📷 Wangz Camera Prompt Studio

Wangz Camera Prompt Studio does exactly what its name promises: it turns the fiddly business of typing "right side, extreme low-angle, wide shot, 28mm lens, shallow depth of field, film grain, golden hour" into seven dropdowns and one text box. The name is also a small lie you'll love - it calls no API, loads no model, and needs no key. It's a prompt assembly node: pure string glue. That's the best kind of ComfyUI tool, because it can't OOM your machine and can't corrupt a latent.

Why you'd reach for it: camera and lens language is the single highest-leverage thing you can put in a modern prompt, and block-structured prompts beat a comma soup on the LLM-encoded models (Flux Klein, Z-Image, that whole family) that dominate local generation now. This node is basically a UI for that block structure - the angle, the camera spec, the subject, and the mood each stay a separate, reusable chunk instead of one uneditable wall of text. The author's README frames it as merging two nodes you'd otherwise chain - Qwen Multiangle Camera plus CR Prompt Text - into one clean box with no spaghetti wires.

How it works

Under the hood it's boring in the best way: a single function reads the dropdowns, builds five blocks, and joins them. The order is fixed - angle block, then camera spec (lens + depth of field + film filter), then your base_prompt, then the mood - separated by whatever separator you pick (comma, newline, or comma+newline), with tail_params stuck on the very end. Any dropdown set to none is dropped entirely, so you're not paying tokens for a filter you didn't want. The output is a single STRING named prompt, and you can wire it into a CLIP Text Encode, into an LLM-encoded model's prompt input, or just read it off the node and paste it somewhere else.

The inputs that actually matter

• horizontal_angle, vertical_angle, shot_distance - these three build the angle block ("front-right quarter slightly high-angle close-up"). The defaults are sane; you'll mostly be twiddling vertical_angle and shot_distance.
• base_prompt - your actual subject ("Japanese Gothic-style ribbon-decorated maid girl" by default). This is where the real content lives.
• tail_params - defaults to --ar 16:9 --v 6.0, which is Midjourney syntax. Great if you're pasting into Midjourney. See the gotcha below if you're not.
• external_angle_prompt (optional) - if you connect Qwen Multiangle Camera's prompt output here, its angle text completely replaces the dropdown angle block. Handy when you want a 3D-angle node driving the shot while this node still handles lens, mood, and assembly.

Install

ComfyUI Manager → "Install via Git URL" → paste https://github.com/wjs0502m-design/Wangz_camera_prompt_studio, or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/wjs0502m-design/Wangz_camera_prompt_studio

Then restart ComfyUI and search "Camera Prompt Studio" or "📷" (it's under category Wangz/camera). There's no requirements.txt and no model download - the whole thing is stdlib Python, so install takes about three seconds.

Where people get burned

• The <sks> prefix is on by default. include_angle_prefix prepends <sks> to every angle block. That token only means something if you're using a LoRA/embedding trained with <sks> as its trigger word. If you aren't, it's dead text burning tokens at the LLM attention cap - flip it off.
• Those Midjourney flags will bite you in ComfyUI. Feed --ar 16:9 --v 6.0 into a ComfyUI prompt encoder and you're literally prompting the word "--ar". Empty tail_params for ComfyUI-internal use.
• The prefix is silently skipped on the external path. In the source, include_angle_prefix only applies to the dropdown branch - connect external_angle_prompt and the <sks> you thought you'd set never gets added. It's not a bug, just easy to miss.
• Don't blank the subject. With an empty base_prompt and everything at none, this node happily emits a near-empty string. Keep at least your subject in there.

This is a natural-language composer, so it shines on LLM-encoded models and Midjourney-style destinations. If your whole stack is Pony or Illustrious danbooru tags, it'll still work - "name a physical piece of equipment" lands on CLIP models too - but it's a different dialect than you're used to. For everything else, this is the one clean node for shot grammar.