Guess you are all familiar with @Lodestone's Chroma models based on FLUX.1-schnell: https://civarchive.com/models/1330309/chroma

Chroma1-HD works in latent space (compressed representation) like traditional AI image generators.

But this is about: Chroma1-Radiance - a radically tuned version that works directly in pixel space, skipping the latent-to-image decode step entirely.

The main advantages of Radiance:

• Sharper output with more accurate colors

• Reduced artifacts and distortion

• Eliminates color artifacts and blurring that can occur when decoding from latent space

In simple terms: HD uses the traditional compressed method (latent space → decode → image), while Radiance generates images directly in pixel space, resulting in higher quality with fewer artifacts.

You'll need a diffusion model file. I did use this one: https://huggingface.co/lodestones/Chroma1-Radiance/blob/main/latest_x0_full_20M_dataset_run_1024.safetensors

And a text encoder. I've been using the same as for Chroma-HD: https://huggingface.co/silveroxides/flan-t5-xxl-encoder-only/resolve/main/flan-t5-xxl-fp16.safetensors

No need for a VAE.

Just pull the showcase images into Comfy. Workflow is embedded.