I started using it back when it was only on GitHub, and even then it was a polished frontend. Since then, everything has continued to improve, but the addition of the API and MCP features—and now Agent being in alpha—has really opened up the creative process for me. It’s fast, and having both cloud and local running has only made it easier to produce things quickly.

Pricing feels fair, and being able to run locally on a 5080 using flux2 klein 9b is great for prototyping, saving token costs, and making your credits count when it’s time for production runs. Onboarding is also very simple: signup and installation are incredibly straightforward.

One thing that has always stood out to me is that this feels made with care and passion—by people who understand the power of creativity and the value of democratizing it. It’s allowed me to create images I never thought I could produce. It’s my go-to software for image creation, and with the number of models and providers available, you can really find something for all pockets. I’m a solo user and an enthusiast, and I can’t imagine wanting to use anything else. Review collected by and hosted on G2.com.

Sometimes I notice a bit of slowdown when things get crowded. I’d also like to see a guided learning path, and an iOS app similar to the LmLink app in lmstudio. Review collected by and hosted on G2.com.

ComfyUI has completely transformed the way I create visual audio content. The node-based workflow gives you an incredible amount of control and flexibility — you can build, customize, and fine-tune your pipelines exactly the way you want instead of being locked into a rigid template. Combined with Comfy Cloud, I can run everything without worrying about hardware limitations, which makes the whole experience smooth and accessible. The community is also fantastic, with tons of shared workflows and custom nodes that keep expanding what's possible. Review collected by and hosted on G2.com.

There is a learning curve, especially if you're new to node-based interfaces. It can feel overwhelming at first, but once it clicks, it becomes incredibly powerful. Better onboarding tutorials or templates for beginners would make it even more welcoming. Review collected by and hosted on G2.com.

Honestly? I totally fell right into the one thing people seem to hate, the visual linking system. I've never wanted to code, but I both like, and need, to create, and Comfy let's me accomplish things that were either science-fiction or in the realms of professional studio costs. Neither of which were ... options. :) Review collected by and hosted on G2.com.

So far, I'm still learning, so not much. I am disappointed to not be able to get some of the nodes that I've been reading about, but I understand why this is tricky. Review collected by and hosted on G2.com.

What I like best is the node-based workflow. It gives me total control to build custom image and video pipelines, run models locally for free, and iterate fast. It handles the newest models like Z-Image Turbo and WAN right away, and the flexibility beats any locked-down app. Review collected by and hosted on G2.com.

What I dislike is the credit cost on the cloud side. It can get expensive fast, and the extra costs get hidden inside the subgraph so they aren't transparent. Support takes too long to get back to you. The workflow templates are often out of date, and when a workflow breaks mid-run you still burn the credits with no refund. Those are the real problems. Review collected by and hosted on G2.com.

Its versatility and multiple options make it a very useful tool for design and developing creative ideas. Additionally, I can choose and test different AI models according to what I need. Review collected by and hosted on G2.com.

In some models, credit consumption can be high. However, the flexibility to adjust credit usage—by modifying size and quality settings, choosing different models, and combining them with various nodes—still provides me with a reasonable degree of control.

Another issue that needs to be resolved is that some images from the 'load image' node cannot be deleted; a way to control this is necessary. Review collected by and hosted on G2.com.

I like that ComfyUI just works most of the time. I also find the features like Lip Sync, Image to Video, and Lora Training to be wonderful. Review collected by and hosted on G2.com.

The CLI Comfy Cloud token usage is an issue for me because it doesn't match my account, preventing me from reloading often. It would be useful to have a token reset button that can clear the cache or resync without needing to reach out for assistance. Also, the initial setup was a bit confusing, especially when I dealt with AMD and over-installed the CLI, but other than that, it turned out to be straightforward and easy. Review collected by and hosted on G2.com.

I truly believe that ComfyUI is highly important because it is open-source, powerful, covers a massive ecosystem, and has a great community. It utilizes hardware remarkably well, and if you don't have the required specs, you can use the web version. Additionally, ComfyUI is always the first interface to update whenever a new model hits the market. And as i have said i also like to have the hability to use my GPU and if i dont want to use the Web version. Pricing is ok but i think you shouldnt rise your prices. I personaly like the UI but maybe a non programmer or other person would find it challenging. Integrations are as i have said one of the best things, every time i find out about a great model it is in ComfyUI. The AI used by models is the best in the market and the Support i dont know how it is i have never asked for support, but i asume it is ok. Review collected by and hosted on G2.com.

If i had to say something i dislike about ComfyUI is that in the Workflows you download or add sometimes is very slow or difficult to install required requisites or packages, besides that everything is great. Review collected by and hosted on G2.com.

ComfyUI's versatility is its most significant strength. While it may not be beginner-friendly right out of the box, the platform offers excellent documentation and a welcoming environment for newcomers, making it surprisingly accessible compared to other complex software. The open-source nature is a major advantage, especially when it comes to model access, allowing users to leverage free resources even if budget constraints are a concern. ComfyUI's forward-thinking approach, with its focus on integrating new models quickly and providing options for custom node creation, demonstrates a strong commitment to future development.

The platform's integrations are top-notch, making workflow building much smoother. While there have been some minor performance issues with Node 2.0, the overall user experience is solid, and these problems are being addressed. For those comfortable setting up the necessary environment (like PyTorch), ComfyUI offers unparalleled flexibility. It’s a powerful tool that rewards patience, but it’s also incredibly well-documented and accessible for newcomers.

ComfyUI's commitment to open source is a huge win, as it allows users to access free models, keeping their budget intact. The platform remains at the forefront of innovation, with new models always prioritized in updates, and the ability to create custom nodes for specific needs shows great foresight. Their integrations are excellent, making connecting different tools seamless.

While there were some minor hiccups with Node 2.0 speed bumps, these issues are being addressed quickly. The platform's power lies in its flexibility and community support. If you're looking for a tool that lets you build exactly what you need, ComfyUI is the place to start—even if you have to learn the basics first. Review collected by and hosted on G2.com.

The only complaint I have ever had is beyond control of ComfyUI, I have an Intel GPU and Nvidia dominates the space, torch-xpu, on Linux and Windows both, has been a royal pain. Anything that can be done to stop the Python circus behind the scene would be greatly appreciated! Review collected by and hosted on G2.com.

What I like most about ComfyUI is the level of control and flexibility it provides over the entire AI generation workflow. The node-based UI has a steep learning curve, but once you understand it, it becomes extremely powerful, transparent and easy to customize.

Performance is another major advantage, especially for users running AI locally on limited hardware. I can optimize workflows, manage VRAM usage and choose models and settings according to my hardware instead of being locked into a fixed cloud pipeline.

The integrations and community ecosystem are excellent. Custom nodes, models and workflows make it possible to connect different tools and build highly specialized pipelines for image and video generation.

From a pricing and ROI perspective, running ComfyUI locally is one of its biggest strengths. There are no mandatory per-generation costs, and the ability to reuse and optimize workflows can significantly reduce the cost of experimentation and production.

The main weakness is onboarding. ComfyUI can be intimidating for beginners, and understanding nodes, dependencies, models, VRAM requirements and workflow errors takes time. Documentation and the community help a lot, but the initial learning curve could be improved.

In terms of AI capabilities, I appreciate that ComfyUI does not hide the process behind a simple prompt box. It gives me direct control over models, conditioning, sampling, upscaling and other stages of the generation pipeline. This makes experimentation much more precise and repeatable. Review collected by and hosted on G2.com.

The biggest downside of ComfyUI is the steep learning curve and the amount of technical knowledge required to use it effectively. The node-based interface is extremely powerful, but it can become difficult to manage when workflows grow large and complex. Finding the right custom nodes, installing dependencies and diagnosing errors can also be frustrating, especially for new users.

Performance can be highly dependent on hardware and workflow configuration. On systems with limited VRAM, even relatively simple workflows may require significant optimization, and understanding VRAM management is not always intuitive.

The ecosystem is both a strength and a weakness: there are many custom nodes and models available, but compatibility issues, outdated nodes and dependency conflicts can make workflows break after updates. Better dependency management and clearer compatibility information would help significantly.

The local approach provides excellent ROI, but the initial setup can involve considerable time and troubleshooting. Better onboarding, more guided workflows and clearer error messages would make ComfyUI much more accessible without sacrificing its advanced capabilities.

Finally, ComfyUI itself does not always feel like a polished AI application for beginners. It exposes a huge amount of technical control, but users have to understand how the underlying AI pipeline works. I would like to see more intelligent assistance that explains errors, recommends compatible nodes or models, and helps optimize workflows while still allowing experienced users to retain full control. Review collected by and hosted on G2.com.

The nodes system is so powerful, it allows to create very powerful workflows and provide better control over your generations than other platforms. UI/UX is awesome, it has all the models for any task, it is fast, reasonable pricing but even if you dont wanna pay the local workflows are very powerful. I have connected all my tools with a single AI agent, comfy ui plays very nicely and i had no issue so far. Review collected by and hosted on G2.com.

Would be great if it had in-app tutorials for the basic stuff so someone can learn how to build workflows. I know documentation already exists and is a good one, but an in-app section with tutorials would be great for a smooth onboarding of new users. Review collected by and hosted on G2.com.