist.

Authorsrebuttal2024-08-11

We appreciate the reviewers for their thorough reading of our paper, and we look forward to engaging in detailed discussions and addressing the questions you have raised. Preparing our response may take some time, so we hope to begin our communication promptly.

Reviewer 6fgc2024-08-13

Thank you for the response

Thanks for the detailed response. Most of my concerns have been well-addressed. I have raised my rating from 4 to 5.

Program Chairsdecision2024-09-25

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC