Reponse and HTTP 400 Error

[marche-obscure]

Two questions nobody has answered yet, and both would help a lot:

1. Does the one-click .ccx install work on a clean machine?
Yes — I tested it on a clean machine and it worked perfectly. The one-click .ccx install completed without any issues.

2. Does any of this work on macOS?
I can't speak to that, unfortunately — I only have a Windows 11 PC

That said, I have another issue.

I completely understand the limitation on workflows, especially for an ALPHA version.

However, I have a ComfyUI installation via ComfyUI-Easy-Install.

Here's what my "models" folder looks like:

I mainly work in Krea2.
I configure each field as follows:

When I launch the generation, I get the following error:

Did I forget something?

[MehranMarxian]

Thanks for the detailed report
and Yes, I think I see what may be happening with the Krea-2 setup.

OpenLayer's current Krea-2 workflow expects the model files in ComfyUI's separate model folders, rather than under models/checkpoints/Turbo/...:

models/
├── diffusion_models/
│   └── krea2_turbo_fp8_scaled.safetensors
├── text_encoders/
│   └── qwen3vl_4b_fp8_scaled.safetensors
└── vae/
    └── qwen_image_vae.safetensors

The Krea-2 workflow specifically uses UNETLoader for the diffusion model, CLIPLoader with type: "krea2" for the Qwen3-VL text encoder, and VAELoader for the Qwen Image VAE. So if the Krea-2 model is currently in models/checkpoints, ComfyUI won't find it for this workflow. for example my path is comfyui.git\app\models\diffusion_models\krea2_turbo_fp8_scaled.safetensors

I'd also recommend:

1. checking the Setup section first, check if Krea2 three files are labeled installed, that way you can see if the model is in the right path (especially diffusion_models, text_encoders, and vae)

2. making sure your ComfyUI installation is fully up to date, because Krea-2 support requires the newer ComfyUI model/node stack.

If the files are already in the folders above and you're still getting the error, plase send me the ComfyUI console output immediately after the 400. That should tell us exactly which node/input ComfyUI is rejecting.

And thanks for confirming the .ccx installation working on a clean Windows machine, that's very useful testing information.

Feel free to notify me.

[marche-obscure]

OMG!
Thank you so much for your quick response and help with this problem.
I've correctly placed all the Models/Clip/Vae files in their respective folders without putting them in subfolders.
And the result is as follows:
[INFO] Prompt executed in 6.91 seconds

Well, it's a simple prompt, but it confirms that it works perfectly.
Thank you so much for this .ccx file; we'll be able to do without Firefly, which is costing us an arm and a leg.

[MehranMarxian]

@marche-obscure, following up, and thanks again for testing the .ccx on a clean Windows 11 machine. Nobody had answered that question until you did.

Also a heads-up, since you're one of the few people actually running this: v0.35 is close, and it adds Qwen-Image 2.1 to Text to Image, Image to Image and Multi-Reference. You get transparent-background layers straight from the model, lettering that comes out spelled right, and instruction editing. One catch: its weights are under a research/non-commercial licence, so those presets are opt-in and labelled that way. I'll post here when it's out, and a second machine's results would be really useful.