ComfyUI-Detail-Daemon

DetailDaemon-example

Video Demo

Here is a video demo showing Detail Daemon in action, provided by MuseMachine:

Watch the video

ComfyUI-Detail-Daemon

A port of muerrilla's sd-webui-Detail-Daemon as nodes for ComfyUI, to adjust sigmas that generally enhance details and possibly remove unwanted bokeh or background blurring, particularly with Flux models (but also works with SDXL, SD1.5, Qwen, Z-Image, and likely other models). If the values are taken too far it results in an oversharpened and/or HDR effect. There are five nodes here, including an interactive sampler GUI. Multiply Sigmas and Lying Sigma Sampler are also included as alternative methods of generally enhancing details.

• Detail Daemon Sampler GUI (sampler and interactive schedule graph in one node)
• Detail Daemon Sampler
• Detail Daemon Graph Sigmas (to graph the sigmas adjustment visually)
• Multiply Sigmas (stateless)
• Lying Sigma Sampler

Note that Detail Daemon and Lying Sigma Sampler nodes work by default with custom sampler nodes such as SamplerCustomAdvanced. If you want to use them with non-custom sampler nodes such as KSampler or KSamplerAdvanced, then you'll need to make a custom sampler preset using the BlehSetSamplerPreset node, so you can select the preset from the list in the sampler node, as discussed here.

Example and testing workflows are also available.

Nodes

Detail Daemon Sampler

Screenshot 2024-10-29 124741

Allows sampling with the Detail Daemon schedule adjustment, which keeps the noise levels injected the same while lowering the amount of noise removed at each step, which effectively adds detail. Detail_amounts between 0 and 1.0 work best. See muerrilla's Detail Daemon repo for full explanation of inputs and methodology. Generally speaking, large features are established in earlier steps and small details take shape in later steps. So adjusting the amount in earlier steps will affect bigger shapes, and adjusting it in later steps will influence smaller fine details. The default adjusts mostly in the middle steps.

Parameters (the graphing node below can help visualize these parameters):

• detail_amount: the main value that adjusts the detail in the middle of the generation process. Positive values lower the sigmas, reducing noise removed at each step, which increases detail. For Flux or Z-Image models, you'll probably want between 0.1–1.0 range, or higher. For SDXL models, probably less than 0.25. You can also use negative values if you want to decrease detail or simplify the image.
• start: when do you want the adjustment to start, in a percent range from 0–1.0, 0 being the first step, 1.0 being the last step. Recommended: 0.1–0.5
• end: when do you want the adjustment to end, in a percent range from 0–1.0, 0 being the first step, 1.0 being the last step. Recommended: 0.5–0.9
• bias: shifts the detail_amount in the middle steps forward or back in the generation process.
• exponent: changes the curvature of the adjustment. 0 is no curvature, 1 is smoothly curved.
• start_offset: start the detail_amount at a particular value at the beginning of the generation process. Not recommended.
• end_offset: end the detail_amount at a particular value at