Ino Get Conditioning

Every image workflow funnels down to the same step: turn text into conditioning tensors the sampler can use. Ino Get Conditioning is that step, packed into one node with a Flux twist. It encodes positive and negative conditioning from text, and it knows about the Flux way of doing things - the dual-encoder positive prompt, the separate guidance value, and the fact that on guidance-distilled models the "negative prompt" is usually better off zeroed out than actually encoded.

It's one of the InoSamplerHelper nodes in ComfyUI-InoNodes, and it's clearly built around the pack's config-driven Flux pipelines (the default base model throughout the pack is flux1dev).

The inputs

• clip - the CLIP model to encode with. Required.
• positive1 - the first positive prompt.
• positive2 - the second one. For Flux this is the T5-XXL side of the dual-encoder call; for plain SD-family models it's unused.
• negative - the negative prompt.
• config - an optional model config JSON that can supply the flags below (like the pack's model configs, whose defaults are Flux-flavored).

The toggles - all optional combos that can also come from config:

• use_flux_encoder - use the Flux dual-encoder path (CLIPTextEncodeFlux) instead of plain CLIPTextEncode.
• use_flux_guidance - apply a Flux guidance value on top of the encoded positive.
• guidance - the guidance number. Default −1, which the config treats as "not set."
• use_negative_prompt - actually encode negative into real negative conditioning. Off by default, which is the right call for Flux-style guidance-distilled models.

Outputs: positive and negative (both CONDITIONING), plus old_config and new_config - the config as it went in and as it came out after the flags were merged.

How it works

It merges the toggles into the model config, then branches: with the Flux encoder on, it runs the Flux dual-encoder call (clip_l from positive1, t5xxl from positive2, plus the guidance value). With Flux guidance on, it applies the Flux guidance node. For the negative side, if use_negative_prompt is on it encodes your negative text; otherwise it zeroes out the positive conditioning (ConditioningZeroOut) - which is the standard, and usually correct, behavior for guidance-distilled models where a literal negative prompt does little. Both positive and negative wire straight into a KSampler.

Where it fits

This replaces the classic "two CLIP Text Encode nodes plus a flux guidance node" trio with one node whose behavior is driven by config - so your sampler pipeline can switch between a Flux stack and a plain SD stack by changing the config, not rewiring the graph. That's the pack's whole philosophy in one node.

Gotchas

The Flux specifics will bite you if you ignore them: with a non-Flux CLIP, leave use_flux_encoder off or the dual-encoder call will fail. And positive2 being empty on the Flux path produces an empty T5 prompt - keep both halves filled for Flux. The guidance default of −1 is a sentinel ("use the config's value"); if neither a config nor a real guidance number is present, downstream may see −1 and behave oddly. Also, a disabled node returns None for both conditioning outputs - wire that into a sampler and you'll get a hard-to-read failure, so keep enabled on.

Installation

Part of ComfyUI-InoNodes:

• ComfyUI Manager: search "ComfyUI Ino Nodes", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/nobandegani/comfyui_ino_nodes && cd comfyui_ino_nodes && pip install -r requirements.txt, restart.

Python 3.10+, ComfyUI 0.18.1+ (V3 schema). No models, no keys - it encodes with whatever CLIP you wire in.