CLIP Text Encode (Prompt Enrichment)

CLIP Text Encode (Prompt Enrichment) is a drop-in replacement for the standard CLIP Text Encode node with a preset dropdown bolted on: pick Cinematic, Cyberpunk, Line Art, White Background, and sixteen others, and the node appends the preset's style vocabulary to your prompt before encoding. You get the conditioning straight out, no extra wiring. It's the "make my generic prompt look intentional" button, and for quick stylistic comparison shots it's genuinely convenient.

How it works

The mechanism is blunt and that's fine: your text is stripped, the preset's suffix is appended, and the combined string goes through the normal CLIP tokenize → encode pipeline (including the pooled output, so SDXL's conditioning routing works correctly). Nothing fancy, which means the behavior is entirely predictable. Three inputs:

• clip - your loaded CLIP/text encoder, same as the stock node.
• text - your prompt. Two useful details: it's a multiline field, and it's marked as a dynamic-prompt input, so {red|blue} hair style alternations work straight out of the box.
• preset - the dropdown. None (default) makes the node behave exactly like a vanilla encode, so you can leave it in place and just never touch the dropdown.

Output is CONDITIONING, ready for the sampler. Twenty presets ship in the pack, from Photographic and 3D Render to Macro Close-up, Vintage Film, and Monochrome/Noir.

The honest take on presets

Here's the part worth understanding before you lean on this. These presets are classic tag vocabulary - masterpiece, best quality, 8k uhd, artstation, color graded. On SDXL-lineage models (including Illustrious, NoobAI, WAI, Pony) that vocabulary is exactly what the model was trained on, and the presets will visibly steer the output. On Flux and other modern models whose text encoder is an LLM wrapped in a chat template, a lot of that vocabulary is inert - masterpiece, best quality means little to an encoder that was never trained on it, and (word:1.3) weights get discarded entirely. So: brilliant on SDXL, mostly harmless on Flux. The presets that do survive the encoder swap are the descriptive ones (Cinematic, White Background, Macro) because naming a scene or lens is the highest-leverage move on LLM-encoded models.

Installing

Part of ComfyUI-Flow-Assistor - ComfyUI Manager (search "Flow Assistor") or:

cd ComfyUI/custom_nodes
git clone https://github.com/Merserk/ComfyUI-Flow-Assistor.git

Restart after cloning. Current ComfyUI required (V3-only pack); no extra dependencies and no model files.

Where people get burned

Since it's just string concatenation, the failure modes are cosmetic: the preset has no commas before it in a few cases, so check the combined prompt isn't gluing words together (your prompt, cinematic lighting vs your promptcinematic lighting - the presets start with a leading comma, but if your own prompt ends awkwardly, trim it). And don't expect a preset to rescue a fundamentally weak prompt - it's seasoning, not the dish. On models where the vocabulary is inert you'll wonder why nothing changed; that's not a bug, that's the encoder swap doing its thing.