Codex Fantastica

A wildcard pack for ComfyUI, tailored for building fantasy scenes. Everything is broken down into categories: characters, emotions, clothing/armor, poses (solo and paired), environments, camera angles, and styles. It works like a construction kit — you pick the files, and the model assembles the scene.

Important before using

These are strictly wildcards, not ready-made prompts or LoRAs. RNG is RNG: different seeds, the order you connect the files, and their combinations will yield very different results. Sometimes everything clicks perfectly on the first try, sometimes you'll need to regenerate or tweak the tag order a bit. I've tested all of this, the pack works stably overall and produces interesting compositions, so I decided to release it to the public.

About models and LoRAs

Keep in mind that the base model and LoRAs you use have a huge impact on the final result. Some models understand camera angles and composition tags really well, while others basically ignore them. If a pose or camera angle isn't working the way you expect, try a different checkpoint before blaming the wildcards. The same goes for LoRAs — certain style LoRAs can override or conflict with the wildcard logic.

What's inside the archive

The zip contains around 50 wildcard files organized into corresponding folders by category (characters, poses, emotions, environments, camera, nsfw, etc.). File names are self-explanatory — you'll figure out what goes where just by looking at them.

Technical note

I use comfyui_local_wildcards, but nothing stops you from using Impact Pack, Dynamic Prompts, ComfyUI-Wildcards, or any other node that can read __filename__. Doesn't matter, just make sure the files are in the correct folder for your specific extension.

My prompt template

Positive:

masterpiece, best quality, amazing quality, very aesthetic, extremely detailed, very detailed, absurdres, newest, highres, score_9, score_8, score_7, explicit,

Wildcard order:

characters, emotions, attire (or nude), (same with 2nd character), solo or paired poses, environments, camera

Negative:

worst quality, low quality, score_1, score_2, score_3, artist name, (lowres:1.2), (worst quality:1.4), (low quality:1.4), (bad anatomy:1.4), bad hands, bad anatomy, multiple views, jpeg artifacts, patreon logo, patreon username, web address, signature, watermark, logo, artist name, censored, mosaic censoring, loli, chibi

A couple of tips from my testing

- If the scene comes out too cluttered or elements conflict, try locking one or two wildcards to reproduce mode, or just manually type in the specific line you want.

- Tag order affects the model's priority. If clothing renders over the body or poses break, move poses closer to the beginning and environments to the end.

- Don't be afraid to change the seed and play with tag weights (tag:1.2) — sometimes that saves a composition way better than rewriting the whole prompt.

A quick apology

Some of the category headers inside the wildcard files are still in Russian (comments like # === MAGICAL ACTIONS === may appear in Cyrillic). This won't cause any issues — the actual prompt content is fully in English, and the model only reads the non-comment lines. I'll fix the headers in a future update, but for now everything works as intended.

The files inside are structured and the names speak for themselves. Mix and match, test things out, and share your results. Happy generating!