SDXL Power Prompt - Simple / Negative (rgthree)

This is the SDXL negative-prompt node: the two-text-stream SDXL Power Prompt with the LoRA machinery removed. It exists for the same reason its SD1.5 cousin does - negative prompts can't use LoRAs, so a node built for negatives shouldn't carry LoRA plumbing. Feed it your "what you don't want" text in SDXL's g and l streams, connect a CLIP, and you get clean negative conditioning out, no clutter.

Why you'd reach for it

SDXL prompting has two text encoders and a set of size parameters, and the full SDXL Power Prompt - Positive wires all of that up plus LoRA support. For a negative prompt, the LoRA support is dead weight - LoRAs don't apply to negatives. This node is the trimmed version: it keeps the SDXL dual-text setup and the size parameters that SDXL conditioning wants, drops the LoRA dropdown and model input, and gives you a focused negative node.

The natural pairing is SDXL Power Prompt - Positive for your positive and this one for your negative. Both understand SDXL's g/l streams and both offer the embedding dropdown so you don't mistype a filename - which, for a negative prompt full of quality embeddings, is genuinely handy.

Inputs and outputs

The two required fields are the SDXL text pair:

• prompt_g - the "global" negative text, multiline.
• prompt_l - the "local" negative text, multiline.

The optional inputs cover the SDXL machinery, minus anything LoRA: opt_clip (connect to encode to conditioning on the node), opt_clip_width / opt_clip_height (default 1024, the encoder reference size), the dropdown helpers insert_embedding and insert_saved (they paste text into your prompt), and the composition controls target_width, target_height, crop_width, crop_height (default -1, meaning "use the default" until you set them). Note there's no insert_lora and no model input - that's the whole point of the negative variant.

Outputs: CONDITIONING (populated when a CLIP is connected - wire it to your KSampler's negative input) plus the raw TEXT_G and TEXT_L strings if you want to route the text elsewhere.

Installing it

The node ships in rgthree-comfy. Install via ComfyUI Manager (search rgthree's ComfyUI Nodes) or clone the repo:

cd ComfyUI/custom_nodes
git clone https://github.com/rgthree/rgthree-comfy.git

Restart ComfyUI afterward. No models to download, no heavy dependencies - the pack is graph and UI code.

Common issues

The main thing is picking the right node for the right slot. This is the SDXL negative (or any LoRA-free SDXL text). If you want LoRA support, you need the positive SDXL node - this one has no model input and can't load LoRAs at all. And if you're on a non-SDXL checkpoint, the two-stream g/l setup is overkill; use the plain Power Prompt - Simple instead.

Second, the SDXL size parameters default to -1 deliberately - that's "use a sensible default." You don't have to set target/crop width and height to get a working negative; leave them alone unless you're specifically tuning composition, which for a negative prompt is uncommon anyway.

Third, if you connected a CLIP but nothing seems to encode, confirm the CONDITIONING output is wired into your sampler's negative slot. Without a CLIP connected the node only emits raw text, which a KSampler can't consume directly.

And the pack-wide caveat: if the node renders wrong, check whether ComfyUI's Nodes 2.0 frontend is enabled. The Vue rewrite broke several rgthree nodes and the standing workaround is to keep it off.