🐰 McBoaty Configurator - v6 /m

If the McBoaty v6 set is a tiled upscaling assembly line, the Configurator is where you set the speed and the fuel: it's the "Upscaler" half, responsible for taking the tiles from a Tiler node and stamping every one of them with the sampling settings and prompt they'll be refined with. The node's own description just calls it "An UPSCALER Node," which undersells it - it does no upscaling itself. It configures.

The flow

The v6 pipeline splits what v5 crammed into three nodes:

Tiler → Configurator → TilePrompter → Refiner

The Configurator takes the MC_BOATY_PIPE from the Tiler (that pipe carries your sliced tiles plus all the geometry - tile size, overlap, grid - needed to rebuild the image later), plus your model, clip, vae, and the prompts as plain strings (note: v5 took conditioning; v6 switched to text you type or wire in). It then writes the sampler settings and prompt onto every tile in the pipe and hands you two outputs:

• McBoaty Pipe - on to the Refiner.
• McPrompty Pipe - a separate pipe carrying just the per-tile prompt data, on to a TilePrompter if you want to edit prompts per tile by hand.
• info - a string with run info.

The inputs that actually matter

The full sampler block is here - sigmas_type, model_type (SD1/SDXL/SD3/FLUX1/SVD), sampler_name, basic_scheduler, steps, seed, cfg, denoise. Defaults are the McBoaty recipe: 10 steps, CFG 2.5, denoise 0.27. The README's guidance holds: 0.27–0.35 denoise is the "minimal modification for printing" zone; go higher and tiles start drifting.

Then the three that trip people up:

• vae_encode (default on) - "tiled" vs standard VAE encode. Keep it on; tile_size_vae (512) governs the tile size for the encode, and it's how McBoaty keeps 4K+ images from OOMing a mid-range card.
• strength / start_percent / end_percent - ControlNet strength applied to every tile (default 0.76, active 0–76% of steps). This is the tile-anchoring that stops per-tile divergence, and the README calls 0.76 the max denoise you'd want with control net active.
• The Ollama block - ollama_url, vlm_model, ollama_keep_alive, vlm_query. If your positive prompt is empty and you have a local Ollama instance with a vision model (the dropdown defaults to llama3.2-vision), the Configurator will caption the whole image, then caption each tile individually, and use those as the per-tile prompts. That's the "assisted prompt generation via LLM" feature from the changelog. Leave the prompt empty with no Ollama running and you'll get an empty prompt, so don't set it and forget it.

Wiring and install

The optional tiles IMAGE input lets you override which tiles get used, and strength/percent live in optional too - meaning you can leave ControlNet completely unconfigured if you set them nowhere. Everything else is required.

Install is the pack once:

cd ComfyUI/custom_nodes
git clone https://github.com/MaraScott/ComfyUI_MaraScott_Nodes

Restart or use ComfyUI Manager → "MaraScott". The Configurator's Ollama features need Ollama installed and running separately; the Florence-2 captioning path downloads the model on first run. And remember the pack-level caveat: the README flags v6 as broken (issue #187), so if the Configurator misbehaves, the v5 Upscaler is the proven alternative - it does the same "stamp every tile with settings" job with conditioning inputs instead of strings.