I will build comfyui and wan video workflows as a production API on runpod GPU
About this gig
Have a ComfyUI workflow that works on your desktop but need it running as a reliable API for your app, agency or SaaS?
I turn ComfyUI, Wan, Flux and Stable Diffusion workflows into production services: serverless GPU workers on RunPod, or hosted inference through fal.ai and Runware when you want zero infrastructure.
What I build:
- ComfyUI workflow to API: your JSON graph behind a clean REST endpoint with queues, webhooks and storage
- Wan 2.x and image to animation pipelines on serverless GPUs, cold start tuned, autoscaling
- fal.ai and Runware integrations for Flux, SDXL, LoRA, upscaling and background removal at scale
- Custom LoRA and model training, fine tuning and dataset prep for your product or brand style
- Docker images with pinned models, CI, health checks and acceptance tests
- Cost control: batching, caching, spot GPUs and fallback providers
Proof: a self hosted avatar rendering worker I built has 62 automated tests and cut cost per render by about 45% versus the commercial API, with automatic fallback.
Stack: Python, Docker, RunPod, ComfyUI, PyTorch, Node.js, Supabase.
Message me first with your workflow or model and I will reply with a plan and a fixed quote.
Get to know Usama Ijaz
AI Developer for AI Agents, n8n Automation, Claude Code and GenAI Video
- FromPakistan
- Member sinceJul 2022
- Last delivery4 years
Languages
English, Urdu
Other AI Development Services I Offer
FAQ
Can you host it for me or do I need my own GPU account?
It runs in your own RunPod, fal.ai or Runware account so you own the keys, data and billing. I set everything up, hand over the Docker image and endpoint, and show your team how to update models.
Which models and workflows do you support?
ComfyUI graphs with any custom nodes, Wan 2.x text and image to animation, Flux, SDXL, LoRA, ControlNet, upscalers, background removal and talking avatar models. If it runs in ComfyUI or on Hugging Face, I can serve it.
How do you keep GPU costs down?
Serverless workers that scale to zero, tuned cold starts with baked in models, batching and caching, spot GPUs where safe, and a fallback to a hosted provider like fal.ai so you never lose a job. You see cost per render in the logs.

