👉 2. Artist Prompt Station (Template)

ArtistPromptComposer is the second node in the Artist-Tester pack, and it's the one that makes the iterator's output actually useful. It takes your base prompt, an artist tag, and a template, then does all the CLIP encoding for you - positive and negative - and hands you conditioning ready for the KSampler. In a stock graph this replaces a CLIP Text Encode (Prompt) node plus the string-concat dance you'd otherwise do by hand.

What it actually does

Wire in a clip, a positive_base, and an artist_tag, and it builds a prompt like {base}, {artist}, which by default comes out as "masterpiece, best quality, 1girl, solo, portrait, artist:wlop". Then it tokenizes and encodes that with ComfyUI's standard clip.tokenize / encode_from_tokens path - the same thing the built-in text-encode node does under the hood - encodes the negative_prompt separately, and returns both conditionings.

The genuinely nice bit is prompt_template. It's simple string replacement with two placeholders: {base} and {artist}. The README's examples show the point:

• {base}, {artist} - artist last, no special treatment
• {base}, (style of {artist}:1.2) - artist emphasized and weighted
• (artist:{artist}), {base} - artist up front, locked in as a subject

Because it's just text substitution, you can move the artist anywhere and apply any syntax your model's encoder understands. Leave the template empty and it falls back to {base}, {artist}.

The inputs that matter

• clip - required, no default. Wire the CLIP output from your checkpoint loader or CLIPLoader; forget it and the node won't even build.
• positive_base - your normal prompt, defaulting to the classic "masterpiece, best quality, 1girl, solo, portrait".
• artist_tag - where the iterator's output lands.
• prompt_template - discussed above.
• negative_prompt - the default is the standard SDXL-era blocklist (lowres, bad anatomy, bad hands, watermark…), a perfectly fine starting point.

Outputs

positive and negative (both CONDITIONING) go straight into your KSampler, and final_text is the fully composed prompt string - that's what you feed to Save Image (Smart Name) as its filename_prefix, so every saved file is named after the artist it was generated with.

The catch: this node lives or dies by your text encoder

This is where the pack shows its target audience. The (style of {artist}:1.2) weight syntax is passed through verbatim and only does anything on CLIP-encoded models - SD 1.5, SDXL, Illustrious, NoobAI, Pony. On the LLM-encoded family (Flux, Anima, Klein, Z-Image), (word:1.3) is silently discarded and the negative prompt is largely ignored, so this composer's whole approach - tag-style prompts with weights and a separate negative - is a CLIP-era tool. If you're on an LLM-encoded model, you'd get more from structured instructions than from this template. For the tag-model crowd it's exactly right, and the negative prompt stays very much alive there.

Installing

Same pack as the iterator: ComfyUI Manager → search "ComfyUI-Artist-Tester", or clone https://github.com/g7b2/ComfyUI-Artist-Tester.git into ComfyUI/custom_nodes and restart. No dependencies, no model downloads - it leans only on ComfyUI's own CLIP encoding. It lives under the "🎨 Artist Tester" category.

Wiring it up

The full loop: ArtistListIterator → artist_tag into this node → positive/negative into KSampler → decode → Save Image (Smart Name) with filename_prefix ← final_text. That's a complete artist batch in five nodes, and nothing about the template is hidden state - the composed string is right there in final_text if you ever want to see exactly what you sent the sampler.

One last thing: it's not smart. There's no LLM and no guesswork - {artist} gets replaced with whatever text arrives, so a typo in your list is a typo in your render. For a batch tool, that's a feature: predictable beats clever when you're comparing forty outputs.