I like making generative models run faster, and run on hardware people actually own. That means hand-written attention kernels for AMD's RX 9070 XT, ComfyUI node packs, one-click RunPod images and tools for running LLMs locally. On the side I write hypervisors.
| Project | What it is | |
|---|---|---|
| SageAttention-RDNA4 | SageAttention for the RX 9070 / 9070 XT, with hand-written HIP attention kernels for fp16 and bf16. Up to 2.3× faster than the existing gfx12 port, and a drop-in for ComfyUI. |  |
| Project | What it is | |
|---|---|---|
| ComfyUI-Kling-Direct | Kling video, image and audio generation inside ComfyUI, with your own API key. |  |
| ComfyUI-NanoBanana2 | The full Gemini API as nodes: text, vision, image, audio, music, video and embeddings. |  |
| ComfyUI-NanoBanana-FaceSwap | Face and head replacement through Gemini image editing. |  |
| ComfyUI-ElevenLabs-Pro | ElevenLabs in ComfyUI: text-to-speech, speech-to-speech, sound effects, music and voice cloning. |  |
| ComfyUI-API-Toolkit | Kling, ElevenLabs and Gemini in one package: 68 nodes. |  |
| ComfyUI-Utility-MegaPack | 156 utility operations in 7 nodes, with 11 visual themes. |  |
| ComfyUI-IxMxAMAR-Masterworks | 12 ready-made workflows that combine the packs above. |  |
| Project | What it is | |
|---|---|---|
| ComfyUI-Ultimate | A batteries-included ComfyUI image for RunPod: 29 curated node packs, SageAttention and FlashAttention, pinned dependencies, CI-gated builds. |  |
| ComfyUI-Ultimate-ST | ComfyUI-Ultimate plus SillyTavern in one pod image. |  |
| runpod-wan22-serverless | Serverless Wan 2.2 14B text-to-video and image-to-video endpoints, with a desktop client. |  |
| ComfyUI-Serverless-FaceSwap | ComfyUI face-swap workflows served as a RunPod API. |  |
| rigma-runpod | Rigma on a RunPod GPU pod: one port behind the proxy, a stock-image template that installs Rigma at boot, and a CI-built image. |  |
| Project | What it is | |
|---|---|---|
| rigma | Hardware-aware local LLM deployment: rigma up probes your GPU and RAM and serves the best-tuned model setup for that exact machine. Community-verified setups live in rigma-registry. |
 |
| raggity | Local-first search and Q&A over your notes, docs and PDFs: hybrid retrieval, reranking and verified citations, as a CLI or an API. |  |
| LoRA-Dataset-Forge | Two photos in, a captioned, aspect-bucketed LoRA training set out, with a review step and one-click export. |  |
| Project | What it is | |
|---|---|---|
| Aria | A Type-1 hypervisor in C that boots through UEFI and virtualizes a running Windows in place, with a full nested VMX engine. |  |
| Visor | A hypervisor experiment in Rust: UEFI loader plus Windows driver, VMCALL dispatch and EPT. |  |
| barevisor | Experiments on top of tandasat's barevisor (Rust). |  |
| SimpleVisor | A modified SimpleVisor, ionescu007's minimal Intel VT-x hypervisor (C). |  |