Hypnodes Control Hub

The Control Hub is where every Hypnodes workflow starts. It's the settings panel, the seed holder, the resolution picker, and the global prompt box all rolled into one node - and the entire point is that you only ever set these things in one place. Change the seed here and every Hypnodes node downstream updates with it, instead of you hunting across the graph for the five samplers that each secretly have their own seed widget.

Here's the trick that makes it work: the Hub doesn't send values down a dozen wires. It packs everything into a single hn_config bus (an opaque HN_CONFIG type) that your other Hypnodes nodes read from. The README calls it a "State Manager," which is fair - this is a settings dictionary in node form. A Stacker watches it to know how many character boxes to show, the Mask Generator reads character_count and layout_mode from it, and the Conditioners pull the global prompts out of it.

The inputs that actually matter

You'll set most of these once and forget them:

• ckpt_name - pulls the list from your models/checkpoints folder, so it doubles as your model loader.
• seed, steps, cfg - standard fare, piped straight into the sampler later. cfg defaults to 7, which is the classic SD 1.5/SDXL zone, though it's worth knowing distilled and Turbo checkpoints want something closer to 1.
• character_count (1–10) - the master dial. Bump this and the Stacker shows more boxes and the Mask Generator carves more regions.
• layout_mode - Horizontal, Vertical, or Free. Be careful here, because the naming is easy to trip on: Vertical means side-by-side columns (left to right), and Horizontal means stacked rows (top to bottom). The README pushes Vertical for group shots where characters stand together, and Horizontal for creative stacking - a character flying above another, lying on the ground, standing on shoulders.
• resolution - 26 preset aspect ratios from 704x1408 to 1728x576. No freeform box; you pick the preset.
• mask_blur (0–64, default 6) - controls how soft the region boundaries are. This is the "sweet spot" dial. Too low and you get visible cut-and-paste seams between characters; too high and color bleeds - one character's hair starts showing up on the other's face. The author recommends starting at 6–8, which lines up with the general community guidance of roughly 4–12px of mask blur for mask-based work.
• global_positive_prompt / global_negative_prompt - the two text fields at the bottom, and honestly the most important inputs in the pack. The positive should carry the big picture: character count (2girls), environment (inside cafe), lighting, quality tags. The negative carries the universal hates (lowres, bad anatomy, text, watermark).

Outputs

• hn_config - the bus. Feed this to the Stacker, Mask Generator, Conditioner, Prompt Encoder, or Converter.
• char_count - the same character_count as a plain INT, for wiring into nodes that want a number rather than the bus.

One subtle thing to know

The global positive prompt doesn't just set the scene - the Conditioner prepends it to every character's regional prompt. That's deliberate (global context plus regional detail), but it means you should keep scene description out of the individual character boxes to avoid "token noise," as the author puts it.

Install

ComfyUI Manager (search Hypnodes or ComfyUI-Hypnodes) is the easy path, or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/hypnichorse/ComfyUI-Hypnodes

Restart ComfyUI. There are no model downloads and no heavy dependencies - the pack's pyproject only lists numpy and pillow, both of which ComfyUI already ships. The pack is young (one commit at time of writing), so expect occasional rough edges and keep the README's Ko-fi link in mind - this is a small solo project.