Prompt With Style V2 (Mikey)

This is the front door of an SDXL graph built the Mikey Nodes way: type your positive and negative prompts, pick a style and an aspect ratio from dropdowns, and get back an empty latent at the right size plus fully-encoded conditioning for both the base and refiner models - four CONDITIONING outputs and a LATENT, from one node, in one step. It exists specifically to replace the usual pile of CLIPTextEncodeSDXL, EmptyLatentImage, and separate ratio-math nodes with a single starting point.

How it works

You type a positive and negative prompt as plain text, then pick a style from the dropdown - the pack ships 86 choices including a none option and Stability AI's own official style set (SAI-Photographic, SAI-Anime, SAI-Digital art, SAI-Cinematic, and dozens more), plus a long tail of additional community-style presets. The style field is separate from the raw prompt syntax the README documents for the pack's main flagship node (Prompt With Style V3) - here you select a style from a list rather than typing <style:name> inline, which is exactly the trade-off the README frames as "if you like to break your workflow up into more nodes." ratio_selected picks one of SDXL's proper trained aspect ratios (1:1 up to 21:9 and their portrait equivalents) so the latent comes out at a resolution SDXL was actually trained on, rather than something stretched.

The node then runs both prompts through CLIP encoding twice - once for the base model's conditioning, once for the refiner's - using the clip_base and clip_refiner inputs you wire in, and hands back a matched set of conditioning ready to drop straight into Mikey Sampler.

The inputs and outputs that matter

• positive_prompt / negative_prompt (multiline STRING) - your actual prompt text.
• style - one of 86 preset styles, none if you'd rather not apply one.
• ratio_selected - one of SDXL's trained aspect ratios (15 presets), the same list the README's ratio nodes use.
• batch_size (default 1, max 64) and seed - standard generation controls.
• clip_base / clip_refiner (CLIP) - required; this node does its own encoding, it doesn't take pre-encoded conditioning.
• Outputs: samples (LATENT, empty, at the chosen ratio), base_pos_cond / base_neg_cond / refiner_pos_cond / refiner_neg_cond (CONDITIONING), and the raw positive_prompt / negative_prompt strings passed through - handy for feeding SaveMetaData or a caption node without retyping anything.

How to install it

Bundled with the rest of Mikey Nodes. ComfyUI Manager: search "Mikey Nodes," install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/bash-j/mikey_nodes

then restart. If you want your own custom styles beyond the built-in 86, the README's mechanism for that is a hand-created user_styles.json in your ComfyUI root - the pack doesn't ship this file, you write it yourself, in the exact {"styles": {"name": {"positive": "...", "negative": "..."}}} shape from the README.

Common issues & troubleshooting

Custom styles or ratios not showing up. Both user_styles.json and user_ratios.json have to be created by hand in the ComfyUI root directory - they're not included, and the README is explicit that you have to make the file yourself. A missing file just means you only see the built-in options, not an error.

Wired this into a Flux or SD 1.5 graph and it looks wrong. The ratio_selected list and the refiner-conditioning design are both SDXL-specific. This node assumes a base/refiner SDXL pair; feeding it a single-model architecture leaves half its purpose unused.

On a hosted ComfyUI, user_styles.json isn't there. Since it has to live in the ComfyUI root and isn't part of the git-cloned pack, a serverless or hosted executor that resets its filesystem between runs - comfy.icu included - won't retain a hand-placed file the way a persistent desktop install does. Stick to the built-in 86 styles unless your hosting setup gives you a way to persist custom files into the container.

No LoRA or wildcard syntax working in the prompt text. That inline <lora:name:weight> and __wildcard__ syntax is the flagship Prompt With Style V3 node's feature, described at the top of the pack's README. V2's positive_prompt/negative_prompt fields are plain text for CLIP encoding - don't expect the same inline parsing here.