Serverless GPU Stable Diffusion XL worker for Runpod, powered by diffusers. Generate images from text, transform existing images, or inpaint masked regions — all from a single serverless endpoint. Supports SDXL Base 1.0, SDXL Turbo, SDXL Lightning, Juggernaut XL, Playground V2, plus the official SDXL refiner for a final detail pass.

Serverless GPU/CPU zero-shot text classification on RunPod. Hand the worker any text and an arbitrary list of candidate labels, get back ranked probabilities in milliseconds — no fine-tuning, no training data, no label schema baked into the model. Powered by NLI-style encoders such as BART-MNLI, DeBERTa-v3-MNLI, and multilingual checkpoints...

Serverless background removal worker powered by rembg on Runpod. Accepts public image URLs or base64-encoded images and returns the cut-out RGBA PNG (or WebP/JPG), an optional alpha mask, an optional composited image on a custom background color, and the foreground bounding box. Supports the U2Net, IS-Net, SAM, and BiRefNet model families with GPU

A RunPod serverless worker for Donut (Document Understanding Transformer), NAVER's OCR-free document AI model. Donut takes a document image plus a task prompt and produces structured JSON directly — no separate OCR step, no detector pipeline. The model is small (~200M params), fast on a single GPU, and remarkably accurate on layout-driven documents

Qwen 2.5 7B Instruct OpenAI-compatible vLLM worker for low-cost RunPod Hub chat deployments

A serverless multi-task NLP toolbox built for scalable deployment on RunPod. Expose a single API endpoint to handle multiple natural language processing tasks with low overhead and production-ready performance.

Serverless, pure-CPU PDF processing worker for Runpod. Extracts text, layout-preserving text, Markdown, tables, metadata, embedded images, hyperlinks, bookmarks/outline, form-field values, redaction/annotation listings, and page renders (PNG/JPG). Also runs in-document text search and produces concatenated output. Backed by PyMuPDF (fitz) and ...

Serverless GPU LLM inference worker built on vLLM with an OpenAI-compatible request shape. One worker, one loaded model — vLLM owns CUDA memory and refuses model swaps at runtime, so picking your backbone is an env-var concern, not a per-request one. Switch via LLM_MODEL and redeploy.

Qwen 2.5 32B Instruct OpenAI-compatible vLLM worker for H100-class RunPod Hub deployments

Serverless CPU-only MediaPipe Tasks worker covering the full vision catalog through a single endpoint: pose estimation, hand landmarks, face mesh, holistic (pose + hands + face), object detection, image segmentation, and gesture recognition — plus an annotation mode that overlays everything onto the input image.

Qwen 2.5 Coder 7B OpenAI-compatible vLLM worker for affordable RunPod Hub coding assistants

A RunPod serverless worker for Microsoft TrOCR, a HuggingFace transformer-based OCR model specialized for single-line text recognition — both printed and handwritten.

A serverless OCR worker for RunPod powered by PaddleOCR and PP-Structure, providing high-accuracy multilingual text recognition and structured document understanding. It delivers stronger performance for Asian scripts (Chinese, Japanese, Korean), Arabic-script languages (Arabic, Persian/Farsi, Urdu), and complex document layouts including tables...

Production-grade serverless GPU worker for Flux.1, Black Forest Labs' open-weights text-to-image model that consistently beats SDXL on prompt adherence, typography, and photorealism. Supports text-to-image, image-to-image, inpainting, ControlNet (Canny & Depth), batched prompts, multi-image output per prompt, LoRA loading...

A production-grade fleet of 35 RunPod serverless inference workers — image generation, video, audio, speech, vision, OCR, document AI, NLP, multimodal, and utility — vendored as git submodules. Each worker is its own standalone repository, ships with a tested handler, a CUDA-or-CPU Dockerfile, pinned dependencies, and a RunPod Hub manifest.

Serverless music source separation on RunPod, powered by Demucs v4 (Hybrid Transformer Demucs, htdemucs). Split any track into vocals / drums / bass / other (4-stem) or vocals / drums / bass / other / guitar / piano (6-stem with htdemucs_6s). Every stem is returned as a base64-encoded audio file (WAV by default, MP3 or FLAC on request).

Serverless GPU long-text summarization on RunPod. Wraps a curated set of HuggingFace seq2seq models behind a single handler.py, with chunked map-reduce for long inputs and a CPU-only extractive fallback.

Serverless GPU ControlNet-conditioned image generation for RunPod. Send a reference image plus a text prompt, pick a control type (or stack several), and the worker auto-preprocesses the image into the appropriate control map (canny edges, depth map, openpose skeleton, scribble, soft edges, MLSD line segments, segmentation, lineart, normal map...

Serverless GPU speaker diarization worker for RunPod, built on pyannote.audio. Answers "who spoke when?" and (optionally) combines its answer with a faster-whisper transcript to produce a speaker-labelled, time-aligned conversation.

Serverless GPU music + sound-effect generation on RunPod, powered by Meta's AudioCraft stack.