MCWW: a replacer of Replacer!

[light-and-ray]

Announcement

I glad to present you a spiritual successor of Replacer extension for A1111 - it's a similar extension, that adds a dedicated ui, but for ComfyUI. I've chosen to switch to Comfy, because it has massive support of every new model, and it's optimized. There are new models that fit perfectly to replace this DINO+SAM+SD1.5/SDXL pipeline: Qwen Image Edit 2509, ChronoEdit, some models for video editing. Also you can add miscellaneous staff, similar to Extras tab in A1111, like upscaling, background removal

But I find ComfyUI not actually comfy. And I'm not speaking about nodes themselves - it has a lot of UI inconveniences and bugs. It works perfectly as a visual programing language for creating pipelines, but not for inference. So I decided to make this extension. There are similar existing projects - SwarmUI, ViewComfy, Flow. But I find them a bit bloated

Minimalistic Comfy Wrapper WebUI

As for Replacer you need to have working A1111, for MCWW you need to have working ComfyUI. You install it as any other extension (but I haven't added it in Comfy Registry yet, so you need to perform git clone and pip install manually), and you will see a button in a corner

Here you'll see all your workflows in familiar layout, you only need to set titles for nodes like <Load Image:prompt:1>, <Prompt:prompt:1> and <Result:output:1> to tell MCWW where ui components should appear. Read more in MCWW readme

Hardware requirements

Newer models require more VRAM, but not so many as some people think. If you have 12GB of VRAM, you can use models like Qwen Image Edit 2509 or video model Wan2.2 14b natively, in the native resolution. I recommend these weights

• wan2.2_i2v_A14b_low_noise_scaled_fp8_e4m3_lightx2v_4step.safetensors
• wan2.2_i2v_A14b_high_noise_scaled_fp8_e4m3_lightx2v_4step_comfyui_1030.safetensors
• qwen_image_edit_2509_fp8_e4m3fn_scaled.safetensors
• Qwen-Image-Edit-2509-Lightning-4steps-V1.0-fp32.safetensors

If you have less then 12GB, you can still use them but for lower resolution or duration. You can also use https://github.com/city96/ComfyUI-GGUF extension and download GGUF models. You can find people with 6GB of VRAM who managed to run this models!

Nevertheless, in ComfyUI MCWW extension you can recreate the same pipeline of DINO+SAM+SD, or use any other old models. For example, in the screenshot you can see "thispersondoesnotexist", real name StyleGAN - very old open-source model from NVidia, I think it was the first harbinger of the current AI explosion. And you can use it here, in the same minimalistic UI

Post scriptum

MCWW already have more commits then Replacer: 502 vs 423 as for today, and there will be much more. It is also much stable then Automatic1111 in general: there is an advanced queue and ui state is kept in the browser, you can reload your browser or even reboot your PC and nothing will happen