Comfy Internals | How We Got Four Rival AI Labs to Fight over Our Code Reviews
Comfy built an automated code‑review pipeline that fans a pull‑request diff out to four large‑language models from OpenAI, Anthropic, Google, and Moonshot, each running two passes (adversarial and edge‑case). The eight raw reviews are distilled by a Claude Opus judge that surfaces the ten most critical findings. Hosted as a GitHub Action, the system runs in CI for a flat $200 per month and has already vetted about 110 PRs, catching bugs that human reviewers missed. The open‑source workflow is now shared across Comfy’s teams and available to the public.
May Wrapped
In May, ComfyUI expanded its ecosystem with 11 new foundation models covering image, video, 3D, audio, and multimodal capabilities. Highlights include Krea 2 for style‑driven image generation, Netflix’s open‑source Void for comprehensive video object removal, and Google DeepMind’s Gemma 4, a 31‑billion‑parameter...
Krea 2 Image Is Now Available via Partner Nodes
Krea has released its first foundation image model, Krea 2, as a Partner Node for ComfyUI. The node supports prompt‑driven generation with adjustable creativity, moodboard conditioning, and optional style‑reference input. Two variants are offered—Medium, a faster, cost‑efficient option tuned for illustration...
Stable Audio 3.0 Day-0 Support in ComfyUI：From Sound Effects to Longer, More Musical Tracks
Stability AI’s Stable Audio 3.0 family is now available in ComfyUI, bringing fully licensed, AI‑generated music and sound effects to a visual workflow platform. The release includes lightweight Small‑SFX and Small‑Music checkpoints that run on CPUs and a Medium model...
Tripo 3.1 in ComfyUI: Production-Ready, High-Detail 3D Asset Generation.
ComfyUI now supports Tripo 3.1 through Partner Nodes, enabling high‑detail 3D asset generation within existing workflows. The integration offers high‑density geometry, PBR‑ready materials, and hero‑level quality for close‑up production use. Users can create models from text, images, or multiview inputs and...
Luma Uni-1 Is Now Available via Partner Nodes
Luma’s Uni‑1 image model is now accessible in ComfyUI via Partner Nodes, allowing users to run the model directly in the popular open‑source workflow or on Comfy Cloud. Unlike diffusion‑based generators, Uni‑1 is a decoder‑only autoregressive transformer that treats text...
HappyHorse 1.0 Is Now Available in ComfyUI
Alibaba’s HappyHorse 1.0 cinematic video model is now integrated into the open‑source ComfyUI platform. The model supports text‑to‑video, image‑to‑video and subject‑to‑video creation, delivering up to 15‑second 1080p clips with multi‑shot consistency. It also introduces V2V and SV2V editing workflows that...
ComfyUI Raises $30M to Scale Open-Source AI for Creative Production
ComfyUI announced a $30 million Series A at a $500 million valuation, raising its total capital to $47 million. The open‑source AI workflow platform now supports a community of 4 million users, 60,000+ community‑built nodes and 150,000 daily downloads. Studios, agencies and enterprise teams are...
GPT Image 2 Is Now Here via Partner Nodes
OpenAI has launched GPT Image 2.0, now accessible through ComfyUI Partner Nodes and Comfy Cloud. The model adds a reasoning step before pixel generation, allowing it to plan composition, verify output, and iterate, which dramatically improves rendering of dense text,...
ACE-Step 1.5 XL: Commercial-Grade Music Generation in ComfyUI
ACE-Step 1.5 XL introduces a 4‑billion‑parameter Diffusion Transformer decoder to the ACE‑Step framework, delivering commercial‑grade music quality that rivals leading models like Suno v4.5 and v5. The suite offers three variants—xl‑base, xl‑sft, and xl‑turbo—each optimized for versatility, audio fidelity, or...
ComfyUI Now Supports Sonilo via Partner Nodes
ComfyUI now integrates Sonilo, an AI‑powered video‑to‑music service, through Partner Nodes. The new node generates frame‑by‑frame soundtracks that align with a video's length, pacing, and emotional cues, delivering full‑length tracks in seconds. All output includes commercial‑ready usage rights, removing licensing...
Seedance 2.0 Is Now Available in ComfyUI
Seedance 2.0, the viral multimodal video generation model, is now available as a native node in ComfyUI. The integration adds text‑to‑video, reference‑to‑video, and first‑last‑frame‑to‑video workflows, letting users create cinematic clips with synchronized audio and consistent characters. The model accepts up...
Topaz Starlight Precise 2.5 Now Available via Partner Nodes
Topaz has released Starlight Precise 2.5, an upgraded video‑upscaling model now offered as a Partner Node in ComfyUI. The new version delivers noticeably sharper 720p‑to‑4K results with fewer artifacts while preserving creative detail. It is a drop‑in replacement for the...
Dynamic VRAM in ComfyUI: Saving Local Models From RAMmageddon
ComfyUI has launched Dynamic VRAM, a memory‑optimization layer that shifts model weights onto GPU memory on demand, dramatically lowering system RAM consumption. The feature, available for Nvidia GPUs on Windows and Linux, eliminates out‑of‑memory crashes and speeds up model loading...
The Complete Style Transfer Handbook: All in ComfyUI
Style transfer tools in AI image generation have matured, offering creators multiple pathways to apply precise visual aesthetics. Out‑of‑the‑box models like Recraft, Nano Banana Pro, Grok Image Edit and Seedream 5.0‑lite enable reference‑based transfers, while custom LoRAs extend capabilities for...