Yehor D.

AI Engineer | ComfyUI Pipelines & Cloud Deployment | Custom LoRA

Specialized in custom AI pipelines: character LoRAs, virtual try-on systems, and identity-preservation engines, plus the deployment layer that keeps them running. Built for AI influencers, startups, YouTube creators, and brands that need consistency across thousands of frames, not generic tools that break on the hard cases. Perfect 5.0 rating across every closed project, 30+ production LoRAs deployed, full handover of files and workflows so you fully own what I build. 🚀 CORE SERVICES ● Pipeline Deployment & Scale — I take a working ComfyUI pipeline and put it into production on your own infrastructure: job queue, on-demand cloud GPU workers, auto-retry, delivery straight into your storage. Your agent posts a job, the finished clip lands where you need it. The last one returns a clip in under 3 minutes at roughly two cents each, around 500 clips a day off a single GPU, and it loses nothing when a worker dies. ● AI Influencer & Character Engines — full identity, style and scene pipeline. Consistent character across thousands of frames, multiple poses, lighting setups, and editorial scenes. ● Virtual Try-On Pipelines — photo-real garment swap for bridal, fashion, and e-commerce. Custom-built, preserves lace, 3D appliqué, sheer mesh, and accessories where IDM-VTON or CatVTON break. ● Custom Character & Style LoRAs — trained for AI influencers, animation studios, and illustrators who want their drawing style replicated at production scale. 🛠 TECHNICAL STACK (depth across the stack, picked per project) ● Diffusion: Flux Klein, Flux.dev, Qwen Image 2512, Z-Image, SDXL,Krea2 ● Identity / Character / Style: native LoRA training on Flux, Qwen, Wan, Z-Image,Krea,LTX,Wan. ● Video: Wan 2.2, LTX-2.5, Seedance 2.0, Kling,MiniMax H3 ● Orchestration: Redis queues, on-demand GPU workers (RunPod, Vast), API endpoints, webhooks, automated delivery ● LLM / VLM: Qwen2.5-VL, Qwen3-VL, Florence-2 ● Segmentation: SAM3, Grounded-SAM ● Training: ai-toolkit, kohya ● Face-aware 4K upscale and restoration chains 💡 THE EDGE 6 years as a professional tattoo artist before AI. The hand-drawn anatomy training catches AI glitches in hands, limbs, faces, and proportions that most pure-engineer freelancers miss. Production-grade identity preservation, not approximations. WHAT YOU GET Final .safetensors files, ComfyUI workflows, and full documentation, yours to run on your own infrastructure. No model lock-in, no recurring license fees. Production-ready from day one.