arXiv subjects

Bulat Suleimanov

Publications and source records attributed to Bulat Suleimanov.

5 recordsLinked to original sources

arXiv2026

VIBE: Visual Instruction Based Editor

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.

cs.CV↗

arXiv2025

NoHumansRequired: Autonomous High-Quality Image Editing Triplet Mining

Recent advances in generative modeling enable image editing assistants that follow natural language instructions without additional user input. Their supervised training requires millions of triplets (original image, instruction, edited image), yet mining pixel-accurate examples is hard. Each edit must affect only prompt-specified regions, preserve stylistic coherence, respect physical plausibility, and retain visual appeal. The lack of robust automated edit-quality metrics hinders reliable automation at scale. We present an automated, modular pipeline that mines high-fidelity triplets across domains, resolutions, instruction complexities, and styles. Built on public generative models and running without human intervention, our system uses a task-tuned Gemini validator to score instruction adherence and aesthetics directly, removing any need for segmentation or grounding models. Inversion and compositional bootstrapping enlarge the mined set by approx. 2.6x, enabling large-scale high-fidelity training data. By automating the most repetitive annotation steps, the approach allows a new scale of training without human labeling effort. To democratize research in this resource-intensive area, we release NHR-Edit, an open dataset of 720k high-quality triplets, curated at industrial scale via millions of guided generations and validator passes, and we analyze the pipeline's stage-wise survival rates, providing a framework for estimating computational effort across different model stacks. In the largest cross-dataset evaluation, it surpasses all public alternatives. We also release Bagel-NHR-Edit, a fine-tuned Bagel model with state-of-the-art metrics.

cs.CV↗

arXiv2023

Typical dropping asymptotics of quasiclassical approximations to solutions of the nonlinear Schrödinger equation

Formal asymptotics are substantiated that describe typical dropping cusp singularity of quasiclassical approximations to solutions of two cases of the integrable nonlinear Schrödinger equation $-i\varepsilonΨ'_{t}=\varepsilon^2Ψ''_{xx}\pm2|Ψ| ^2Ψ$, where $\varepsilon$ is a small parameter. The substantiation uses the ideology and facts of the mathematical catastrophe theory and the part of the theorem of Yu. F. Korobeinik, concerning analytical as $h\to 0$ solutions $G(h,u)$ of the mixed type linear equation $hG''_{hh}=G''_{uu}$ to which the hodograph images of both cases of the systems of equations of these quasiclassical approximations are equivalent.

math-ph↗

arXiv2015

"Quantization" of higher hamiltonian analogues of the Painleve I and Painleve II equations with two degrees of freedom

We construct a solution of an analog of the Schrödinger equation for the Hamiltonian $ H_I (z, t, q_1, q_2, p_1, p_2) $ corresponding to the second equation $P_1^2$ in the Painleve I hierarchy. This solution is produced by an explicit change of variables from a solution of the linear equations whose compatibility condition is the ordinary differential equation $P_1^2$ with respect to $z$. This solution also satisfies an analog of the Schrödinger equation corresponding to the Hamiltonian $ H_{II} (z, t, q_1, q_2, p_1, p_2) $ of Hamiltonian system with respect to $t$ which is compatible with $P_1^2$. A similar situation occurs for the $P_2^2$ equation in the Painleve II hierarchy.

nlin.SI↗

arXiv2013

"Quantum" linearization of Painlevé equations as a component of their $L,A$ pairs

The procedure of the "quantum" linearization of the Hamiltonian ordinary differential equations with one degree of freedom is introduced. It is offered to be used for the classification of integrable equations of the Painleve type. By this procedure and all natural numbers $n$ we construct the solutions $Ψ(\hbar,t,x,n)$ to the non-stationary Shrödinger equation with the Hamiltonian $H = (p^2+q^2)/2$ which tend to zero as $x\to\pm\infty$. On the curves $x=q_n (\hbar, t) $ defined by the old Bohr-Sommerfeld quantization rule the solutions satisfy the relation $i\hbar Ψ'_x\equiv p_n (\hbar, t) Ψ$, where $p_n (\hbar, t) = (q_n (\hbar, t)) '_t $ is the classical momentum corresponding to the harmonic $q_n (\hbar, t) $.

nlin.SI↗