Organizations: Northwestern University, Evanston, USA · Northeastern University, Seattle, USA · South China University of Technology, Guangzhou, China · Hong Kong Baptist University, Hong Kong SAR, China · Beijing Normal - Hong Kong Baptist University, Zhuhai, China
Video color grading is a critical post-production process that transforms flat, log-encoded raw footage into emotionally resonant cinematic visuals. Existing automated methods act as static, black-box executors that directly output edited pixels, lacking both interpretability and the iterative control required by professionals. We introduce LumiVideo, an agentic system that mimics the cognitive workflow of professional colorists through four stages: Perception, Reasoning, Execution, and Reflection. Given only raw log video, LumiVideo autonomously produces a cinematic base grade by analyzing the scene's physical lighting and semantic content. Its Reasoning engine synergizes an LLM's internalized cinematic knowledge with a Retrieval-Augmented Generation (RAG) framework via a Tree of Thoughts (ToT) search to navigate the non-linear color parameter space. Rather than generating pixels, the system compiles the deduced parameters into industry-standard ASC-CDL configurations and a globally consistent 3D LUT, analytically guaranteeing temporal consistency. An optional Reflection loop then allows creators to refine the result via natural language feedback. We further introduce LumiGrade, the first log-encoded video benchmark for evaluating automated grading. Experiments show that LumiVideo approaches human expert quality in fully automatic mode while enabling precise iterative control when directed.
Figures & tables
Figure 1. From Pixel Generation to Parameter Reasoning. ①: The perception semantic stream JSON is part of an example.
Figure 2. Architecture of LumiVideo. As an example, we set up max iteration times to 5 here.
Figure 3. Qualitative comparison with other SOTA methods.
Technical Quality
Aesthetic Quality
Grading-Specific (LLM-Judge)
Human Eval
Method
MANIQA ↑
DeQA ↑
Q-Align ↑
CLIP-AS ↑
Harmony ↑
Tonal ↑
Skin ↑
Avg. ↑
Win% ↑
Pro Usab. ↑
GPT-5.4 (DALL·E)
0.441
2.84
3.21
5.52
4.6
4.2
4.1
4.30
1.8
1.3
Gemini 3.1 (NanoBanana 2)
0.418
2.67
3.07
5.38
4.3
3.9
4.0
4.07
1.2
1.2
Camera Official LUT
0.607
3.72
3.64
6.01
5.9
6.0
5.4
5.77
9.4
3.2
Diffusion Lut + CST
0.553
3.41
3.52
6.08
5.5
5.6
5.1
5.40
5.8
2.6
Human Expert †
0.672
4.15
4.45
6.68
7.9
7.6
7.9
7.80
43.6
4.7
Table 1. Quantitative comparison on the LumiGrade benchmark. ↑ : higher is better. Best method in bold , second best underlined .
Figure 4. Demonstration of reflective iteration, the automatic base grade is refined via a single natural language directive.
Technical Quality
Aesthetic Quality
Grading-Specific (LLM-Judge)
Configuration
MANIQA ↑
DeQA ↑
Q-Align ↑
CLIP-AS ↑
Harmony ↑
Tonal ↑
Skin ↑
Avg. ↑
Full LumiVideo
0.681
4.22
4.38
6.79
8.1
7.8
7.8
7.90
w/o ToT (CoT only)
0.634
3.52
3.91
6.34
6.8
6.5
6.5
6.60
w/o RAG
0.651
3.81
4.05
6.43
7.2
6.8
6.8
6.93
w/o Protected Tones
0.654
4.06
4.12
6.51
7.6
7.3
6.2
7.03
w/o Adaptive Lift
0.632
3.68
4.21
6.58
7.7
7.4
7.2
7.43
Table 2. Ablation study on the LumiGrade benchmark. All metrics are no-reference. ↑ : higher is better.
Color affects how we interpret image style and emotion. Previous color grading methods rely on patch-wise recoloring or fixed filter banks, struggling to generalize across creative intents or align with human aesthetic preferences. In this study, we propose AceTone, the first approach that supports multimodal conditioned color grading within a unified framework. AceTone formulates grading as a generative color transformation task, where a model directly produces 3D-LUTs conditioned on text prompts or reference images. We develop a VQ-VAE based tokenizer which compresses a 3×323 LUT vector to 64 discrete tokens with ΔE<2 fidelity. We further build a large-scale dataset, AceTone-800K, and train a vision-language model to predict LUT tokens, followed by reinforcement learning to align outputs with perceptual fidelity and aesthetics. Experiments show that AceTone achieves state-of-the-art performance on both text-guided and reference-guided grading tasks, improving LPIPS by up to 50% over existing methods. Human evaluations confirm that AceTone's results are visually pleasing and stylistically coherent, demonstrating a new pathway toward language-driven, aesthetic-aligned color grading.
Tianren Ma, Mingxiang Liao, Xijin Zhang +1
University of Chinese Academy of Sciences · ByteDance
Independent control of tonescale regions (e.g., shadows, highlights) is essential for painters, photographers and cinematographers to bring 2D images to life. In image manipulation software this is most directly addressed by color grading modules, which use intensity thresholds to segment distinct illumination regions for local manipulation. In this work we develop an open source color grading tool and use it to annotate a large dataset of video frames with tonescale region thresholds. Using these thresholds we conduct modeling experiments with strategies based on both practitioners' conventional wisdom and machine learning. Results show that K-nearest neighbors is an effective prediction strategy, outperforming state-of-the-art end-to-end methods for image enhancement. This outcome demonstrates the benefit of focusing on a compact set of core parameters when modeling creative stylization processes. Our adaptive color grading interface and data are available at https://github.com/SamsungLabs/adaptive-color-grading.
Trevor D. Canham, Abhijith Punnappurath, Michael S. Brown
Modern black-box Image-to-Video (I2V) models offer powerful capabilities in automated content creation, yet their lack of fine-grained control and reliability presents significant challenges in professional workflows. Their inherent stochasticity causes minor variations in textual prompts or hyperparameters to yield drastically different outputs often necessitating inefficient, brute-force trial-and-error processes. To address these limitations, we introduce the ``Agentic Self-Improvement" framework, which reframes video synthesis into a closed-loop, goal-directed optimization. Our framework systematically navigates the generation parameter space using a novel two-stage approach. In the first stage, an iterative prompt optimization loop uses a multimodal Large Language Model (mLLM) to refine the input prompt. This refinement implements two automated evaluations: Davidsonian Scene Graph (DSG) queries ensure semantic adherence, and Common Mistake Questions (CMQ) for artifact detection. At the second stage, we use Bayesian optimization to efficiently co-optimize stochastic seeds and CFG scales. This search is guided by a suite of quality metrics, including the novel Video-Text Adherence (VTA) score derived from the DSG and CMQ evaluations. Our framework significantly outperforms unguided search methods: in human preference studies, videos generated via our agentic approach were strongly preferred over baseline outputs, achieving win rates up to 69%. This work provides a practical and extensible methodology for enhancing the predictability and control of state-of-the-art video generation models, moving the field beyond speculative curiosities toward reliable, production-ready tools.