acceptodds
Withdrawn from ICLR 2027

HarnessPainter: Grounding Visual Reasoning in Executable Processes for In-Context Stroke Organization

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

Stroke-based rendering algorithms generate complete painting programs via pixel-level reconstruction objectives, but they frequently produce disordered execution sequences, inverted structural layers, and redundant stroke fragments. We present HarnessPainter, a tool-augmented framework that formulates stroke reorganization as closed-loop, in-context program editing. Using an initial stroke sequence, a vision–language agent inspects target images, dynamic intermediate keyframes, and localized visual evidence to propose discrete structural edits, such as reordering procedural stages and consolidating adjacent contour fragments. A deterministic execution harness verifies these proposals: an integrated differentiable rendering backend locally optimizes continuous curve parameters while preserving global topology and unedited execution orders. Proposals are deterministically replayed and accepted only upon passing local, global, and event-aligned consistency checks. Evaluations show HarnessPainter rectifies fragmented artifacts and enforces coherent long-range drawing dependencies while preserving fidelity, demonstrating how executable visual environments ground high-level reasoning for procedural graphics generation.

Primary area:
applications to computer vision, audio, language, and other modalities
Keywords:
Stroke-based rendering, Execution harnesses, Tool-augmented agents, Differentiable rendering

Reviews

Among:
Shape:
1234this paper, Bottom 1%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.00 over 3 reviews, above 0% of the others.
Reviewer_eZvWOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
2
Read review
Paper summary

The paper argues that stroke-based rendering (SBR) should consider not only reconstruction fidelity but also the organization of the strokes. Treating the output as an ordered stroke sequence (a "painting program"), HarnessPainter proceeds in three stages: (1) semantic ordering, which fits strokes in six stages toward coarse-to-fine targets (2) alignment, which rotates, bends, and translates strokes to follow a flow field derived from edge tangent flow and guide curves specified by a vision-language model (VLM) (3) consolidation, which merges compatible strokes into fewer curves. The VLM makes high-level decisions, local tools perform the geometric edits, and a "harness" accepts an edit only if it passes geometric checks and PSNR bounds.

Critical strengths

Viewing strokes as a program is an interesting perspective. The paper treats an SBR result as an executable stroke program whose organization matters in its own right, and refines it through VLM decisions that are executed and verified by tools. This viewpoint is the main reason I find the submission interesting.

Critical weaknesses
  1. The formulation and presentation make it hard to see what the method is and what is new.

Related work: Section 2 is very condensed. Each prior method is summarized in a short phrase relying on terms specific to the cited paper, and the paper's own contribution is described in similarly specialized terms (e.g., "stable IDs, protected context, replacement lineage, and rendered acceptance gates"). It is therefore hard to grasp the core idea of each prior work, what HarnessPainter builds on, and what is new. Beyond the works cited, stroke order and the alignment of strokes with contours have also been discussed in the non-photorealistic rendering literature.

Formulation: The abstract states that stroke organization is formulated "as an executable program refinement problem", but the program is simply an ordered list of stroke parameters, and no objective for a well-organized program is defined. The desired properties appear only implicitly in the acceptance checks.

Section 3: Each step is easy to follow, but the purpose of the procedures and the choice of numerical values are rarely explained. For example, "Locations with support at least 0.08 provide reliable guidance" relies on a quantity, "support", that is defined only verbally, with an unexplained threshold. The PSNR tolerances in Eqs. (3)–(4) and the many thresholds in Appendices A.1.3–A.1.4 are similar. The section reads more like an implementation with hard-coded constants than a method with motivated design choices.

Reproducibility: The VLM, which makes key decisions at every stage, is not identified, and its prompts and input/output formats are not provided.

  1. Parts of the manuscript appear unfinished.

Several passages read as working notes or plans. Appendix A.1.3 states that the conversion of the width values used in several thresholds to the renderer's half-width "remains to be confirmed". Appendix A.1.6 states that a clamping surrogate "and its threshold must be recorded" without reporting them. Appendix A.2 lists "planned extensions" for which "no measurements or qualitative outcomes are yet available". Some statements refer to things absent from the paper, such as "comparisons with fixed-rule control and one-pass planning", "generated states as targets", a "generator", and the "in-context" organization in the title.

  1. The evaluation does not substantiate the main claims.

Section 4.1 notes that stroke organization, artistic character, and drawing order matter beyond pixel fidelity, and the abstract claims "improved stroke coherence and process-level quality over existing SBR approaches". However, these qualities are neither defined nor measured. The evidence consists of visual comparisons and a single PSNR comparison in the ablation. Moreover, the only baseline is "unguided SBR", which is not described in the paper and appears to have been constructed by the authors. None of the SBR methods discussed in Section 2 is compared against.

Questions for authors
  1. What exactly is "unguided SBR"? Why are none of the existing SBR methods discussed in Section 2 used as baselines?
  2. Where is differentiable rendering used? The abstract mentions "differentiable SBR operators" and Appendix A.1.6 derives a gradient, but the alignment updates appear rule-based, and the loss and optimizer for the stage fitting and the appearance fit are not specified.
  3. How were the thresholds chosen, and how sensitive are the results to them?
  4. Which VLM is used, and could the prompts and input/output formats be provided?
Reviewer_5EjkOpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

This paper proposes HarnessPainter, a system for organizing stroke into images. The paper is very hard to understand and the experiments are very limited. Several sentences in this paper do not seem to be written by human authors.

Critical strengths

None

Critical weaknesses
  1. Some statements do not seem to be written by human authors. The reviewer noted the following passages:
  • Line 729: “Their conversion to the renderer’s half-width w_i remains to be confirmed.”
  • Line 887: “the following are planned extensions of Section 4; no measurements or qualitative outcomes are yet available.”
  1. Insufficiently justified motivation and contribution. The paper does not convincingly establish the practical value of the studied “stroke organization” task. Selected visual examples suggest different stroke arrangements, but do not demonstrate improved usability, artist preference, or execution quality over simpler alternatives.
  2. Unclear and incomplete methodological specification. The connection between VLM decisions and numerical operations is difficult to reconstruct, with important details missing about stroke initialization, fitting objectives, and replacement-curve construction.
  3. Insufficient experimental support. The evaluation relies heavily on qualitative examples and provides limited quantitative results. It also fails to isolate the contribution of VLM reasoning from the geometric tools and heuristic rules, and Line 887 states additional experiments are planned but have not been run yet. Overall, the current evidence does not support the breadth of the paper’s claims.
Questions for authors

None

Reviewer_6GuZOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

The paper proposes HarnessPainter, which achieves a VLM-guided framework to perform stroke-based rendering (SBR) by leveraging semantic ordering, alignment, and consolidation. The paper looks incomplete based on the information in the main paper and supplementary material.

Critical strengths
  1. The paper starts with a very novel idea of organizing strokes through visual reasoning.
  2. The methodology is well motivated by the idea of separating VLM planning from numerical stroke manipulation.
Critical weaknesses
  1. The paper looks unfinished. I’m a bit concerned about the additional experiments in App. A.2. The authors mention that “no measurements or qualitative outcomes are yet available” (L887-888), and the paragraphs that follow only describe planned experiments with no results at all. There are also other places where work is left open, for example, “remain to be measured” (L427-428) and “remains to be confirmed” (L729-730).
  2. I couldn’t find any comparison with existing SBR methods, even though the abstract claims it (L29-30). The only baseline is “unguided SBR”, and the authors don’t provide any clarification about it.
  3. The results are not convincing and are limited. The only number shown in paper is one PSNR value on a single image (L397–399), and only four test images are used.
  4. VLM is missing from the paper. The authors don't provide any information about VLM at all (i.e, model name, prompt used).
Questions for authors
  1. See weakness 1-4.
  2. Please clarify all weaknesses, especially 1 and 2.

p.s. I like the idea, but the current results just aren’t convincing enough, either in numbers or visually. I’d suggest polishing the paper further, showing more varied visual examples and adding more baselines before resubmitting.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.