CoCo: Code as CoT for Text-to-Image Preview and Rare Concept Generation

Haodong Li; Chunmei Qing; Huanyu Zhang; Dongzhi Jiang; Yihang Zou; Hongbo Peng; Dingming Li; Yuhong Dai

cs.AIMar 9, 2026

CoCo: Code as CoT for Text-to-Image Preview and Rare Concept Generation

Authors: Haodong Li, Chunmei Qing, Huanyu Zhang, Dongzhi Jiang, Yihang Zou, Hongbo Peng, Dingming Li, Yuhong Dai,

Organizations: The Chinese University of Hong Kong · South China University of Technology · Institute of Automation, Chinese Academy of Sciences · Nanyang Technological University

Abstract

Recent advancements in Unified Multimodal Models (UMMs) have significantly advanced text-to-image (T2I) generation, particularly through the integration of Chain-of-Thought (CoT) reasoning. However, existing CoT-based T2I methods largely rely on abstract natural-language planning, which lacks the precision required for complex spatial layouts, structured visual elements, and dense textual content. In this work, we propose CoCo (Code-as-CoT), a code-driven reasoning framework that represents the reasoning process as executable code, enabling explicit and verifiable intermediate planning for image generation. Given a text prompt, CoCo first generates executable code that specifies the structural layout of the scene. The code is then executed in a sandboxed environment to render a deterministic draft image. Subsequently, the model refines this draft through fine-grained image editing to produce the final high-fidelity result. To support this training paradigm, we construct CoCo-10K, a curated dataset containing structured draft-final image pairs designed to teach both structured draft construction and corrective visual refinement. Empirical evaluations on StructT2IBench, OneIG-Bench, and LongText-Bench show that CoCo achieves improvements of 68.83%, 54.8%, and 41.23%, respectively, over direct generation, while also outperforming other CoT-enhanced generation methods. These results demonstrate that executable code is an effective and reliable reasoning paradigm for precise, controllable, and structured text-to-image generation.

Figures & tables

Figure 1: Comparison of reasoning paradigms for text-to-image generation. (a) Direct generation without explicit reasoning. (b) Text-only Chain-of-Thought (CoT) reasoning prior to image synthesis. (c) Textual CoT with an intermediate visual draft. (d) Multi-turn reasoning with iterative textual and visual drafts. (e) CoCo : our approach generates executable code as a structured visual draft that explicitly specifies object layouts and attributes, which is then rendered and refined to produce the final image.

Figure 2: Framework of CoCo. CoCo operates in three stages: (1) generating executable code from the input prompt to explicitly specify structural layouts, (2) executing the code in a sandboxed environment to render a deterministic draft image, and (3) performing draft-guided refinement to produce the final high-quality output.

Figure 3: Visual comparison between code-generated drafts and final results. Left: draft image synthesized from code generated by CoCo . Right: final image refined by CoCo based on the draft image and input prompt.

Figure 4: Construction Pipeline and Examples of CoCo-10K. We design specialized data pipelines targeting three atomic correction capabilities: general editing, scientific diagrams, and complex text. Using Gemini-3-Pro [ 36 ] and Nano Banana [ 37 ] , we generate code and final images from the collected prompts. The resulting data are organized into two training categories: Text-Code Pairs and Text-Draft Image–Final Image Pairs.

Figure 5: Detailed Visualization of CoCo-10K. We present the prompt, generated code, draft image, and final image. CoCo -10K demonstrates high fidelity and strong semantic alignment between draft and final images.

Figure 6: Contrastive visualization of the experiment. CoCo generates accurate layouts and high-fidelity text, outperforming baselines, and supports adaptive aspect ratios that align images with prompt semantics.

Model	Chart $\uparrow$	Graph $\uparrow$	Math $\uparrow$	Puzzle $\uparrow$	Science $\uparrow$	Table $\uparrow$	Overall $\uparrow$
Close-resource models
Seedream 4.0 [ 100 ]	35.79	54.08	63.33	50.89	62.59	68.94	47.52
Nano banana [ 37 ]	35.55	58.96	64.81	63.87	60.75	67.20	48.45
GPT-Image [ 101 ]	37.09	57.00	63.25	59.42	60.94	83.31	49.58
Open-resource models
UniWorld-V1 [ 10 ]	1.71	5.52	4.72	1.58	8.82	5.25	3.20

Table 1: Quantitative comparison on StructT2IBench. All values are reported as Accuracy (%). Orange and Champagne cells indicate the Top 1 and Top 2 performers respectively.

Gen. Only Models
Method	OneIG-Bench [ 40 ]			LongText-Bench [ 39 ]
Method	English	Chinese	Overall	English	Chinese	Overall
FLUX.1-dev [ 103 ]	0.523	-	-	0.607	0.005	0.306
HiDream-I1 [ 102 ]	0.707	0.205	0.456	0.543	0.024	0.284
Kolors 2.0 [ 106 ]	0.427	0.502	0.465	0.258	0.329	0.294
Unified Models

Table 2: Comparison of text rendering capability. All values are reported as Accuracy (%). Orange and Champagne cells indicate the Top 1 and Top 2 performers respectively.

T2I Generation

Text-to-image generation creates images from natural-language prompts using generative models, with research focused on fidelity, control, efficiency, and safety.

See all 299 papers

IV-CoT: Implicit Visual Chain-of-Thought for Structure-Aware Text-to-Image Generation
Jun 23, 2026Zixuan Li, Haokun Lin, Yicheng Xiao +10T2I Generation Multimodal Large Language Models
NLPR, Institute of Automation, Chinese Academy of Sciences · The University of Hong Kong · Ant Group
StruVis: Enhancing Reasoning-based Text-to-Image Generation via Thinking with Structured Vision
Mar 6, 2026Yuanhuiyi Lyu, Kaiyu Lei, Ziqiao Weng +7T2I Generation VLM Reasoning
Hong Kong University of Science and Technology (Guangzhou) · Hong Kong University of Science and Technology · East China Normal University +2
Tree-of-Thoughts Reasoning for Text-to-Image In-Context Learning
Jul 8, 2026Stepanida Alekseeva, Jenifer Kalafatovich, Seong-Whan LeeTree of Thoughts Multimodal ICL
Department of Artificial Intelligence, Korea University, Seoul, Korea.

Jun 23, 2026J/K move · Enter open · S save

IV-CoT: Implicit Visual Chain-of-Thought for Structure-Aware Text-to-Image Generation

Zixuan Li, Haokun Lin, Yicheng Xiao, Zhiwei Li+9

NLPR, Institute of Automation, Chinese Academy of Sciences · The University of Hong Kong · Ant Group

T2I Generation Multimodal Large Language Models Compositional T2I Generation Latent Visual Reasoning

Open paper AskPDF arXiv

Unified multi-modal large language models (MLLMs) have achieved strong text-to-image generation quality, but still struggle with structure-aware prompt following, where object counts, spatial relations, attribute bindings, and coarse layouts must be preserved. We attribute this limitation in part to the entanglement of structural planning and appearance rendering within a single conditioning stream. To address this issue, we propose Implicit Visual Chain-of-Thought (IV-CoT), a latent visual reasoning framework for query-conditioned image generation. IV-CoT decomposes the visual conditioning queries into a structural-to-semantic cascade, where structural queries first form a latent visual plan and semantic queries then render appearance conditioned on this plan. To guide the structural queries, we introduce training-only sketch supervision, which encourages them to capture structure from sketches without requiring sketch extraction or intermediate decoding at inference time. IV-CoT performs implicit CoT reasoning in a single forward pass and achieves superior results on GenEval and T2I-CompBench. Visualizations and analyses demonstrate that the learned structural and semantic queries play complementary roles in structure-aware generation.

Abstract

T2I Generation

Explore similar work

IV-CoT: Implicit Visual Chain-of-Thought for Structure-Aware Text-to-Image Generation

StruVis: Enhancing Reasoning-based Text-to-Image Generation via Thinking with Structured Vision

Tree-of-Thoughts Reasoning for Text-to-Image In-Context Learning