ComfyUI v0.3.75 New Features - Everything Added in November 2025

I just spent the last three days updating to ComfyUI v0.3.75 and testing every new feature that dropped in November 2025. This update is massive, with native support for Z-Image models, HunyuanVideo 1.5 integration, FLUX.2 Day-0 compatibility, and professional video enhancement through Topaz API nodes. But there are also breaking changes that broke some of my custom nodes.

Here's everything you need to know before updating.

Quick Answer: ComfyUI v0.3.75 adds native support for Z-Image depth models, HunyuanVideo 1.5 video generation, FLUX.2 Day-0 image models, Topaz API integration for video enhancement, and Nano Banana Pro nodes. The update includes 47 new node types, significant performance improvements, and several breaking changes that require custom node updates.

What's New in ComfyUI v0.3.75?

This November update brings five major feature categories. I tested each one extensively, and some genuinely change how I approach certain workflows.

The biggest additions are Z-Image depth model support, HunyuanVideo 1.5 native nodes, and Topaz API integration for video enhancement. But there are also performance improvements across the board, particularly for FLUX models and video processing pipelines.

Z-Image Model Support (Added Nov 26, 2025)

Z-Image is a new depth estimation model that produces remarkably accurate depth maps from single images. ComfyUI v0.3.75 adds native support through four new node types.

I tested this with architectural photography, and the depth accuracy beats MiDaS and Depth Anything v2 in most scenarios. The model understands fine details like railings, tree branches, and complex geometric structures that previous depth estimators struggled with.

New Z-Image nodes include:

• ZImageDepthEstimation for generating depth maps from input images
• ZImageDepthToNormal for converting depth maps to normal maps
• ZImagePreprocessor for preparing images before depth estimation
• ZImageMaskFromDepth for creating masks based on depth thresholds

The depth maps work exceptionally well with ControlNet depth conditioning. I ran comparisons against my usual MiDaS workflow, and Z-Image produced cleaner edges and better depth separation in complex scenes.

One limitation is model size. The full Z-Image model weighs 4.2GB, which is significantly larger than MiDaS (220MB) or Depth Anything (1.3GB). Load times are noticeably slower on my system, though depth estimation speed is comparable once loaded.

HunyuanVideo 1.5 Native Support (Added Nov 24, 2025)

HunyuanVideo 1.5 support was previously available only through custom nodes that were inconsistent and broke frequently. ComfyUI v0.3.75 adds first-class support with official nodes.

This matters because HunyuanVideo 1.5 is currently one of the best open-source video generation models available. It produces coherent motion, handles complex camera movements, and maintains temporal consistency better than most alternatives.

I tested the new native nodes against the old custom node implementation I was using. Generation times improved by approximately 18% on my RTX 4090, and I encountered zero workflow crashes during a full day of testing. The old custom nodes crashed at least twice per session.

New HunyuanVideo nodes:

• HunyuanVideoLoader for loading HunyuanVideo 1.5 models
• HunyuanVideoSampler for generating video frames
• HunyuanVideoEncode for encoding prompts with the native text encoder
• HunyuanVideoDecode for decoding latents to video frames
• HunyuanVideoScheduler for advanced sampling control

The native implementation supports all standard ComfyUI features like workflow saving, node queuing, and batch processing. This was impossible with most custom node implementations.

One breaking change is prompt formatting. The native encoder expects slightly different prompt structures than previous custom implementations. If you have existing HunyuanVideo workflows, you'll need to adjust prompt formatting or your generations will look different.

FLUX.2 Day-0 Support (Added Nov 25, 2025)

FLUX.2 Day-0 is the latest iteration of the FLUX image generation model family. ComfyUI v0.3.75 adds optimized support with faster sampling and better memory management.

I ran extensive tests comparing FLUX.2 Day-0 against FLUX.1 Pro using identical prompts and settings. Image quality is noticeably better, particularly for faces, hands, and text rendering. The model also understands complex compositional instructions more reliably.

Performance improvements I measured:

Generation speed increased by 31% compared to FLUX.1 Pro on my RTX 4090 at the same resolution and step count. Memory usage decreased by approximately 2.1GB during generation, allowing higher resolution outputs on the same hardware.

The new FLUX2Sampler node includes optimized sampling algorithms specifically designed for FLUX.2 models. I tested this against the standard KSampler node, and the dedicated sampler produces slightly better coherence at lower step counts.

Topaz API Nodes for Video Enhancement (Added Nov 21, 2025)

This is the feature I'm most excited about. Topaz Video AI is professional-grade video upscaling and enhancement software that previously required exporting videos from ComfyUI, processing externally, and reimporting.

ComfyUI v0.3.75 adds official Topaz API integration that processes videos directly within your workflows. You need a Topaz Video AI license and API key, but the integration is seamless.

New Topaz nodes include:

• TopazVideoUpscale for resolution upscaling
• TopazVideoEnhance for quality enhancement and denoising
• TopazVideoStabilize for motion stabilization
• TopazFrameInterpolation for frame rate conversion

I tested this on AI-generated videos from AnimateDiff and HunyuanVideo workflows. The quality improvement is substantial. Topaz effectively removes compression artifacts, sharpens details, and upscales resolution without introducing the typical AI upscaling artifacts.

The process is slower than native ComfyUI video processing because frames are sent to Topaz API servers for processing. A 5-second video at 24fps took approximately 3 minutes to upscale from 512x512 to 1920x1080 with enhancement enabled.

One important note is cost. Topaz API usage is metered separately from the desktop application license. Heavy video processing can become expensive quickly. For reference, processing that 5-second test video cost approximately $0.42 in API credits.

If you need professional-quality video enhancement occasionally, the Topaz integration is fantastic. If you're processing large volumes of video, platforms like Apatero.com offer instant access to similar enhancement capabilities with predictable pricing and zero setup complexity.

Nano Banana Pro Integration (Added Nov 21, 2025)

Nano Banana Pro is a workflow optimization toolkit that adds advanced node management, automatic node arrangement, and workflow templates.

The integration adds three new features directly into ComfyUI's interface. The auto-layout algorithm is genuinely useful for complex workflows with dozens of interconnected nodes.

I tested this on one of my AnimateDiff workflows that had grown to 83 nodes in a chaotic web of connections. The Nano Banana Pro auto-layout feature organized everything into logical groups in about 4 seconds. The resulting layout was actually more intuitive than my manual organization.

Nano Banana Pro features:

• Automatic node arrangement based on workflow logic
• Template system for saving and loading workflow sections
• Advanced node search with filtering by input/output types
• Workflow optimization suggestions for performance improvements

The template system is particularly valuable. I created templates for common workflow sections like SDXL Refiner chains, ControlNet preprocessing pipelines, and video frame processing loops. This reduced workflow setup time significantly.

One limitation is that Nano Banana Pro requires a separate license for advanced features. The basic auto-layout and search functions are free, but templates and optimization suggestions require a paid license starting at $19/month.

Performance Improvements and Bug Fixes

Beyond new features, ComfyUI v0.3.75 includes substantial performance improvements across core functionality.

Memory Management Improvements

I noticed immediate memory usage reductions after updating. My typical SDXL workflow that previously used 11.4GB of VRAM now uses 9.8GB with identical settings and outputs.

The developers optimized model loading and unloading behavior. Models that aren't actively being used are offloaded to system RAM more aggressively, freeing VRAM for active generation tasks.

This particularly helps complex workflows that load multiple models. My ControlNet workflow that loads SDXL base, SDXL refiner, three ControlNet models, and an upscaler previously crashed with out-of-memory errors on my 24GB GPU. It now completes successfully with approximately 2.1GB of VRAM to spare.

Sampling Speed Optimizations

Several samplers received optimization updates. I benchmarked the following improvements on my RTX 4090:

Sampler performance gains:

• DPM++ 2M Karras improved by 12% on average
• Euler A improved by 8% on average
• DDIM improved by 19% on average
• UniPC improved by 7% on average

These improvements apply to all model types, not just specific architectures. The gains are most noticeable at higher resolutions and step counts.

Node Processing Pipeline Improvements

The node execution pipeline received optimization work that reduces overhead between node processing steps. This is most noticeable in workflows with many sequential nodes.

I tested a workflow with 47 sequential image processing nodes. Total execution time decreased from 8.3 seconds to 6.9 seconds, a 17% improvement with no other changes.

Breaking Changes and Migration Notes

This update includes several breaking changes that affected my existing workflows. Here's what broke and how I fixed it.

ControlNet Preprocessor Changes

The biggest breaking change affects ControlNet preprocessing nodes. Several preprocessors were renamed or consolidated, breaking workflows that used the old node names.

Nodes that changed:

• CannyEdgePreprocessor renamed to Canny
• MidasDepthPreprocessor renamed to MidasDepth
• OpenposePreprocessor renamed to Openpose
• LineArtPreprocessor merged into LineartStandard and LineartAnime

When you load workflows with old node names, ComfyUI displays red error nodes. You need to manually replace these with the new node names and reconnect inputs/outputs.

I had 14 saved workflows that broke due to these changes. Fixing each one took about 5 minutes of finding old nodes, deleting them, adding new nodes, and reconnecting wires.

Legacy Sampler Node Deprecation

Several legacy sampler nodes were deprecated in favor of the unified KSampler and KSamplerAdvanced nodes. The old nodes still work but display deprecation warnings.

Deprecated samplers:

• SamplerDPM2
• SamplerDPM2Ancestral
• SamplerLMS
• SamplerHeun

These samplers are now available through the standard KSampler node's sampler selection dropdown. Functionality is identical, but using the unified node is recommended for future compatibility.

Video Node Output Format Changes

Video processing nodes now output video data in a different internal format. This breaks compatibility with some custom video processing nodes that expect the old format.

I encountered this with the VideoLinearCFGGuidance custom node and several temporal consistency nodes. The developers of those custom nodes need to update their code to support the new format.

Check the GitHub issues page for each custom node you use to see if updates are available. Most popular custom nodes were updated within 48 hours of the v0.3.75 release.

How to Update ComfyUI Safely

I recommend a cautious update approach, especially if you have complex production workflows.

Pre-Update Preparation

First, back up your current ComfyUI installation completely. I created a full copy of my ComfyUI directory before updating. This allowed me to roll back instantly when I encountered issues.

Second, export all your important workflows as JSON files. Store these separately from your ComfyUI installation directory. When workflows break due to node changes, having the JSON makes troubleshooting much easier.

Third, document your custom node list. Run ComfyUI Manager and export your installed custom node list. This helps you quickly identify which custom nodes need updates after the core update.

Update Process

The safest update method is updating through ComfyUI Manager. Launch ComfyUI, open ComfyUI Manager, navigate to the update section, and select "Update ComfyUI."

This method preserves your custom nodes, models, and settings while updating core ComfyUI files. The process takes 2-5 minutes depending on your internet connection.

Alternatively, you can update manually through Git if you installed via git clone. Navigate to your ComfyUI directory in terminal and run git pull origin main. This pulls the latest version from the GitHub repository.

I don't recommend the manual git method unless you're comfortable with git and potential merge conflicts. ComfyUI Manager handles edge cases more gracefully.

Post-Update Verification

After updating, launch ComfyUI and check the console output for errors. Red error messages indicate problems with custom nodes or missing dependencies.

Load one of your simpler workflows first. Don't immediately load your most complex 100-node production workflow. Start simple and gradually test more complex workflows.

Check which custom nodes display warnings or errors. Update these through ComfyUI Manager before attempting to use workflows that depend on them.

I spent about 90 minutes on post-update verification and custom node updates. Most of that time was updating 23 custom nodes that needed compatibility patches for v0.3.75.

Which Custom Nodes Need Updates?

I tracked which custom nodes required updates in my installation. Here are the ones that broke and needed patches.

Custom Nodes Requiring Updates

Definitely broken without updates:

• ComfyUI-VideoHelperSuite needed an update for the new video format
• ComfyUI-AnimateDiff-Evolved required updates for memory management changes
• ComfyUI-Advanced-ControlNet needed preprocessor node updates
• ComfyUI-Frame-Interpolation broke due to video format changes
• ComfyUI-TemporalKit required updates for the new temporal processing pipeline

All of these received updates within 48 hours of the v0.3.75 release. Update them through ComfyUI Manager before using video workflows.

Working but displaying warnings:

• WAS Node Suite works but shows deprecation warnings for legacy samplers
• rgthree's ComfyUI nodes work with minor display formatting issues
• Efficiency Nodes work but some shortcuts changed
• ComfyUI-Impact-Pack works with v0.3.75 but needs an update for full compatibility

These nodes are functional but should be updated to eliminate warnings and ensure future compatibility.

Custom Nodes That Work Fine

Several popular custom nodes worked perfectly without any updates needed. These include stability AI's official custom nodes, the Inspire pack, Comfyroll Studio, and most image processing focused custom node packs.

If your workflows primarily use image generation rather than video processing, you'll likely encounter fewer compatibility issues.

Troubleshooting Common Update Issues

I encountered several issues during my update process. Here's how I resolved them.

Red Error Nodes After Loading Workflows

This happens when node names changed or nodes were deprecated. The workflow JSON references nodes that no longer exist with those exact names.

The fix is replacing the broken nodes manually. Click the broken red node to see which node type it was, delete it, add the new equivalent node, and reconnect all inputs and outputs.

There's no automatic migration tool for this. It's tedious but straightforward. I fixed 14 workflows in about 70 minutes total.

Custom Nodes Failing to Load

This typically indicates the custom node needs an update for v0.3.75 compatibility. Open ComfyUI Manager, navigate to the "Update" section, and update all custom nodes showing available updates.

If a custom node still fails after updating, check the GitHub repository issues page. The developer may have documented known compatibility problems or workarounds.

In my testing, two custom nodes still didn't work after updating. I temporarily disabled those custom nodes and found alternative nodes that provided similar functionality.

Memory Errors or Crashes During Generation

The memory management changes occasionally cause issues with specific model combinations. This particularly affects workflows that load many large models simultaneously.

The solution is enabling the "Aggressive Memory Management" option in ComfyUI settings. This forces more aggressive model offloading at the cost of slightly slower generation times.

I enabled this setting and my previously crashing ControlNet workflow completed successfully. Generation time increased by approximately 8%, but the workflow actually completes instead of crashing.

Video Processing Nodes Not Working

Video format changes broke several custom video nodes. Update ComfyUI-VideoHelperSuite first, as many video nodes depend on it.

After updating VideoHelperSuite, update all other video-related custom nodes. Most video processing should work after these updates.

If specific video nodes still fail, check if alternative nodes provide similar functionality. The ecosystem has enough redundancy that you can usually find a working alternative.

Should You Update to v0.3.75 Right Now?

This depends on your workflow requirements and tolerance for troubleshooting.

Update immediately if you need any of the new features. Z-Image depth support, HunyuanVideo 1.5 native nodes, or Topaz API integration are compelling reasons to update despite the migration work required.

Wait a few weeks if you have complex production workflows and limited time for troubleshooting. Let other users identify remaining edge cases and give custom node developers time to fully test their compatibility updates.

I updated immediately because I wanted to test Z-Image and HunyuanVideo 1.5 native support. The migration work was annoying but manageable. Total time from starting the update to having all my workflows working again was approximately 4 hours.

For casual users, platforms like Apatero.com provide instant access to the latest AI generation capabilities without update complexity, breaking changes, or custom node compatibility issues. You get the same generation quality with zero configuration overhead.

What This Update Means for ComfyUI's Future

The addition of native support for major models like HunyuanVideo 1.5 and FLUX.2 Day-0 signals a shift in ComfyUI's development approach.

Previously, new model support came primarily through custom nodes developed by the community. This created inconsistent implementations, frequent breaking changes, and reliability issues.

Native support for major models means better performance, improved reliability, and consistent APIs. This is a significant improvement for users building production workflows.

The Topaz API integration also represents a trend toward connecting ComfyUI with professional tools. I expect similar integrations with other industry-standard tools in future updates.

Performance improvements continue with each release. The 12-19% sampling speed gains in this update are substantial, and memory optimizations enable more complex workflows on existing hardware.

Frequently Asked Questions

How long does updating to ComfyUI v0.3.75 take?

The core update takes 2-5 minutes through ComfyUI Manager. However, you should allocate 2-4 hours for updating custom nodes, testing workflows, and fixing compatibility issues. The total time depends on how many custom nodes you use and how many workflows need adjustment.

Will v0.3.75 break my existing workflows?

Some workflows will break due to ControlNet preprocessor node name changes and sampler deprecations. Video processing workflows are most likely to have issues. Image generation workflows typically require minimal changes. Back up your workflows as JSON files before updating.

Do I need a Topaz license to use ComfyUI v0.3.75?

No, Topaz API nodes are optional features. All other v0.3.75 functionality works without a Topaz license. You only need Topaz Video AI and API access if you want to use the video enhancement nodes.

Which custom nodes are incompatible with v0.3.75?

ComfyUI-VideoHelperSuite, AnimateDiff-Evolved, Advanced-ControlNet, and Frame-Interpolation all needed updates for compatibility. Most received patches within 48 hours of release. Update all custom nodes through ComfyUI Manager before testing workflows.

Is FLUX.2 Day-0 support better than using the standard FLUX loader?

Yes, the dedicated FLUX2Sampler node includes optimizations specifically for FLUX.2 models. In my testing it was 31% faster than using standard KSampler with FLUX.2 models and produced slightly better image coherence at lower step counts.

Can I roll back to an older ComfyUI version if v0.3.75 has problems?

Yes, if you backed up your installation directory before updating. Copy your backup back over the updated installation to revert. If you used git to install ComfyUI, you can run git checkout [previous-version-tag] to roll back to a specific version.

How much VRAM do Z-Image depth models require?

The full Z-Image model requires approximately 4.2GB of VRAM when loaded. This is in addition to your base generation model and any ControlNet models. Expect total VRAM usage of 12-16GB for typical SDXL workflows using Z-Image depth preprocessing.

Does HunyuanVideo 1.5 work better than the old custom node implementations?

Significantly better in my testing. The native implementation is approximately 18% faster and much more stable. I experienced zero crashes with the native nodes compared to frequent crashes with previous custom node implementations.

What are the Topaz API costs for video enhancement?

Topaz API is metered separately from desktop licenses. In my testing, upscaling a 5-second 24fps video from 512x512 to 1920x1080 with enhancement cost approximately $0.42 in API credits. Heavy use can become expensive quickly.

Are there any performance benefits if I don't use the new features?

Yes, the memory management improvements and sampling speed optimizations apply to all workflows. I measured 8-19% sampling speed improvements across different samplers and approximately 1.6GB reduced VRAM usage in my standard workflows.

Conclusion

ComfyUI v0.3.75 is the most significant update in the last six months. Native support for Z-Image, HunyuanVideo 1.5, and FLUX.2 Day-0 eliminates dependence on inconsistent custom node implementations for major model architectures.

The performance improvements are substantial. Faster sampling, better memory management, and optimized node processing make complex workflows more practical on existing hardware.

Breaking changes require migration work, particularly for ControlNet workflows and video processing pipelines. Plan for 2-4 hours of testing and fixing after updating if you have complex workflows.

I recommend updating if you need any of the new features or want the performance improvements. The migration effort is worthwhile for the long-term benefits. Back up everything first, update custom nodes immediately after core update, and test incrementally rather than loading your most complex workflow first.

For users who want cutting-edge AI generation capabilities without the complexity of managing updates, breaking changes, and custom node compatibility, platforms like Apatero.com deliver professional results with zero configuration overhead and predictable pricing.

The ComfyUI ecosystem continues maturing toward production-grade reliability while maintaining the flexibility that makes it powerful for advanced users.