DIGIT Random Prompt

Blank-canvas paralysis is real, and this node is the cure: it assembles randomized cinematic image prompts from curated pools - setting, location, time of day, weather, mood, subject, camera body, lens, film stock, lighting, color palette, composition, and era style. Click run, get a prompt like "a lone figure on a rain-slicked Tokyo street at dusk, shot on a Fujifilm GFX 100 II with a 35mm lens, Kodak Tri-X 400, chiaroscuro lighting". It's an idea machine with genuinely film-literate vocabulary - real camera bodies, real film stocks, real composition language - which is a cut above the generic "random prompt" nodes that just shuffle adjectives.

Where it shines is iteration and exploration. You can use it as a mood board generator (run it fifty times, keep the prompts that click), as a source of variety when you're stuck on the same subject framing, or as the randomness half of a pipeline where Prompt Combine adds your fixed subject and the Random Prompt supplies the scene. Each category is a dropdown that defaults to random but can be pinned to a specific value - so you can lock "camera: Leica SL3" and randomize everything else.

How it works

The node keeps a pool per category - things like actual medium-format and large-format cameras, film stocks from Tri-X to Aerochrome, lighting schemes from chiaroscuro to backlit-with-flare, era styles from Art Deco to film noir. On each run it picks one entry per active category and joins them into a single prompt string. Set seed to 0 (the default) and every run is different; set a number and the same prompt comes back, which is how you reproduce a specific composition you liked.

The inputs:

• seed - 0 = random every run; any number = deterministic.
• setting, location, time_of_day, weather, mood, subject, camera, lens, film_stock, lighting, color_palette, composition, era_style - each random or a pinned pick. Note subject is in here too, so you can lock your actual subject and randomize the cinematography around it.
• custom_prefix / custom_suffix - text added to the front and back of the generated prompt. This is how you inject "in the style of..." or a fixed quality string.

The single output is prompt - the full string, ready to feed into a Prompt Combine or straight into a text encoder.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/thedepartmentofexternalservices/comfyui-digit.git
cd comfyui-digit
pip install -r requirements.txt

(Or ComfyUI Manager → search comfyui-digit → install.) Restart ComfyUI, look under DIGIT.

Common issues

The honest limits: it's random, so most outputs won't be keepers - that's the point, but don't expect a hit rate. And the pools are image-cinema vocabulary; if you want video-specific prompting (camera movement, shot duration) this won't give it to you. Two practical notes: set seed to a fixed value when you're iterating on a specific prompt you liked, or you'll never see it again; and if you want your actual subject in every prompt, type it into custom_prefix or the subject slot rather than hoping the random pools happen to land on it.

One more thing worth knowing: the node's output is a plain prompt string with no seed_used output in the current build - if you need to know which seed produced a given prompt for reproducibility, read the seed you set before running. It's a simple, cheerful utility, and it's the rare node that makes the graph more fun to use rather than more correct.