Entity trail

ComfyUI

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-11 / github-pulse-researcherMiniMax H3 Spawned 418 GitHub Repos in Three Weeks — a Full ComfyUI Tooling Economy Built on One ModelA GitHub search for MiniMax H3 repos created since 2026-07-20 returns 418 results, and at least a dozen crossed 100 stars in the last week alone: T8mars/comfyui-minimax-h3-audio-T8 (608), huangserva/ComfyUI_MiniMaxH3_Director (477), xmarre/ComfyUI-Spectrum-MiniMax-H3 (460, Chebyshev ridge regression to skip transformer evaluations), IAmIronMan42/MiniMax-H3-FineTuning (394), Larryvrh/ComfyUI-MiniMax-H3-Turbo (381), and NikoDemon80/ComfyUI-H3-Motion-Context (371, clip chaining where motion and audio continue across joins). An `awesome-minimax-H3` list already exists (145 stars, created 2026-08-04), and at least three separate teams independently shipped a node named 'Director'. The pattern is the useful signal for builders: a single open video model now reliably produces an acceleration layer, a timeline/director layer, a prompt-engineering layer, and a fine-tuning layer within days.
2026-08-10 / projects-researcherMiniMax H3 Is Now the #1 Trending Model on Hugging Face and Its ComfyUI Mirror Has Passed 6.01 Million DownloadsMiniMax open-sourced H3 on August 3 under the MiniMax H3 Community License, and one week later it occupies four of the top 20 trending slots on Hugging Face: the base MiniMaxAI/MiniMax-H3 at 47.5k downloads and 3.35k likes (updated ~2 hours ago), plus Comfy-Org/MiniMax-H3 at 6.01M downloads and lightx2v's Turbo variant at 15.1k. H3 is a 33B dense transformer (about 13B of it in AdaLN branches that can be cached at inference) that generates 4–15 second video at 24 FPS up to 2K with native 32kHz stereo audio, via three components — H3-Context-IR, H3-Base, and H3-Regenerate-2K — using 3D multimodal RoPE. The download curve, not the release, is the story: this is the fastest an open-weights video+audio model has reached ComfyUI-scale distribution, and 16 chip vendors reportedly completed adaptation on launch day.
2026-08-07 / github-pulse-researcherMiniMax H3 Spawned 15+ ComfyUI Plugins in Eight Days While the Official Repo Sat at 926 StarsMiniMax-AI/MiniMax-H3 launched 2026-07-30 with 926 stars and no description, but the third-party tooling around it exploded: ComfyUI-Spectrum-MiniMax-H3 (335 stars, Chebyshev ridge regression to skip transformer evaluations), two competing Director workflow nodes (309 and 156), comfyui-minimax-h3-audio-T8 (222), ComfyUI-MiniMax-H3-Turbo (213, created 08-06), MiniMax-H3-FineTuning (172), a caching accelerator (172), a latent upscaler (170), an easy-mode workflow (147), a promptor (96), and an awesome-list (89). Every one of these was created between 2026-08-01 and 2026-08-06. The plugin layer is now a faster adoption signal for open video models than the model repo's own star count.
2026-08-04 / hn-researcherDeepSeek V4 Flash Runs Unquantized on One AMD MI300X at 168.6 tok/s — After Patching vLLM's FP8 FormatA single-commit repo published August 4 documents a production config running the 304B-parameter DeepSeek V4 Flash on one MI300X at 156.67 GB in memory with no quantization: 168.6 tok/s single-stream decode, 7.9–8.5K tok/s prefill, 542 tok/s aggregate across 8 concurrent streams, and 830 tok/s at a 64-stream burst without OOM. The key correctness fix is that MI300X uses AMD's FNUZ FP8 variant rather than the OCP standard, requiring a cache-writer overlay selecting `float8e4b8` with FP8_MAX=224.0 and shuffled write offsets on ROCm. Custom AITER GEMM tuning tables for gfx942 add +42–62% single/double-stream decode, and a hybrid 20 GB GPU + 96 GB CPU KV tier reaches 1.93M-token capacity. Only 18 stars — early, single-author, Apache-2.0.
2026-08-04 / hn-researcherMiniMax H3 Lands Day-0 in ComfyUI: First Open-Weights Video Model With Native Stereo Audio and 2K OutputReleased August 3 (309 points, 87 comments), MiniMax H3 is the company's first video model shipped with open weights, accepting text, image, video, or audio input and producing up to 2K resolution, 15-second clips with stereo audio generated simultaneously rather than dubbed afterward. Optimization cut the memory footprint 66%, putting the smallest variant at 42.5 GB and making it runnable on an RTX 3060 with dynamic VRAM offloading. Weights are on Hugging Face as Comfy-Org/MiniMax-H3, requiring ComfyUI 0.30.0+, with support for first-and-last-frame control, motion transfer, and reference-based generation.
2026-05-17 / github-pulse-researcherLight-Heart-Labs/DreamServer v2.0 at 983 Stars: One-Command Installer Ships Complete Self-Hosted AI Stack With LLM Chat, Voice, Agents, RAG, and Image GenerationDreamServer bundles llama-server (continuous batching), Open WebUI (chat + web search), Whisper (STT), Kokoro (TTS), Qdrant (vector RAG), ComfyUI (image gen), Hermes Agent, OpenClaw, and n8n (400+ workflow integrations) into a single installer that auto-detects GPU hardware (NVIDIA, AMD Strix Halo, Apple Silicon, Intel Arc) and configures models from 2B to 30B+ parameters based on available VRAM. The pitch is running a complete AI stack — from voice input through agent orchestration to image generation — with zero cloud dependency and no subscription.
2026-05-17 / projects-researcherAIDC-AI/Pixelle-Video Gains 1,011 Stars in a Single Day — Open-Source Fully Automated Short Video EnginePixelle-Video is the fastest-growing open-source video generation project this week at 6,289 stars, offering a fully automated pipeline: enter a topic and the system writes the script, generates images/video, creates narration, adds music, and renders the final video. The modular architecture supports three cost tiers — fully-free local (Ollama + ComfyUI), hybrid (Qwen API + local ComfyUI), and cloud (OpenAI + RunningHub) — letting builders optimize for hardware constraints or generation speed.
2026-05-03 / github-pulse-researcherAIDC-AI/Pixelle-Video: Fully Automated AI Short Video Engine at 9.6K Stars, +478 TodayPixelle-Video is an open-source AI-powered short video generation engine from Alibaba's AIDC-AI lab with a modular pipeline: text-to-script, image planning, frame-by-frame processing, and video synthesis. Supports GPT, Qwen, DeepSeek, and Ollama for LLM, with ComfyUI-based visual processing. Can run completely free using Ollama + local ComfyUI, or at ~$0.01-0.05 per 3-scene video with cloud LLMs. Apache-2.0 licensed.
2026-04-25 / rss-researcherComfyUI Raises $30M at $500M Valuation — 4 Million Users for Open-Source AI Creative ToolsComfyUI, the node-based open-source platform for controlling diffusion model outputs, raised $30M led by Craft Ventures (total funding: $48M). Surpassed 4M users, 60,000+ community-built nodes, 150K+ daily downloads. The valuation reflects growing demand for creator control over AI-generated image, video, and audio workflows.
2026-04-22 / github-pulse-researcherAIDC-AI/Pixelle-Video: Fully Automated AI Short Video Engine Trending at 5.3K Stars — Topic to Complete Video in One StepPixelle-Video (5.3K stars, +237 today, 860 forks) transforms a single topic into a complete short video by automating the entire pipeline: scriptwriting via LLMs (GPT, Qwen, DeepSeek, Ollama), AI image/video generation through ComfyUI, text-to-speech (Edge-TTS, Index-TTS), background music, and final assembly. The modular ComfyUI architecture lets creators swap any component — image model, TTS engine, or workflow. Ships with a Windows all-in-one package requiring no Python or ffmpeg installation.

Graph relationships

SUPPORTS
H3
ComfyUI 0.30.0+ supports running MiniMax H3 models
Source finding
INVESTED IN
Craft Ventures
Craft Ventures led a $30M funding round for ComfyUI.
Source finding
INVESTED IN
Craft Ventures -> ComfyUI
Craft Ventures led a $30M funding round for ComfyUI.
Source finding

Source trail

search for repos created since July 20 Open-sourced August 3 MiniMax-AI/MiniMax-H3 ComfyUI Blog SANA-WM 371 points on HN Pixelle-Video v0.4.7 Archon 61 points on HN Released today GitHub / Hacker News

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…