Analyzing and Improving Training-Free Fast Sampling of Text-to-Image Diffusion Models

arXiv:2603.00763 · cs.CV · Submitted 2026-02-28 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: We're looking at the paper: "Analyzing and Improving Training-Free Fast Sampling of Text-to-Image Diffusion Models".

Jane: Here's the gist of the paper. Text-to-image diffusion models still struggle to produce high-quality results under limited sampling budgets <ref:2603.00763#pg2> This paper proposes a constant total rotation schedule (TORS), a scheduling strategy that ensures uniform geometric variation along the sampling trajectory, which outperforms previous training-free acceleration methods and produces high-quality images with 10 sampling steps on Flux.1-Dev and Stable Diffusion 3.5 <ref:2603.00763#pg3> Systematic Analysis of Acceleration Components The authors bridge the gap by systematically elucidating the design space of ODE solvers, time schedules, and feature caching methods <ref:2603.00763#pg5> They identify the time schedule as the most pivotal factor affecting model performance <ref:2603.00763#pg6> The paper quantifies the impact of each component on sampling acceleration and pinpoints the outer time schedule as the dominant factor <ref:2603.00763#pg6> Geometric Foundation for Scheduling The proposed TORS strategy is inspired by the geometric properties of diffusion models revealed through the Frenet-Serret formulas <ref:2603.00763#pg10> The authors define curvature κ and torsion τ, which quantify the geometric properties of a curve through two scalar functions They define the total rotation of a segment as Theta

s 1, s 2: = integral s from s 1 to s 2 sqrt(kappa 2(s) + tau 2(s)) ds TORS schedule leverages these geometric insights to allocate more compute steps to regions where the curvature and torsion are high Performance Gains through Optimized Scheduling Extensive experiments demonstrate that TORS consistently yields significant performance gains while exhibiting architecture-agnostic adaptability, hyperparameter robustness, and seamless task transferability <ref:2603.00763#pg6> The results show that TORS achieves nearly 5× acceleration on the Flux model, closely approaching the performance of the 50-step baseline Furthermore, TORS with 10 steps significantly outperforms the 10-step baseline and even surpasses the 50-step baseline in overall score on GenEval Compatibility and Robustness Evaluation The compatibility evaluation across various training-free acceleration methods demonstrates that our proposed TORS exhibits strong robustness, consistently achieving enhanced performance when combined with diverse acceleration strategies The outer schedule predominantly drives the performance improvements among all acceleration strategies Neither the cache object nor the cache cycle significantly affects performance in most cases The method exhibits strong generalization ability to unseen models, maintaining high image quality under constrained sampling budgets The work concludes that the outer time schedule impacts the performance most, and the default uniform schedule is suboptimal leading to slow convergence The TORS strategy provides a principled indicator for adaptive discretization by ensuring that regions of high bending or twisting receive proportionally denser sampling steps The pursuit of ultimate training-free sampling acceleration through incorporation and refinement of multiple components is an important direction for future exploration References 1.

Tom: BlackForestLabs: Flux.1-dev <ref:2603.00763#pg2> 2.

Jane: Chen, D., Zhou, Z., Mei, J.P., Shen, C., Chen, C., Wang, C.: A geometric perspective on diffusion models <ref:2603.00763#pg3> 3.

Tom: Chen, D., Zhou, Z., Wang, C., Lyu, S.: Geometric regularity in deterministic sampling dynamics of diffusion-based generative models 4.

Title and authors: Tom: We’re moving on to the next section of the paper, which is actually just laying out their systematic analysis of these acceleration components. They're going deep into how they designed this entire investigation.

Jane: So, what does that look like for us? Are they just listing off solvers and caching methods?

Lu: No, they’re bridging the gap by systematically elucidating the design space for ODE solvers, time schedules, and feature caching methods. They want to show how these pieces interact.

Meng: I’m curious about what the actual findings are here. Are they just saying "this solver is better" or something more concrete?

Lalam: They quantify the impact of each component on sampling acceleration, and they pinpoint the outer time schedule as the dominant factor affecting model performance across all five components they tested.

Tom: So, it’s not just one thing winning; it’s this outer schedule that dictates how well everything else performs. That's a strong claim.

Jane: It sounds like a lot of work to quantify the impact of every single element, but they are trying to show why the time schedule is so much more important than the rest.

Lu: They do quantify it, showing that by focusing on the outer time schedule, they can isolate its effect and prove it’s what drives most of the performance improvement in their experiments.

Meng: If we were deploying this, I’d want to know if this quantification is based on real traffic or just controlled experiments. That's where I get skeptical about benchmark data.

Tom: The paper says they did comprehensive experiments on five key components of training-free acceleration methods, but it doesn't mention the specific dataset for that analysis.

Jane: That’s a point we should watch out for, because if the results aren't generalizable to real use cases, then the performance gains might not hold up.

Lu: The methodology is designed to be comprehensive, but as you saw in the summary of this paper, their goal was to bridge that gap by looking at what each component contributes.

Lalam: Based on what I've processed from this paper, I'm pretty sure they are confident in pointing out the time schedule as the most pivotal factor for performance.

The paper's summary: Tom: So, let’s go over that summary again of what they actually achieved with this analysis. What’s the main message they want us to get from all this deep diving?

Jane: They are saying that by looking at all these components together, we can finally understand how training-free methods work and where the bottlenecks are.

Lu: The main takeaway is that the default uniform time schedule isn't good enough because it causes slow convergence in image structure, so they introduce TORS to ensure uniform geometric variation along the sampling trajectory.

Tom: So we’re back to TORS being the solution, but now we know *why* it works geometrically. It's not just a lucky guess.

Lalam: They are showing that this strategy doesn't just speed things up; it ensures that regions of high bending or twisting get proportionally denser sampling steps, which is a principled indicator for adaptive discretization.

Jane: That gives us a reason to use TORS instead of just sticking to a fixed number of steps, because it’s based on the actual geometry of the diffusion process itself.

Meng: If we can move away from that slow uniform schedule, does that mean we can reduce our overall sampling budget significantly?

Tom: The paper shows they achieve nearly five times acceleration on the Flux model, and they show this with just ten steps <ref:2603.00763#pg2>. That’s a massive jump from the previous baseline of fifty steps <ref:2603.00763#pg2>.

Jane: So, even though it’s only ten steps, it still gives us quality comparable to a much longer process, which is what we need for practical applications <ref:2603.00763#pg2>.

Lu: They demonstrate this improvement across Flux and Stable Diffusion three point five with architecture agnostic adaptability and hyperparameter robustness <ref:2603.00763#pg1>.

The paper's improvements: Tom: Let’s talk about the specific improvements they suggest in the methodology itself—what is TORS doing differently compared to what came before?

Jane: The main improvement is this new time scheduling strategy, TORS, which leverages curvature and torsion from the Frenet-Serret formulas to guide the sampling process.

Lu: It’s not just a schedule; it’s using those geometric properties—curvature and torsion—to determine where to put more compute steps based on those properties.

Tom: So instead of a fixed schedule, we are adapting our sampling in real time based on the geometry of the path we're taking through the diffusion process. That sounds sophisticated.

Meng: From an engineering perspective, how does this adaptation actually translate into reduced computational load? Does it just mean faster steps or something more efficient?

Lu: It means that by focusing compute where it’s needed most—where curvature and torsion are high, we can achieve the same level of quality with fewer total steps.

Tom: So the improvement is in being smarter about how we spend our sampling budget, not just finding a faster way to run the math itself.

Jane: And they showed that this method performs well even when we test it under different guidance scales, which adds another layer of reliability for us.

Conclusion: Tom: So we’re wrapping up this paper on Analyzing and Improving Training-Free Fast Sampling of Text-to-Image Diffusion Models. What’s the final word?

Jane: They conclude that training-free sampling acceleration methods have improved efficiency, but they largely developed in isolation, so the main thing they't pointing out is we need to move away from the common uniform schedule.

Lu: They emphasize that TORS provides a principled indicator for adaptive discretization by ensuring that regions of high bending or twisting receive proportionally denser sampling steps.

Meng: From a production standpoint, it sounds like this gives us a clear path forward: stop relying on the default settings and start looking at the geometry of the trajectory when trying to speed up generation.

Lalam: I'm confident they’re showing that TORS is a solid strategy because it achieves nearly five times acceleration on Flux with only ten steps <ref:2603.00763#pg2>.

Tom: So, we’re leaving this paper with the idea that for future work, incorporating and refining multiple components remains an important direction for exploring how to achieve truly efficient sampling.

Jane: It’s a lot of work, but it shows that looking at the geometry of the diffusion process can unlock significant performance gains when you have limited sampling budgets.

State Key Laboratory of Blockchain and Data Security, Zhejiang University · Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security · University of California, Berkeley 4 State University of New York at Buffalo

cs.CV

Submitted: 2026-02-28

Updated: 2026-11-11

Comments: Accepted by ECCV 2026

DOI: 10.1007/978-3-032-37044-0_20

Code: https://github.com/zju-pi/TORS

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

Importance score: 91/100

The gist: Text-to-image diffusion models still struggle to produce high-quality results under limited sampling budgets <ref:2603.00763#pg2> This paper proposes a constant total rotation schedule (TORS)…

Key concepts

Constant Total Rotation Schedule (TORS)
TORS is a scheduling strategy that enforces a constant total rotation schedule during the sampling process. This ensures uniform geometric variation along the trajectory, which leads to faster structural convergence and better image quality than standard uniform schedules, especially when using few sampling steps.
Curvature and Torsion
These are geometric properties of a curve used to define the TORS strategy. Curvature measures how sharply a path bends, while torsion measures how much that bend twists in space. The paper uses these metrics to decide where to allocate more sampling steps for better detail.
Architecture-Agnostic Adaptability
The proposed TORS method is designed to work across different diffusion model architectures and training methods. It shows strong generalization ability, meaning the schedule remains effective even when applied to unseen models or different fine-tuned variants, proving its robustness.
Outer Time Schedule Dominance
Systematic analysis reveals that the outer time schedule is the most pivotal factor influencing performance gains in acceleration. While other components like feature caching matter, optimizing this outer schedule provides the most significant improvements across various acceleration techniques.

Terminology

Summary

Text-to-image diffusion models still struggle to produce high-quality results under limited sampling budgets <ref:2603. p. 0

Where the paper says it

Analyzing and Improving Training-Free Fast Sampling of Text-to-Image Diffusion Models Zhenyu Zhou1,2 , Defang Chen3,4† , Siwei Lyu4 , Chun Chen1,2 , and Can Wang1,2 1State Key Laboratory of Blockchain and Data Security, Zhejiang University 2Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security 3University of California, Berkeley 4State University of New York at Buffalo Abstract. Text-to-image diffusion models have achieved unprecedented success but still struggle to produce high-quality results under limited sampling budgets.

00763#pg2> This paper proposes a constant total rotation schedule (TORS), a scheduling strategy that ensures uniform geometric variation along the sampling trajectory, which outperforms previous training-free acceleration methods and produces high-quality images with 10 sampling steps on Flux. p. 0
Where the paper says it

Inspired by the geometric properties of diffusion models revealed through the Frenet-Serret formulas, we propose constant total rotation schedule (TORS), a scheduling strategy that ensures uniform geometric variation along the sampling trajectory. TORS outperforms previous training-free acceleration methods and produces highquality images with 10 sampling steps on Flux.

1-Dev and Stable Diffusion 3. p. 0
Where the paper says it

TORS outperforms previous training-free acceleration methods and produces highquality images with 10 sampling steps on Flux. 1-Dev and Stable Diffusion 3.

p. 0
Where the paper says it

TORS outperforms previous training-free acceleration methods and produces highquality images with 10 sampling steps on Flux. 1-Dev and Stable Diffusion 3.

p. 0
Where the paper says it

TORS outperforms previous training-free acceleration methods and produces highquality images with 10 sampling steps on Flux. 1-Dev and Stable Diffusion 3.

5 <ref:2603. p. 0
Where the paper says it

To improve deployment efficiency without additional training, recent work has shifted toward training-free approaches, such as fast ODE solvers [13, 24, 41, 55–57], optimized time schedules [4, 29, 35, 52], and feature caching [17, 21, 27, 48]. † Corresponding author arXiv:2603.

00763#pg3>

Systematic Analysis of Acceleration Components

The authors bridge the gap by systematically elucidating the design space of ODE solvers, time schedules, and feature caching methods <ref:2603.00763#pg5> They identify the time schedule as the most pivotal factor affecting model performance <ref:2603. p. 1

Where the paper says it

In this paper, we bridge this gap by elucidating the design space of ODE solvers, time schedules, and feature caching methods. Through comprehensive experiments on five key components of training-free acceleration methods, we identify the time schedule as the most pivotal factor affecting model performance, while the default uniform time schedule is suboptimal, leading to slow convergence in image structure.

00763#pg6> The paper quantifies the impact of each component on sampling acceleration and pinpoints the outer time schedule as the dominant factor <ref:2603. p. 1
Where the paper says it

Finally, we conduct a comprehensive compatibility evaluation across various training-free acceleration methods, demonstrating performance improvements and strong robustness of our method. In summary, our contributions are as follows: – We elucidate the design space of mainstream training-free acceleration methods, quantify the impact of each component on sampling acceleration, and pinpoint the outer time schedule as the dominant factor.

00763#pg6>

Geometric Foundation for Scheduling

The proposed TORS strategy is inspired by the geometric properties of diffusion models revealed through the Frenet-Serret formulas <ref:2603.00763#pg10> The authors define curvature κ and torsion τ, which quantify the geometric properties of a curve through two scalar functions They define the total rotation of a segment as Theta[s 1, s 2] = integral s from s 1 to s 2 sqrt(kappa 2(s) + tau 2(s)) ds TORS schedule leverages these geometric insights to allocate more compute steps to regions where the curvature and torsion are high

Performance Gains through Optimized Scheduling

Extensive experiments demonstrate that TORS consistently yields significant performance gains while exhibiting architecture-agnostic adaptability, hyperparameter robustness, and seamless task transferability <ref:2603.00763#pg6> The results show that TORS achieves nearly 5× acceleration on the Flux model, closely approaching the performance of the 50-step baseline Furthermore, TORS with 10 steps significantly outperforms the 10-step baseline and even surpasses the 50-step baseline in overall score on GenEval

Compatibility and Robustness Evaluation

The compatibility evaluation across various training-free acceleration methods demonstrates that our proposed TORS exhibits strong robustness, consistently achieving enhanced performance when combined with diverse acceleration strategies The outer schedule predominantly drives the performance improvements among all acceleration strategies Neither the cache object nor the cache cycle significantly affects performance in most cases The method exhibits strong generalization ability to unseen models, maintaining high image quality under constrained sampling budgets

Conclusion on Design Space

The work concludes that the outer time schedule impacts the performance most, and the default uniform schedule is suboptimal leading to slow convergence The TORS strategy provides a principled indicator for adaptive discretization by ensuring that regions of high bending or twisting receive proportionally denser sampling steps The pursuit of ultimate training-free sampling acceleration through incorporation and refinement of multiple components is an important direction for future exploration

References

  1. BlackForestLabs: Flux. p. 0

    Where the paper says it

    Inspired by the geometric properties of diffusion models revealed through the Frenet-Serret formulas, we propose constant total rotation schedule (TORS), a scheduling strategy that ensures uniform geometric variation along the sampling trajectory. TORS outperforms previous training-free acceleration methods and produces highquality images with 10 sampling steps on Flux.

    1-dev <ref:2603. p. 0
    Where the paper says it

    To improve deployment efficiency without additional training, recent work has shifted toward training-free approaches, such as fast ODE solvers [13, 24, 41, 55–57], optimized time schedules [4, 29, 35, 52], and feature caching [17, 21, 27, 48]. † Corresponding author arXiv:2603.

    00763#pg2>

  2. Chen, D., Zhou, Z., Mei, J.P., Shen, C., Chen, C., Wang, C.: A geometric perspective on diffusion models <ref:2603. p. 0

    Where the paper says it

    In this paper, we bridge this gap by systematically elucidating the design space, and our comprehensive experiments identify the sampling time schedule as the most pivotal factor. Inspired by the geometric properties of diffusion models revealed through the Frenet-Serret formulas, we propose constant total rotation schedule (TORS), a scheduling strategy that ensures uniform geometric variation along the sampling trajectory.

    00763#pg3>

  3. Chen, D., Zhou, Z., Wang, C., Lyu, S.: Geometric regularity in deterministic sampling dynamics of diffusion-based generative models

  4. Chen, D., Zhou, Z., Wang, C., Shen, C., Lyu, S.: On the trajectory regularity of ODE-based diffusion sampling

  5. Chen, P., Shen, M., Ye, P., Cao, J., Tu, C., Bouganis, C.S., Zhao, Y., Chen, T.: Delta-dit: A training-free acceleration method tailored for diffusion transformers

  6. Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F.: Scaling rectified flow transformers for high-resolution image synthesis

  7. Farouki, R.T.: Pythagorean-hodograph curves: algebra and geometry inseparable

  8. Ghosh, D., Hajishirzi, H., Schmidt, L.: Geneval: An object-focused framework for evaluating text-to-image alignment

  9. Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., Hochreiter, S.: GANs trained by a two time-scale update rule converge to a local Nash equilibrium

  10. Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models

  11. Hyvärinen, A.: Estimation of non-normalized statistical models by score matching

  12. Ju, X., Zeng, A., Bian, Y., Liu, S., Xu, Q.: Pnp inversion: Boosting diffusion-based editing with 3 lines of code

  13. Karras, T., Aittala, M., Aila, T., Laine, S.: Elucidating the design space of diffusionbased generative models

  14. Labs, B.F., Batifol, S., Blattmann, A., Boesel, F., Consul, S., Diagne, C., Dockhorn, T., English, J., English, Z.: Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space

  15. Lewiner, T., Gomes, J.D., Lopes, H., Craizer, M.: Curvature and torsion estimators based on parametric curve fitting

  16. Li, L., Li, H., Zheng, X., Wu, J., Xiao, X., Wang, R., Zheng, M., Pan, X., Chao, F., Ji, R.: Autodiffusion: Training-free optimization of time steps and architectures for automated diffusion model acceleration

  17. Li, S., Hu, T., Khan, F.S., Li, L., Yang, S., Wang, Y., Cheng, M.M., Yang, J.: Faster diffusion: Rethinking the role of unet encoder in diffusion models

  18. Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., Zitnick, C.L.: Microsoft coco: Common objects in context

  19. Lipman, Y., Chen, R.T., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling

  20. Liu, B., Akhgari, E., Visheratin, A., Kamko, A., Xu, L., Shrirao, S., Lambert, C., Souza, J., Doshi, S.: Playground v3: Improving text-to-image alignment with deep-fusion large language models

  21. Liu, J., Zou, C., Lyu, Y., Chen, J., Zhang, L.: From reusing to forecasting: Accelerating diffusion models with taylorseers

  22. Liu, L., Ren, Y., Lin, Z., Zhao, Z.: Pseudo numerical methods for diffusion models on manifolds

  23. Liu, X., Gong, C., Liu, Q.: Flow straight and fast: Learning to generate and transfer data with rectified flow

  24. Lu, C., Zhou, Y., Bao, F., Chen, J., Li, C., Zhu, J.: Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

  25. Lu, C., Zhou, Y., Bao, F., Chen, J., Li, C., Zhu, J.: Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models

  26. Ma, X., Fang, G., Bi Mi, M., Wang, X.: Learning-to-cache: Accelerating diffusion transformer via layer caching

  27. Ma, X., Fang, G., Wang, X.: Deepcache: Accelerating diffusion models for free

  28. Meng, C., Rombach, R., Gao, R., Kingma, D., Ermon, S., Ho, J., Salimans, T.: On distillation of guided diffusion models

  29. Nichol, A.Q., Dhariwal, P.: Improved denoising diffusion probabilistic models

  30. Oksendal, B.: Stochastic differential equations: an introduction with applications

  31. Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., Müller, J., Penna, J., Rombach, R.: Sdxl: Improving latent diffusion models for high-resolution image synthesis

  32. Pressley, A.N.: Elementary differential geometry

  33. Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J.: Learning transferable visual models from natural language supervision

  34. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models

  35. Sabour, A., Fidler, S., Kreis, K.: Align your steps: Optimizing sampling schedules in diffusion models

  36. Song, J., Meng, C., Ermon, S.: Denoising diffusion implicit models

  37. Song, Y., Dhariwal, P., Chen, M., Sutskever, I.: Consistency models

  38. Song, Y., Ermon, S.: Generative modeling by estimating gradients of the data distribution

  39. Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., Poole, B.: Scorebased generative modeling through stochastic differential equations

  40. Xu, J., Liu, X., Wu, Y., Tong, Y., Li, Q., Ding, M., Tang, J.: Imagereward: learning and evaluating human preferences for text-to-image generation

  41. Xue, S., Liu, Z., Chen, F., Zhang, S., Hu, T., Xie, E., Li, Z.: Accelerating diffusion sampling with optimized time steps

  42. Ye, Z., Chen, Z., Li, T., Huang, Z., Luo, W.: Schedule on the fly: Diffusion time prediction for faster and better image generation

  43. Yin, T., Gharbi, M., Zhang, R., Shechtman, E., Durand, F., Freeman, W.T., Park, T.: One-step diffusion with distribution matching distillation

  44. Zhang, Q., Chen, Y.: Fast sampling of diffusion models with exponential integrator

  45. Zhao, W., Bai, L., Rao, Y., Zhou, J., Lu, J.: Unipc: A unified predictor-corrector framework for fast sampling of diffusion models

  46. Zhou, Z., Chen, D., Wang, C., Chen, C.: Fast ode-based sampling for diffusion models in around 5 steps

  47. Zhou, Z., Chen, D., Wang, C., Chen, C., Lyu, S.: Simple and fast distillation of diffusion models

Appendix B Experimental Details

The authors collected 100 sampling trajectories using text prompts randomly sampled from the MS-COCO dataset The total latency for trajectory collection was 40 minutes for Flux.1-Dev and 15 minutes for Stable Diffusion 3. p. 0

Where the paper says it

TORS outperforms previous training-free acceleration methods and produces highquality images with 10 sampling steps on Flux. 1-Dev and Stable Diffusion 3.

5 medium

Figure Analysis

Figure 1 shows that TORS attains image quality comparable to the 50-step baseline using only 10 steps Figure 4 illustrates that as the number of sampling steps increases, the image structure continues to change and only stabilizes after around 30 steps with a uniform outer schedule Figure 5 shows that TORS yields considerably faster structural convergence compared with the uniform outer schedule Figure 17c shows that TORS significantly improves performance and achieves faster structural convergence on Stable Diffusion 3.5 medium

Qualitative Results

Figure 9 provides qualitative results of image editing using Flux.1-Kontext, demonstrating that TORS preserves superior layout consistency relative to 50-step reference generations Figure 18 and Figure 19 show that our proposed TORS demonstrates significant improvements when operating under constrained sampling budgets The results for both Flux. p. 8

Where the paper says it

(b) A 10-step example of our proposed TORS, which ensures a constant total rotation change R S 0 |ω(s)|ds along the sampling trajectory. (c) TORS yields considerably faster structural convergence compared with the uniform outer schedule (Figure 4).

1-Dev and Stable Diffusion 3.5 are shown in Figure 18 and Figure 19

Table Analysis

Table 7 shows that Velocity cache requires 38.88 GB, 41.44 GB and 44.00 GB of VRAM for batch sizes of 2, 3 and 4 respectively on Flux. p. 6

Where the paper says it

48 GB to 42. 78 GB of VRAM, even surpassing the memory needed for a doubled batch size with velocity cache (38.

1-Dev Table 2 demonstrates that TORS with 10 steps significantly outperforms the 50-step baseline in overall score on GenEval Table 3 shows that TORS achieves state-of-the-art results among sampling acceleration methods on Flux. p. 1
Where the paper says it

Our proposed TORS attains image quality comparable to the 50-step baseline using only 10 steps. However, existing training-free methods have largely been developed in isolation, and no systematic study, particularly for state-of-the-art text-to-image diffusion models, has examined their design components or compared their relative contributions to overall performance.

1-Dev Table 4 demonstrates that our method maintains its efficacy regardless of the guidance scale employed on TORS CFG Table 5 provides comprehensive ablations on the prompt distributions, sample sizes and step counts used for schedule generation

Final Summary

The paper concludes that training-free sampling acceleration methods have improved the efficiency of textto-image generation but have largely been developed in isolation The authors reveal the defects of the commonly used uniform schedule and introduce TORS, an improved, efficient outer schedule that leverages the geometric properties of diffusion sampling trajectories The results for both Flux.1-Dev and Stable Diffusion 3.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Constant Total Rotation Schedule (TORS) for Sampling Schedules: TORS ensures uniform geometric variation along the sampling trajectory by enforcing a constant total rotation schedule, which leads to faster structural convergence compared to default uniform schedules, especially in few-step generation where image structure otherwise fluctuates before stabilizing.

  2. Geometric Regularity Analysis for Schedule Generation: The method uses PCA on 100 sampling trajectories to identify strong trajectory regularity, allowing the calculation of curvature and torsion across different models and prompts to derive a schedule that allocates more compute steps to regions where the curvature and torsion are high.

  3. Performance Gain in Constrained Budgets: The TORS strategy has been shown to yield nearly 5x acceleration on the Flux model while maintaining image quality comparable to 50-step baselines using only 10 steps, demonstrating superior performance under limited sampling budgets.

  4. Architecture and Model Agnostic Adaptability: The proposed method exhibits architecture-agnostic adaptability and strong generalization ability to unseen models, successfully transferring TORS schedules to different LoRA-fine-tuned variants and distinct architectures like QwenImage.

  5. Robustness Across Components: The compatibility evaluation shows that TORS demonstrates strong robustness, consistently achieving enhanced performance when combined with diverse acceleration strategies, indicating that the outer schedule is the dominant factor for performance gains regardless of other training-free components.

Abstract

Text-to-image diffusion models have achieved unprecedented success but still struggle to produce high-quality results under limited sampling budgets. Existing training-free sampling acceleration methods are typically developed independently, leaving the overall performance and compatibility among these methods unexplored. In this paper, we bridge this gap by systematically elucidating the design space, and our comprehensive experiments identify the sampling time schedule as the most pivotal factor. Inspired by the geometric properties of diffusion models revealed through the Frenet-Serret formulas, we propose constant total rotation schedule (TORS), a scheduling strategy that ensures uniform geometric variation along the sampling trajectory. TORS outperforms previous training-free acceleration methods and produces high-quality images with 10 sampling steps on Flux.1-Dev and Stable Diffusion 3.5. Extensive experiments underscore the adaptability of our method to unseen models, hyperparameters, and downstream applications. Code is available at https://github.com/zju-pi/TORS.

Sources

Related papers