rocca

/

chroma-nunchaku-test

like 26

I think in latest versions of Nunchaku you need to use the 'nunchaku merge node' (it's called something like that) to merge the folder (which contains multiple safetensors, plus config) into a single safetensors file. I could be wrong though.

I think in latest versions of Nunchaku you need to use the 'nunchaku merge node' (it's called something like that) to merge the folder (which contains multiple safetensors, plus config) into a single safetensors file. I could be wrong though.

I was just looking for the same info. The merge node is there but for some reason I got the error "module 'torch' has no attribute 'float8_e8m0fnu'" when trying to run it. Would it be possible to you to also post the merged file alongside the directory?

I'm using this and it works fine:

And the quality is pretty good:

And the quality is pretty good:

wow looks very good. can you share your prompt?

Thanks
1 and 2:

Perspective: A sweeping aerial shot from dragonback (partial dragon wing visible), with the citadel dominating the horizon. • Foreground: Wisps of cloud curling around the viewer’s vantage point, maybe the silhouette of a wing or a wind-swept cloak. • Midground: The citadel itself—spires, bridges, and floating gardens, all glowing faintly with runes. • Background: Distant peaks, a setting sun casting golden light, and a storm brewing below. • Color Palette: Misty whites, emerald greens, and soft golds with ethereal glows. • Lighting: Diffused sunlight through clouds, with magical glows from the citadel’s core structures. • Texture: Ethereal softness of clouds contrasted with crystalline architecture and ancient stone. High-resolution, 8k, hyper-realistic, extremely high contrast, cinematic lighting, detailed textures, award-winning photography, sharp focus, crisp detail, vibrant colors, ultra-HDR photography, photorealism, (breathtaking, analog photo, Pentax 645Z Digital). High-resolution, 8k, hyper-realistic, extremely high contrast, cinematic lighting, detailed textures, award-winning photography, sharp focus, crisp detail, vibrant colors, ultra-HDR photography, photorealism, (aesthetic 10, breathtaking, analog photo, Pentax 645Z Digital).

3:
panoramic view of a pristine alpine lake at sunrise, perfectly reflecting snow-capped mountain peaks, vibrant wildflowers in the foreground, mist rising off the crystal-clear water, photorealistic, high dynamic range, peaceful atmosphere, Ansel Adams style, wide aspect ratio High-resolution, 8k, hyper-realistic, extremely high contrast, cinematic lighting, detailed textures, award-winning photography, sharp focus, crisp detail, vibrant colors, ultra-HDR photography, photorealism, (aesthetic 10, breathtaking, analog photo, Pentax 645Z Digital).

neg:
cartoon style, anime, low resolution, poor anatomy, extra limbs, abnormal proportions, bland lighting, generic background, modern clothing, text, watermark, logo, symmetry, smiling faces, dull colors, cluttered composition, flat shading

I have the same issue as the original poster, how we are supposed to load two files when you can only load one with Nunchaku Flux Dit Loader?

I have the same issue as the original poster, how we are supposed to load two files when you can only load one with Nunchaku Flux Dit Loader?

You load the path/folder where they are stored, Nunchaku Flux Dit Loader supports that. See my screenshot where it says model_path.

BTW:
The latest version of ComfyUI-nunchaku that supports folder loading of models is v0.3.4, as indicated by the GitHub releases page for ComfyUI-nunchaku. This version is the most recent and would include support for loading models from directories.
It is important to ensure you are also using the corresponding Nunchaku wheel (v0.3.1 or newer as suggested in some discussions) for proper functionality with the ComfyUI-nunchaku plugin.

I already tried using the folder, and it doesn't work. It's like it doesn't exist. Error no file named config.json found in directory E:\AI\ComfyUI-Nunchaku\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\Chroma_Nunchaku. (and yeah, I have 0.3.1, I can run nunchaku Koncet and Krea without issues).

Not sure what's going on then. I'm using Comfy_Nunchaku 0.3.3 and it loads the folder just fine. Alternatively you can use the merger workflow from https://github.com/nunchaku-tech/ComfyUI-nunchaku/blob/main/example_workflows/merge_safetensors.json to merge the files from the folder into a .safetensors