Usage

Usage — Nodes & Pipelines

Radiance exposes modular "nodes" for constructing image generation pipelines (similar to ComfyUI/custom-node flows). This page documents common nodes, parameters, and example pipeline patterns.

Node categories

• Data nodes — load images, masks, and reference assets
• Model nodes — load checkpoints / diffusion / denoiser models
• Scheduler nodes — control sampling, timesteps, and schedulers
• Color nodes — high-precision color transforms and conversions (linear / display-referred)
• Post-processing — tonemapping, LUTs, ICC profile conversions

Common node parameters

• seed — integer seed for deterministic outputs
• steps — number of diffusion steps
• guidance_scale — classifier-free guidance scale
• precision — fp16 or fp32 (fp32 recommended for color-critical workflows)

Example pipeline (conceptual)

1. ImageLoader -> ModelEncoder
2. ConditioningNode (text/clip/embedding)
3. DiffusionSampler (scheduler + steps + guidance)
4. HighPrecisionColorTransform (linear->display)
5. Tonemap -> SaveImage

Saving with ICC profiles

• To preserve color fidelity, save intermediate outputs as 16-bit or floating-point formats and apply ICC transforms with the Color node before converting to sRGB for display.

Notes on nodes

• Use fp32 for color-critical transforms. fp16 is fine for exploratory runs but may introduce banding.
• When chaining many transforms, keep intermediate results in floating point formats to minimize quantization artifacts.

Reference

• See the repository's docs/ or the nodes/ directory for a full list of implemented nodes and examples.