Nourivex Prompt Forge XL

A normal SDXL workflow front end is four nodes: a text box, a CLIP encoder for the positive, another for the negative, and an empty-latent node that decides your canvas size. Nourivex Prompt Forge XL collapses all of it into one node - prompt assembly, encoding, size selection, and latent creation - and bolts on nineteen style presets and five character "example prompts" as a bonus. It's the closest thing in this pack to a command center for an SDXL generation, and it's the node you'd reach for when you want to iterate on a character fast without shuffling four widgets.

What comes out and what it saves you

Seven outputs, which is the tell that this is several core nodes' jobs in one:

• positive and negative - finished CONDITIONING, straight to your KSampler. The node encodes them itself using CLIP's pooled output, which is the bit SDXL specifically needs; you don't run a separate CLIP Text Encode.
• latent - an empty latent already sized to your resolution and batch count, ready for the sampler.
• width and height - the resolved integers, so you can drive other size-aware nodes from the same choice.
• positive_text and negative_text - the assembled strings, for auditing exactly what got built or feeding a different encoder.

The inputs worth your attention

• character_prompt - what your subject is. Notice the default is Indonesian ("[deskripsi baju/karakter lu]", roughly "describe your character's clothes") because the author's workflows are Indonesian-language. Replace it immediately.
• positive_base / negative_base - your own base prompts. The defaults reference lazypos and lazyneg, which are quality embeddings you almost certainly don't have installed. This is the number-one gotcha: load the node, hit Run, get junk, because the defaults assume embeddings that aren't there. Swap them for your own base before doing anything else.
• style_preset - nineteen baked-in style blobs, from Photoreal Editorial (default) through Cyberpunk Neon and Manga Ink to Minimal Flat Illustration. Each one injects a pre-written positive chunk and a matching negative chunk. They're decent starting points and you'll outgrow them; treat them as seasoning, not a promise.
• example_prompt - five preset character concepts (Nuwa SciFi Goddess, Streetwear Heroine, and so on) used when character_prompt is empty. The Nuwa ones reference the author's own LoRA series, so results without those LoRAs installed will be generic.
• size_preset - five SDXL-safe sizes: 1:1 1024, 2:3 832×1216, 3:2 1216×832, 9:16 768×1344, 16:9 1344×768. These are the standard SDXL aspect ratios, all multiples of 64, which is exactly what keeps the latent happy. Flip use_custom_size and you get custom_width / custom_height (256–2048, stepping in 64s) plus batch_size (1–8).

The dialect caveat

This node is SDXL, and its whole vocabulary assumes the tag-and-weight CLIP dialect where commas, quality tags, and negatives still mean something. On the pack's Flux Klein workflow, the README deliberately does not use this node for conditioning - Flux gets its text run through CLIPTextEncodeFlux with guidance, because the prompting rules there are a different language entirely. So think of Prompt Forge XL as your SDXL front end, not a universal one.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Nourivex/Nour_Comfy

Restart ComfyUI (or search Nour_Comfy in ComfyUI Manager). No dependencies, no model downloads - just the node. It's the biggest node in the pack and the one with the most personality; you'll either love the one-stop layout or find it fiddly, but either way, swap those lazypos defaults on day one.