🧩 SF FurySampler

The Director Engine renders every entity on the bus in one go. FurySampler is the opposite instinct: it renders one entity, by ID, so you can iterate on a single character before you commit to the full cast. If you're coming from a plain ComfyUI KSampler habit, this node is a KSampler that's been given bus access and a memory - it grabs the pre-encoded conditioning that Add Entity already baked, samples, decodes, and then writes the result back onto the bus as the current_render.

That last part is what makes it a "smart" sampler rather than a rebranded KSampler. After a run, bus["current_render"] holds the image, the latent, and the entity name - which is exactly what Smart Saver looks for when you want to persist the result. So the loop you'll actually use is: Add Entity → FurySampler → tweak → Smart Saver. Iterate on the prompt and settings until Hero looks right, then flip over to the batch engine for the whole set.

The inputs that matter:

• target_entity_id - the entity_id string you gave the character in Add Entity. This is the whole point of the node: you're selecting one passenger off the bus. Misspell it and the node politely returns a black image and a warning instead of crashing.
• resolution - five fixed options: 512x512, 768x768, 1024x1024, 1280x720, 1920x1080. Note this is a square-ish preset list, and it ignores whatever aspect ratio you set on the entity. Want a 21:9 hero render? Not here - that's Director Engine territory.
• steps, cfg, denoise, seed - ordinary sampler controls. The sampler_name list is short and sane (euler, dpmpp_2m, dpmpp_sde, lms) and scheduler has just normal/karras/simple. Enough to get good results, not enough to rabbit-hole.
• model - your checkpoint. Unlike the bus-injected stuff, the model is still wired in directly. VAE comes from the bus, model doesn't.

The outputs: three. bus_result (SF_LINK, now carrying current_render), image (the decoded IMAGE), and latent (the LATENT samples). Wire image to a preview, bus_result into Smart Saver or Generative Fusion.

Install

Same pack, same path: ComfyUI Manager → ComfyUI-Studio-Fury → Install → restart, or:

cd ComfyUI/custom_nodes/
git clone https://github.com/FuryNocturn/ComfyUI-Studio-Fury

Dependencies are torch/Pillow/numpy/aiohttp - nothing extra to install, no models to download.

Gotchas

The VAE has to be in the bus, same as CLIP - feed it to the Project Manager or this node raises. Beyond that, the main trap is expecting this node to respect entity aspect ratios when it doesn't; use it as a character-development sandbox and it's great. One more quirk worth knowing if you ever read the source: the node's registration block is written inside the generate method (after the return, so it's dead code), and it only shows up in your node list because StudioFury's __init__.py falls back to auto-detecting any class with INPUT_TYPES and RETURN_TYPES. It works - the node appears as "🧩 SF FurySampler" - but it's a sign this pack's internals are a little hand-rolled. Fine for a small project suite, just don't expect upstream polish.