JK Big Context

This is the same idea as JKLilContext - bundle a pile of connections into one wire instead of running a dozen lines across your canvas - scaled up for a full SDXL-style pipeline. Where the "Lil" version carries the basics (model, clip, vae, conditioning, latent), this one adds sampler settings, a checkpoint name, dual text-encoder fields, a refiner step count, and even a control-net stack slot. It's clearly built for the base+refiner, G/L-conditioning workflows SDXL popularized, the same territory as rgthree-comfy's larger Context Big node, just this pack's own take on it.

What it carries

Every input is optional, same as JKLilContext - you fill in what you have at that point in your graph, and anything you skip just doesn't populate that slot in the bundle. Beyond the base set (base_model/model, base_clip/clip, vae, positive/negative, latent, images, seed), this node adds:

• wildcard_seed - a second seed slot, useful if you're pairing this with wildcard-prompt nodes that need their own seed separate from your sampling seed.
• steps and step_refiner - two step counts, the tell that this is meant for a two-stage base-then-refiner pipeline rather than a single pass.
• cfg, sampler, scheduler - your KSampler settings, bundled in so downstream context-flavored samplers can pull them without separate wires.
• ckpt_name - populated at runtime from your installed checkpoints.
• clip_width / clip_height - resolution conditioning, the numbers SDXL's conditioning wants to know your target size for.
• text_pos_g / text_pos_l / text_neg_g / text_neg_l - SDXL's dual-encoder text split. G and L are the two CLIP encoders SDXL runs simultaneously; a lot of SDXL-tuned workflows feed them slightly different text for better results, and this context carries both pairs.
• mask - for inpainting-flavored graphs.
• control_net_stack (typed cnet_stack) - a slot for a bundled ControlNet stack from whatever ControlNet-stacking node you're using upstream, rather than wiring individual ControlNets in one at a time.

Outputs

Every one of the fields above comes back out as its own output too - CONTEXT (the bundle itself) plus individually named passthroughs for all fourteen-plus fields, so you can tap exactly what a given downstream node needs without unpacking the whole thing by hand. SAMPLER and SCHEDULER come back out re-typed as the full ComfyUI sampler/scheduler enum lists, which matters if you're wiring into a stock KSampler node that expects those exact enum types rather than a loose string.

Installing it

Part of kostenickj/jk-comfyui-helpers - already on comfy.icu's shared image, or cd ComfyUI/custom_nodes && git clone https://github.com/kostenickj/jk-comfyui-helpers.git and restart for a self-hosted setup. Pure graph plumbing, no model downloads or extra dependencies specific to this node. It's worth repeating for a node this deep in the pack: this is a small, low-visibility personal toolkit, not something you'll find much community discussion around - the README is your main reference.

Common issues

• Overkill for a non-SDXL pipeline. If you're not running a base+refiner SDXL setup with G/L text encoding, you almost certainly want JKLilContext instead - this node's extra fields just sit empty and add clutter.
• Empty ckpt_name or sampler/scheduler dropdowns. Populated from what's actually installed; if the dropdown looks empty or short, check your checkpoints and that ComfyUI's core sampler list loaded normally.
• G/L text left blank. If you don't fill text_pos_g / text_pos_l separately, downstream SDXL conditioning nodes may end up with an empty half of the split - decide up front whether you actually need per-encoder text or can just duplicate the same string into both.
• control_net_stack type mismatch. cnet_stack is a custom type that has to come from a compatible ControlNet-stacking node upstream - a plain CONTROL_NET output won't plug in here directly.