GenArtist: Multimodal LLM as an Agent for Unified Image Generation and Editing

Despite the success achieved by existing image generation and editing methods, current models still struggle with complex problems including intricate text prompts, and the absence of verification and self-correction mechanisms makes the generated images unreliable. Meanwhile, a single model tends to specialize in particular tasks and possess the corresponding capabilities, making it inadequate for fulfilling all user requirements. We propose GenArtist, a unified image generation and editing system, coordinated by a multimodal large language model (MLLM) agent. We integrate a comprehensive range of existing models into the tool library and utilize the agent for tool selection and execution. For a complex problem, the MLLM agent decomposes it into simpler sub-problems and constructs a tree structure to systematically plan the procedure of generation, editing, and self-correction with step-by-step verification. By automatically generating missing position-related inputs and incorporating position information, the appropriate tool can be effectively employed to address each sub-problem. Experiments demonstrate that GenArtist can perform various generation and editing tasks, achieving state-of-the-art performance and surpassing existing models such as SDXL and DALL-E 3, as can be seen in Fig. 1. Project page is https://zhenyuw16.github.io/GenArtist_page.

Paper

Similar papers

Peer review

Reviewer tVAw6/10 · confidence 4/52024-07-11

Summary

This paper presents a method by which an MLLM (gpt 4) is used as a task planner for image editing tasks. A catalog of off the shelf models is used as a set of tools and the planner constructs a task tree, executes proposed tasks, performs verification, and optionally backtracks. The authors claim the following contributions: * a unified image generation and editing system * a planner which can construct a task tree and perform verification * the ability to perform tool selection

Strengths

I really like the concept. It matches intuition that given a strong enough controller / planner, you might be able to perform editing more easily with a bag of tools. Quantitative evaluations show substantial improvements. Good writing.

Weaknesses

Probably should cite Gupta, et. al. "Visual Programming: Compositional Visual Reasoning Without Training", CVPR 2023. No comparison with other MLLMs. Contributions could be further refined and clarified.

Questions

I would recommend picking one or two more MLLMs and doing a comparison. I think it's fine that this paper serves as an existence proof that there exists an MLLM with planning capabilities suitable for multi-step image editing, but I didn't get a very good sense of how that performance is impacted by product-specific fine tuning and alignment that might happen behind the scenes with gpt 4.

Rating

6

Confidence

4

Soundness

3

Presentation

3

Contribution

2

Limitations

seems fine

Reviewer TXUv5/10 · confidence 4/52024-07-12

Summary

This paper introduces a new multi-modal agent for image generation and editing that break down these tasks into subproblems to solve with external tools, including self correction module with verification feedback.

Strengths

- While the idea of augmenting a large (multimodal) language model with tools and turning it into an agent for complex tasks is not new, this work presents a novel multi-modal agent for image generation and editing that combines different existing ideas such as tool use, tree planning, and verification, etc. - The authors demonstrate strong positive results on two benchmarks for image generation and editing respectively compared to baselines, which wouldn’t have been possible without great execution. - The authors conducted ablation studies and present evidence that shows the importance of the different components e.g. planning tree vs. chain in the system. - The paper is well written with informative and clear illustrations.

Weaknesses

- While the experiments authors conducted are well-motivated and support their claims, the paper would be stronger if it performed a deeper analysis, for example, on common error cases, or via more finegrained ablation studies on the tools / positive-aware tool execution. - It would also be stronger if the authors evaluated the system on additional benchmarks such as GenAI-Bench: A Holistic Benchmark for Compositional Text-to-Visual Generation, although the reviewer understands that it might be impossible due to short rebuttal period. - the paper lacks some technical details about the method such as the underlying model.

Questions

- Did authors use language-only GPT4 or GPT4-V or GPT4o? For reproducibility, which version of GPT4 was used? - One of the common failures of such multi-modal agent systems is due to errors in tool outputs. I wonder if the same issue exists in this system, and if so how the authors address such issues e.g. when the bounding boxes output by the localization/detection module are wrong? Nits: table3: why is smart Edit missing L2 number?

Rating

5

Confidence

4

Soundness

3

Presentation

3

Contribution

2

Limitations

- No, the authors have not addressed the limitations of the work.

Area Chair HJzT2024-08-12

Concerns addressed?

Dear reviewer, thank you for providing constructive feedback to the authors! Are your questions satisfactorily addressed by the rebuttal? Would you like to revise your rating or do you need any more information from the authors to make that decision?

Reviewer TXUv2024-08-14

Thanks for the response and additional evaluations! My questions have been addressed and I have no other concerns. I will keep my rating as it is. Please update the revision with additional technical details e.g. GPT4 version for reproducibility.

Reviewer tJSg6/10 · confidence 5/52024-07-13

Summary

This paper presents GenArtist, a system that utilizes MLLMs as agents for image generation and editing, especially for complex language descriptions. The key idea is to first use MLLM to decompose the generation task as an execution tree of various tools such as SDXL and LMD, and then utilize all the tools in a predefined tool library. Extensive experiments on T2I-CompBench and MagicBrush shows promising results.

Strengths

- The problem of unified image generation and editing over arbitrary language instructions is very relevant for both academia and industry. How to leverage the existing powerful LLMs such as GPT-4 and other tools is also very important for the deployment of current academic progress. - The proposed method is quite simple and is reasonable. The paper is clearly written and the results look good to me.

Weaknesses

Overall, I think this paper has made some good contributions and shown some promising results. However, I hold some concerns as follows, **Major concern.** My major concern is about the possible limitation in the decomposition tree. All images generated for the complex text is by first generating an initial image, and then utilizing editing tools to match the text description. This could be very limited if the images initially generated are not very decent. This is because even if the final edited image matches the language description better in semantics, the overall quality and layout may not be very natural in some cases. The editing trace may be easily detected and this may be attributed the weakness of the editing models. The editing operation in the paper is mostly editing local regions while keeping the original overall structure, and this could be the problem. For example, the generated result with "hot dogs" in Figure 1 is not natural or of decent quality to me when compared to results generated by models like Playground. The image is blurred and the hot dogs look a little fake. **Human-aligned evaluation.** The current evaluation is mainly conducted on T2I-CompBench and MagicBrush. However, it has been well known that traditional metrics such as DINO scores are not aligned with humans. This could cause a result that even if the image quality is not as good as another single-expert result, the score may still be better. However, this might be misaligned with human preference in some cases. Can authors conduct more evaluations on more advanced benchmarks like DreamBench++ [1]? Minor suggestion: I think the paper organization could be improved by trying to put each full paragraph on one page or two. Try your best not to split one section or subsection into two pages. For example, Section 4.1 and 4.3 are all put under the bottom of one page with very few contents, this could cause a sense of reading fragmentation. [1] DreamBench++: A Human-Aligned Benchmark for Personalized Image Generation.

Questions

As stated in weakness, I am curious about the results of modern human-aligned benchmarks, such as DreamBench++. I would be happy to see more human-aligned evaluation, if possible. I am looking forward to the authors' response.

Rating

6

Confidence

5

Soundness

2

Presentation

4

Contribution

3

Limitations

Yes, the authors have discussed the limitations. However, I would suggest the authors discuss the potential limitation brought by the initial generated results and editing artifacts.

Reviewer tJSg2024-08-11

Thanks, rating kept

Thank you for your rebuttal efforts and response. The hot dog case does look better but it is still a little bit fake. However, this issue can be alleviated by using more advanced image editions and generative tools. I appreciate you for letting me know about the aesthetic quality correction procedure, which is helpful. I strongly suggest that the authors discuss this issue properly in the revised paper and incorporate all rebuttal experiments into the final paper. Besides, I suggest the authors use the latest GPT-4o instead of GPT-4V and conduct all experiments again since GPT-4o is a more advanced GPT tool to date, and I think it would be valuable to keep using state-of-the-art tools. All details, including the precise GPT-4o version (eg, `turbo-2024-04-09`), should be explicitly recorded in the paper. Overall, I think my concern is resolved, and I will keep my rating for a supportive assessment for this paper.

Reviewer tVAw2024-08-12

Rebuttal response

Rebuttal looks good, thank you for the MLLM comparison. I'm bumping my rating up to weak accept.

Authorsrebuttal2024-08-14

Thanks

Dear Reviewer We are happy that our reply address your concerns on MLLM comparison. We are sorry to bother you, but It seems that the score has not been changed yet. It will be nice to change the score when you are in convenience. Thanks

Authorsrebuttal2024-08-14

Looking forward to Feedback as Discussion Deadline Approaches

Thanks for your thorough reviews, which are very helpful to improving the quality of our paper. We apologize for any inconvenience caused, but as the deadline for discussion (Aug 13 11:59 pm AoE) draws near, we would like to provide an update on our progress. If you need further clarification or have additional questions, please don't hesitate to contact us. Again, we sincerely thank you for your time and effort in reviewing our paper. Thanks

Program Chairsdecision2024-09-25

Decision

Accept (spotlight)

© 2026 NYSGPT2525 LLC