CLIPTextEncodeWithWeight (Inspire)

This is a CLIP Text Encode with two extra knobs bolted on: a strength multiplier and an add_weight offset applied to the encoded conditioning. It does the same core job as the standard node - turn a text prompt into CONDITIONING for the sampler - but gives you a coarse, whole-prompt intensity control that sits outside the usual (token:weight) prompt syntax.

When's that handy? When you want to scale a whole prompt's influence up or down as one unit, without editing the text or wrapping everything in weight parentheses. It's a blunt instrument compared to per-token weighting, but blunt is sometimes exactly right - for A/B testing how strongly a prompt bites, or for pushing a conditioning harder in a multi-encode setup where you're balancing several text inputs against each other.

How it works

It encodes your text through the given clip model, then post-processes the resulting embedding: strength multiplies it, and add_weight adds a constant bias. Multiplying scales the prompt's overall pull; the additive term shifts it. The result comes out as ordinary CONDITIONING, indistinguishable to the sampler from any other conditioning - the difference is purely in how you shaped it on the way out.

Because this scales the encoded vector directly, it's a different mechanism from prompt-internal weights like (masterpiece:1.3), which reweight individual tokens before encoding. This reweights the finished embedding as a whole. You can use both together.

The inputs and outputs that matter

• text (multiline STRING) - the prompt.
• clip (CLIP) - the text encoder from your checkpoint or CLIP loader.
• strength (FLOAT, default 1.0, range 0–10) - overall multiplier on the encoded conditioning. 1.0 is neutral; below 1 softens the whole prompt, above 1 pushes it harder.
• add_weight (FLOAT, default 0, range -10 to 10) - an additive offset on the encoding. 0 is neutral; small non-zero values nudge the conditioning.

The single output is CONDITIONING - feed it into your sampler's positive or negative input like any encode.

How to install it

ComfyUI Manager: search ComfyUI Inspire Pack, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Inspire-Pack

then restart. It's a Dr.Lt.Data pack (ComfyUI-Manager, Impact Pack).

Common issues & troubleshooting

High strength fries the image. Overscaling a conditioning behaves a lot like cranking CFG - crunchy contrast, blown colors, mangled output. Back strength off toward 1.0. This is a fine-tuning dial, not a "more is better" one.

add_weight seems drastic at small values. It's an additive bias on the raw embedding, so even small numbers move things noticeably and it can distort the prompt in odd ways. Leave it at 0 unless you specifically know what you're chasing, and change it in small steps.

No difference at strength 1.0 / add_weight 0. That's the neutral setting - at those values you've essentially got a plain CLIP Text Encode. The extra knobs only matter when you move them off neutral.

Wrong CLIP for the model. The clip has to match your base model (SD1.5 vs SDXL vs newer). A mismatched encoder produces conditioning the model can't use properly, regardless of the weight settings. Pull clip from the same checkpoint you're sampling with.