ManimCode: Learning Visual Self-Correction for Educational Animation

Zhengyang Liang; Lizi Liao

cs.CVNov 5, 2026

ManimCode: Learning Visual Self-Correction for Educational Animation

Authors: Zhengyang Liang, Lizi Liao

Organizations: Singapore Management University

Abstract

Manim supports precise, programmatic video generation, but a successfully rendered educational animation can still have visual defects, missing explanations, or incorrect content. Rendering does not directly identify the code changes needed to correct these visual failures, so feedback-driven revision is needed. Existing agentic frameworks use iterative rendering and revision but typically rely on large models, often proprietary. Existing Manim datasets primarily contain final programs and often lack quality verification and the trajectories of feedback and revisions needed to learn this refinement process. We introduce ManimData and ManimCode, a data and training framework for feedback-driven educational animation generation. We construct ManimData through a render-guided pipeline that iteratively generates, executes, reviews, and revises Manim programs with retrieved API documentation. Its 24,034 render-verified, quality-filtered records preserve intermediate programs, execution errors, visual critiques, and corrective revisions. We use role-conditioned supervised fine-tuning on these trajectories to train ManimCode, a single 4B model, to write, review, and revise animation programs. To improve correction of errors that differ from those in the collected trajectories, we then train the Reviser with reinforcement learning on self-generated states, using rendering outcomes and reference-free quality feedback. ManimCode achieves 96.6% and 97.0% render success on ManimData-test and ManimBench, compared with 52.6% and 82.0% for the untrained 4B model under the same inference setup, while also improving video quality scores from 2.48 to 2.94 and from 3.00 to 3.44. Integrating ManimCode into two long-form educational video generation systems increases completed videos from 19/117 to 107/117 and from 53/239 to 239/239, alongside improvements in educational effectiveness and visual quality.

Figures & tables

Dataset	Size	Purpose	Target Video	Render	Quality	Revision
Dataset	Size	Purpose	Target Video	Verified	Filtered	Trajectories
ManimBench ( Rammuni Silva et al., 2026a )	416	Train/Eval	Scene-level Animation	✓	✗	✗
JanusCode-800K (Manim) ( Sun et al., 2026 )	8,316	Train	Scene-level Animation	✗	✗	✗
HF Community (7 datasets)	20,976	Train	Scene-level Animation	✗	✗	✗
MMMC (Code2Video) ( Chen et al., 2025b )	117	Eval	Long-form Explanation	n/a	n/a	✗
TheoremExplainBench ( Ku et al., 2025 )	239	Eval	Long-form Explanation	n/a	n/a	✗

Table 1: Comparison with existing Manim datasets and benchmarks. n/a marks a property that does not apply to the evaluation-only long-form benchmarks.

Figure 1: ManimData construction pipeline. (1) Request collection: Wikipedia STEM requests, community programs, and the API knowledge base. (2) Render-guided generation: Qwen3.8-27B writes or revises the program from retrieved API cards, and each version is rendered, checked for layout defects, and reviewed on eight frames; an error or a critique (a crash and an overlapping label) starts the next revision. (3) Curation: an input whose best version reaches a mean review score of at least 3.75 is admitted with its code, video, and trajectory.

Figure 2: ManimData statistics. Top: domain shares of the 24,034 records. Bottom: per-record distributions of final program length, video length, and number of renders.

Figure 3: Overview of ManimCode. (1) Final programs and intermediate steps of ManimData trajectories supervise five tasks of one Qwen3.5-4B model. (2) Reinforcement learning further trains the Reviser on repair and critique states that the supervised model produces itself: the Reviser samples eight revisions per state, Qwen3.8-27B scores each rendered revision (request coverage for repair, two order-swapped comparisons with the draft for critique), and revisions that score above the rest of their group become more likely (Sec. 4.2 ). The two states shown are real training states; their revisions and scores are illustrative. (3) At inference, our model runs the render–review–revise loop. Qwen3.8-27B is used only to build the data and to compute training rewards.

Large general-purpose models
		ManimData-test		ManimBench
Model	Size	RS $\uparrow$	VQ $\uparrow$	RS $\uparrow$	VS $\uparrow$	VQ $\uparrow$
GPT-4o	–	60.2	2.87	79.0	71.0	3.34
GLM-5.3 Flash	MoE	57.2	3.18	83.0	73.7	3.51
DeepSeek-V4 Flash	MoE	62.4	3.05	81.0	72.9	3.33
Small general-purpose models

Table 2: Scene-level animation generation on ManimData-test and ManimBench. VQ is averaged over successfully rendered videos. † 71 of the 100 ManimBench requests appear in JanusCoder’s training data.

Code2Video on MMMC
Metric (higher is better)	Native	+ Ours
Videos completed	19/117	107/117
TeachQuiz gain ( $\times$ 100)	$-$ 2.6	27.2
Aesthetics (0–100)	35.5	58.7
TheoremExplainAgent on TEB
Videos completed	53/239	239/239

Table 3: Scene-module substitution in long-form systems. Both runs of a system share one lecture plan written by the untrained Qwen3.5-4B. Native writes and repairs the scenes with the same untrained model, and + Ours generates them with our system.

Figure 5: One revision step of our system on ManimData-test. Our Reviewer flags that the title overlaps the table header; the Reviser moves the table down and places the current-state label below it, and the revised video is kept as the final output. Dashed boxes mark the defect (red) and the repaired region (green). Under each frame are our Reviewer’s mean score and Gemini 3.8 Flash score. More examples are in Appendix C.5 .

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Table 9

Setting	Value
Renderer	ManimCE v0.19.1 with eight domain plugins, 480p
Budget	10 renders per input
Writer	Qwen3.8-27B; writes the full program from retrieved cards and feedback
Retrieval	2–4 queries per step; Qwen3-Embedding-8B search, Qwen3-Reranker-8B reranking; top 10 cards
Layout checks	Objects outside the frame, overlapping text, labels too small to read, at animation pause points
Reviewer	Qwen3.8-27B; request, code, and 8 uniformly sampled frames; four dimensions scored 1–5

Appendix

Table 7: Construction loop settings. Qwen3.8-27B plays every role. The Writer row also applies to the Reviser, which rewrites the full program from the error or the critique. Acceptance decides when a candidate replaces the best candidate

c^{*}

, and Admission decides whether a record enters ManimData.

Supervision	Input context	Target output	Examples
Writing	Request; retrieved API cards	Final program selected during construction	24,034
Execution repair	Request; failed code; execution error; API cards	Renderable repaired program	14,587
Quality revision	Request; code; scores and critique; layout diagnostics; API cards	Higher-rated revised program	32,683
Reviewing	Request; code; sampled video frames	Scores on the four review dimensions and a critique	46,965
Retrieval queries	Request; error or critique when present	Two to four API search queries	187,183
Total			305,452

Appendix

Table 8: Role-conditioned supervision. Input context, target output, and number of examples for each supervision type extracted from the 24,034 ManimData trajectories.

Supervised fine-tuning		Revision RL
Base model	Qwen3.5-4B	Initialization	supervised model
LoRA	rank 32, $\alpha=64$ , dropout 0.05	LoRA	rank 32, $\alpha=64$
Target modules	all linear layers	Advantage	leave-one-out (RLOO)
Learning rate	$1\times 10^{-4}$ , cosine	Learning rate	$1\times 10^{-4}$
Warm-up	10 steps	Batch	16 states $\times$ $G=8$ samples
Batch size	16 sequences	Policy updates	2 per step (8 states each)

Appendix

Table 9: Training hyperparameters. Both stages train LoRA adapters [ Hu et al., 2021 ] ; the RL adapter starts from the merged supervised model.

Gemini VQ gap	Pairs	Gemini’s higher	About the same	Gemini’s lower	Agreement (%)
0 (equal scores)	14	–	6	–	–
$(0,0.25)$	19	4	5	10	29 [7, 50]
$[0.25,0.5)$	26	15	5	6	71 [52, 90]
$[0.5,1)$	31	22	3	6	79 [61, 93]
$\geq 1$	8	5	2	1	83 [50, 100]
All	98	46	21	23	67 [55, 78]

Appendix

Table 10: Human choices by VQ gap. The gap is the absolute difference between the VQ scores that the judge (Gemini) gave the two videos of a pair (98 pairs, one annotator). Among pairs in which both the annotator and the judge chose a side, agreement is the share in which the annotator picked the video the judge scored higher (95% bootstrap interval in brackets). In the 14 pairs with equal judge scores, the annotator answered about the same in 6 and chose a side in 8; in the All row, those 8 count only in the Pairs column.

Opponent	Benchmark	Pairs	Human W / T / L	Human win rate	Gemini W / T / L
GPT-4o	ManimData-test	31	18 / 4 / 9	0.65 [0.48, 0.81]	17 / 7 / 7
GPT-4o	ManimBench	17	7 / 7 / 3	0.62 [0.44, 0.79]	9 / 3 / 5
Qwen3.8-27B	ManimData-test	36	11 / 2 / 23	0.33 [0.19, 0.49]	14 / 3 / 19
Qwen3.8-27B	ManimBench	14	5 / 8 / 1	0.64 [0.50, 0.79]	6 / 1 / 7

Appendix

Table 11: Human and judge (Gemini) verdicts on the same pairs. Wins, ties, and losses of our system against each opponent; for the judge, a tie means equal VQ scores. The human win rate counts a tie as half (95% bootstrap interval in brackets).

Figure 6: More revision steps of our system on ManimData-test. Each row is one revision step; the best-candidate rule decides which version the loop returns. From the top: a crash repaired from the renderer’s error message; a wrong sum corrected in the explanation text; an overlap removed by moving the result below the derivation; and two consecutive revisions of the same request. In the last two rows, moving the transaction list down clears the progress bar but pushes the list past the bottom edge, and moving it back up restores the overlap; our Reviewer scores the second revision lower, so the loop keeps the first. Under each frame are our Reviewer’s mean score and the score that the judge, Gemini 3.8 Flash, gives the re-rendered video (1–5).

Video Generation

Video generation studies models that create and edit moving images from text, images, audio, and structured controls.

See all 376 papers

See Before You Code: Learning Visual Priors for Spatially Aware Educational Animation Generation
May 15, 2026Yuejia Li, Ke He, Junheng Li +5Execution-Guided Code Generation Code Generation
Wuhan University · University of Chinese Academy of Sciences
Training and Agentic Inference Strategies for LLM-based Manim Animation Generation
Apr 20, 2026Ravidu Suien Rammuni Silva, Ahmad Lotfi, Isibor Kennedy Ihianle +2RL for Code Generation T2V Generation
Department of Computer Science, Nottingham Trent University, Nottingham, United Kingdom · Department of Physics and Mathematics, Nottingham Trent University, Nottingham, United Kingdom
ManimAgent: Self-Evolving Multimodal Agents for Visual Education
Jun 29, 2026Wenjia Jiang, Zongyuan Cai, Yuanhang Shao +7Continual Learning for LLM Agents LLM Agent Self-Improvement
University of Alberta · Southeast University · Virginia Tech +2

May 15, 2026J/K move · Enter open · S save

See Before You Code: Learning Visual Priors for Spatially Aware Educational Animation Generation

Yuejia Li, Ke He, Junheng Li, Shutong Chen+4

Wuhan University · University of Chinese Academy of Sciences

Execution-Guided Code Generation Code Generation Generative AI in Education Video Generation

Open paper AskPDF arXiv

Large language models can generate executable code for educational animations, but the resulting renders often exhibit visual defects, including element overlap, misalignment, and broken animation continuity. These defects cannot be reliably detected from the code alone and become apparent only after execution. We formalize this problem as render-feedback-aware constrained code generation: given a natural language specification, the model must generate executable code whose rendered output satisfies structured quality criteria that can be evaluated only after rendering. To address this problem, we introduce OmniManim, a render-feedback-aware educational animation generation framework built around a shared scene state, explicit visual planning, structured post-render diagnostics, and localized repair. Within OmniManim, the Vision Agent is a task-specific visual planning module: it predicts sparse keyframe layouts with coarse-to-fine bounding-box denoising and optimizes an interpolation-aware objective to reduce intermediate-frame failures induced by downstream animation interpolation. We further construct two datasets, ManimLayout-1K and EduRequire-500, and provide a reproducible evaluation protocol covering executability, instructional quality, visual quality, and efficiency. On EduRequire-500, OmniManim improves measured render quality over both single-model baselines and existing multi-agent frameworks. Systematic ablation studies further verify that explicit visual planning, especially its coarse spatial prior, bounding-box refinement, and interpolation-aware optimization, is central to these gains.

Abstract

Video Generation

Explore similar work

See Before You Code: Learning Visual Priors for Spatially Aware Educational Animation Generation

Training and Agentic Inference Strategies for LLM-based Manim Animation Generation

ManimAgent: Self-Evolving Multimodal Agents for Visual Education