Paper 2601.02242
VIBE: Visual Instruction Based Editor
- Published
- Jan 2026
- Research lab
- Independent
- Citations
- 2
- GitHub
- 55 stars
01 In brief
Summary
VIBE is a compact, high-throughput instruction-based image editing pipeline that combines a 2B-parameter Qwen3-VL model for instruction interpretation with a 1.6B-parameter Sana1.5 diffusion model for image generation.
The architecture uses channel-wise concatenation for reference image guidance and learnable meta-tokens processed by a lightweight connector to bridge the VLM and diffusion conditioning spaces.
Training follows a four-stage pipeline: connector alignment on text-to-image data, large-scale pre-training, supervised fine-tuning, and Diffusion-DPO preference alignment.
The model is trained on approximately 15 million triplets and 48 million text-to-image pairs, with a focus on strict source consistency and real-world instruction distributions.
On the ImgEdit benchmark, VIBE achieves an overall score of 3.85, ranking second and leading in categories like Adjust, Remove, and Background.
On GEdit-Bench-EN, it scores 6.81 overall with a high semantic consistency of 7.91.
The model fits in 24 GB of GPU memory and generates 2K images in about 4 seconds on an NVIDIA H100 in BF16, matching or exceeding larger baselines while maintaining efficiency.
02 From the paper
Abstract
Instruction-based image editing is among the fastest developing areas in generative AI. Over the past year, the field has reached a new level, with dozens of open-source models released alongside highly capable commercial systems. However, only a limited number of open-source approaches currently achieve real-world quality. In addition, diffusion backbones, the dominant choice for these pipelines, are often large and computationally expensive for many deployments and research settings, with widely used variants typically containing 6B to 20B parameters. This paper presents a compact, high-throughput instruction-based image editing pipeline that uses a modern 2B-parameter Qwen3-VL model to guide the editing process and the 1.6B-parameter diffusion model Sana1.5 for image generation. Our design decisions across architecture, data processing, training configuration, and evaluation target low-cost inference and strict source consistency while maintaining high quality across the major edit categories feasible at this scale. Evaluated on the ImgEdit and GEdit benchmarks, the proposed method matches or exceeds the performance of substantially heavier baselines, including models with several times as many parameters and higher inference cost, and is particularly strong on edits that require preserving the input image, such as an attribute adjustment, object removal, background edits, and targeted replacement. The model fits within 24 GB of GPU memory and generates edited images at up to 2K resolution in approximately 4 seconds on an NVIDIA H100 in BF16, without additional inference optimizations or distillation.