Prompt Adapter v2

This is the node that takes all your prompt text - positive and negative, across a few different slots - and folds it into the Searge SDXL data stream so the samplers downstream can use it. If you're wondering why Searge splits a prompt into six inputs instead of the usual positive/negative pair, that's SDXL's design showing through, and it's worth understanding before you wire this up.

SDXL runs two text encoders (OpenCLIP-ViT/bigG and CLIP-ViT/L). The second encoder is where a lot of SDXL's extra control over composition and lighting comes from. Searge exposes that by giving you separate main, secondary and style prompt lanes rather than one box, on both the positive and negative side - six lanes total. SeargePromptAdapterV2 is where those lanes get collected and packed into the Magic Box stream.

How it works

The Magic Box pipeline carries the whole workflow's state on one wire. This node is a prompting stage on that wire: it reads in your text inputs, attaches them to the data stream as the prompt set, and passes the stream along. You type your prompts elsewhere - in SeargeTextInputV2 boxes - and route them into this adapter's slots. The adapter doesn't encode anything itself; it's the collector that hands a tidy prompt bundle to the rest of the pipeline.

In practice you don't have to fill all six. Many people just use the main positive and main negative and leave the rest empty, which behaves like an ordinary prompt/negative setup. The extra lanes are there when you want to steer the two encoders or a style layer separately - a power-user feature that the workflow surfaces without forcing on you.

The inputs and outputs that matter

Every input is optional and typed SRG_PROMPT_TEXT - you feed them from Searge text nodes, not raw strings:

• main_prompt and negative_main_prompt - your primary positive and negative. If you only touch two inputs, these are the two.
• secondary_prompt / negative_secondary_prompt - the second lane, for feeding different text to SDXL's second conditioning path.
• style_prompt / negative_style_prompt - a dedicated style lane.
• data (SRG_DATA_STREAM) - the incoming Magic Box stream from the previous stage.

Outputs are data and prompts, both SRG_DATA_STREAM - the stream carried onward, plus the prompt bundle broken out so a later stage can consume it directly.

How to install it

Ships with Searge SDXL. ComfyUI Manager: search SeargeSDXL, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/SeargeDP/SeargeSDXL.git

and restart. Manual installers must run python -m pip install opencv-python once in ComfyUI's Python environment - the pack won't load without it.

Common issues

Two lanes get people. First, the inputs take SRG_PROMPT_TEXT, not plain strings, so you can't drop a generic text primitive on them - use SeargeTextInputV2 (or another Searge prompt node). Second, don't overthink the six slots. If your results feel muddled, collapse back to just main_prompt and negative_main_prompt; the secondary and style lanes are for deliberate multi-encoder steering, and filling them with near-duplicate text muddies the conditioning more than it helps. As always with this pack, if the node throws after a ComfyUI update, update Searge before you debug - the v4.3.x releases were pure compatibility fixes, and version drift is the pack's most common failure.