You will get a personalized LoRA for Stable Diffusion using ComfyUI
Project details
I specialize in training high-quality custom LoRA models using ComfyUI for Stable Diffusion. Whether you need a consistent AI character, brand model, product visuals, or a unique art style, I handle the full pipeline — dataset review, preprocessing, training, optimization, and delivery.
My workflows are built for professional use, ensuring strong prompt responsiveness, visual consistency, and production-ready outputs. You’ll receive a tested LoRA model with guidance for generating the best results.
Perfect for creators, brands, agencies, and AI teams looking for reliable custom models.
My workflows are built for professional use, ensuring strong prompt responsiveness, visual consistency, and production-ready outputs. You’ll receive a tested LoRA model with guidance for generating the best results.
Perfect for creators, brands, agencies, and AI teams looking for reliable custom models.
AI Applications
AI-Generated Art, Image Analysis, Image Processing, Image Recognition, Image UpscalingWhat's included
| Service Tiers |
Starter
$80
|
Standard
$180
|
Advanced
$440
|
|---|---|---|---|
| Delivery Time | 2 days | 5 days | 7 days |
Number of Revisions | 2 | 2 | 3 |
AI Model Integration | - | - | |
Batch Normalization | - | - | - |
Database Integration | - | - | - |
Detailed Code Comments | - | - | - |
Image Upscaling | |||
MLOps | - | - | |
Model Deployment | - | - | |
Model Documentation | - | - | |
Model Monitoring | - | - | - |
Model Testing & Optimization | - | ||
Model Tuning | |||
Natural Language Processing | - | - | - |
NLP Tokenization | - | - | - |
Pre-Training | - | ||
Prompt Engineering | |||
Setup File | |||
Source Code | - |
Optional add-ons
You can add these on the next page.
Fast Delivery
+$10 - $80Frequently asked questions
2 reviews
(1)
(0)
(1)
(0)
(0)
This project doesn't have any reviews.
AG
Aravind G.
Jun 20, 2025
Figma development
AG
Aravind G.
May 14, 2025
Figma Designer Needed for Web Application Design
About Devansh
AI Graphics & Video Systems Engineer |ComfyUI, SDXL, Flux, RunPod, API
33%
Job Success
Kanpur, India - 7:57 pm local time
My work spans the full technical stack: model selection, ComfyUI workflow architecture, LoRA and DreamBooth training, serverless deployment on RunPod, and API integration so your pipeline connects to the rest of your product. I don't hand off a JSON file and disappear I make sure it works in your environment, with your data, at your volume.
What separates my work from most ComfyUI freelancers is that I operate at the intersection of image generation, video synthesis, and LLM-backed orchestration. I've run inference optimization work, built fine-tuning pipelines using ai-toolkit and kohya, and worked through real production issues CUDA compatibility, node dependency conflicts, driver mismatches, cold-start latency the problems that don't show up in tutorials.
What I deliver
- ComfyUI workflows built for your specific use case product photography, AI influencers, virtual try-on, headshots, ad creatives designed to be reproducible and scalable, not just impressive in a single run
- Custom LoRA training on FLUX, SDXL, and WAN for character consistency, product identity, brand style, or artistic replication with dataset curation, training configuration, and evaluation
- Video pipelines using WAN 2.2, LTX-Video, and Kling with temporal consistency, pose control, and character anchoring across frames
- RunPod serverless deployment and optimization GPU selection, cold-start tuning, API wrapping, cost-per-generation analysis
- Identity preservation and face-aware pipelines using IP-Adapter, PuLID, ControlNet, and 4K upscale + restoration chains
- End-to-end automation via n8n or direct API integration so AI generation slots into your existing product or workflow
Technical stack
Diffusion: FLUX.1 dev/schnell, SDXL, WAN 2.2, Z-Image, Qwen-Image
LoRA / Training: ai-toolkit, kohya_ss, DreamBooth
Video: WAN 2.2, LTX-Video, Kling, Seedance
Identity / Character: IP-Adapter, PuLID, ControlNet, Redux, native LoRA
Segmentation: SAM2, Grounded-SAM, Florence-2
Upscaling: face-aware 4K chains, tiled upscale, restoration
Deployment: RunPod (serverless + persistent), CUDA 12.x, ComfyUI API, n8n
LLM integration: prompt routing, output evaluation, pipeline orchestration
Common builds
- AI influencer and digital avatar pipelines with locked character identity across scenes
- Product lifestyle image generators for e-commerce brands (Shopify, Amazon, ad creatives)
- Virtual try-on systems for fashion and bridal preserving fabric texture, lace, and accessories where off-the-shelf tools break
- Headshot and persona apps with consistent lighting, style, and identity
- Bulk ad creative variation systems for Meta, TikTok, and Google Ads
- Sketch / blueprint / photo-to-render workflows for architecture, fashion, and product design
How I work
Short scoping call to understand your use case, data, and success criteria. A pilot run to validate output quality and cost-per-asset before full build. Production handover includes workflow files, model weights, documentation, and a working deployment — yours to run independently with no lock-in.
If you need AI visuals and video that behave like reliable software, not a demo you can't reproduce, message me.
Steps for completing your project
After purchasing the project, send requirements so Devansh can start the project.
Delivery time starts when Devansh receives requirements from you.
Devansh works on your project following the steps below.
Revisions may occur after the delivery date.
Requirements & Dataset Review
Review images, use-case, and training goals.
Dataset Preparation
Cleaning, tagging, normalization, and optimization.