ComfyUI v0.28.0: SeedVR2 Upscaling, PID 1.5 Support, and New Native Nodes

news

ComfyUI v0.28.0 adds native SeedVR2 upscaling, NVIDIA PID 1.5 PixelDiT support, convrot int4 optimizations, new Save 3D and Text Overlay nodes, and partner node updates including sync.so sync-3 and Seedream 5 Pro.

ComfyUI v0.28.0 is now available, bringing native support for SeedVR2 upscaling, NVIDIA's PixelDiT PID 1.5 model, convrot int4 quantization, and several new built-in nodes. This release follows closely after v0.27.0's convrot int8 launch and continues the focus on model format optimization and expanded model support.

SeedVR2 Native Support

ComfyUI v0.28.0 adds native SeedVR2 image and video upscaling support. SeedVR2 is an advanced upscaling model that can enhance both images and videos directly within ComfyUI workflows without requiring custom nodes or external scripts. The integration is available through the built-in workflow templates.

NVIDIA PID 1.5 (PixelDiT) Support

The release adds support for NVIDIA's PID 1.5 model, a PixelDiT architecture image generation model. PID 1.5 brings improved image quality and efficiency, and its checkpoint was released by NVIDIA alongside the ComfyUI support.

Convrot int4 Models and Turing Optimizations

Building on the convrot int8 support from v0.27.0, ComfyUI v0.28.0 now supports convrot int4 models with specific optimizations for Turing architecture GPUs (GTX 16xx, RTX 20 series). The int4 format offers further memory savings and speed improvements. The release also includes more optimized int8 and int4 kernels for Turing cards.

New Built-in Nodes

Save 3D (Advanced)

A new family of 3D saving nodes has been added: Save 3D (Advanced), Save Splat, and Save Point Cloud. These nodes enable direct export of 3D outputs from ComfyUI workflows, supporting the growing ecosystem of 3D generative models.

SaveText Node

The new SaveText node (CORE-176) allows saving text outputs to .txt, .md, or .json files directly from your workflow. This is useful for logging prompts, saving analysis results, or exporting structured data.

Text Overlay Node

The Text Overlay node (CORE-137) enables adding text overlays to images and videos directly within ComfyUI, eliminating the need for post-processing in external tools for basic text annotations.

Text Node Re-enabled

The built-in Text node has been undeprecated and re-enabled, providing a simple text input for workflows.

Partner Node Updates

  • sync.so sync-3: New sync-3 model support for improved lip-sync and talking image generation
  • Seedream 5 Pro: Added support for ByteDance's Seedream 5 Pro model
  • Seedream Thinking Control: New widget to disable thinking on Seedream nodes for faster inference
  • Gemini Image: Gemini Image preview models moved to release versions
  • Ideogram: Removed IdeogramV1 and V2 nodes (legacy cleanup)
  • StabilityAI: StabilityAI nodes removed

Performance and Infrastructure

  • GQA (Grouped Query Attention) support enabled on all attention backends
  • ROCm/AMD users get automatic comfy-kitchen Triton backend on matrix-core GPUs
  • Dropped support for PyTorch 2.4
  • Security fix for four vulnerabilities (GHSA-779p-m5rp-r4h4)
  • --models-directory launch argument added for custom model folder paths
  • Various fixes for video encoding, VAE loading, and multi-GPU stability

For the full changelog, visit the official GitHub release or the ComfyUI docs changelog.

ComfyUI v0.28.0: SeedVR2 Upscaling, PID 1.5 Support, and New Native Nodes | ComfyUI Wiki