APNext Cinematic

If you only ever install one node from dagthomas's builder family, make it this one. It's the cinematic-language dropdown: directors, specific films to reference, film stocks and resolutions, genres, shot types, color grading, cinematic effects, and a frankly unhinged 1,027-entry "aesthetics" list. You pick, it appends to your prompt. "In the style of Blade Runner (1982), 35mm film, cool-toned color grading, dolly shot" is three clicks away instead of three lines of typing you half-remember.

The reason this one beats the rest of the family isn't the size of the lists - it's that cinematic terms are concrete physical referents. A film stock, a lens, a color-grade, a shot size: these name real optical facts, and that's the single most reliable kind of prompt phrase there is. Naming physical equipment and photographic technique works across encoders - it's one of the few things that transfers cleanly from SDXL onto the 2026 LLM-encoded models, where quality-word spam does nothing. So this builder ages better than the artist- or brand-name ones.

How it works

Pure Python, no model, no key. Your prompt gets joined with the selected terms via the separator (comma default) into a string. Every menu opens with None / Random / Multiple Random, so with a seed set it's also a cinematic randomizer - great for generating a batch of the same scene under different grades and shots.

The inputs and outputs that matter

• prompt - your base scene; the cinematic terms fold in around it.
• The menus that move the image most: directors (50 - Kurosawa, Nolan, the Coens), movies (146 - reference a specific film's look), film_type (23 - 16mm, 35mm, 70mm, anamorphic, resolutions), color_grading (38 - cross-processed, day-for-night, desaturated), shot_type (57 - aerial, close-up, dolly, crane) and cinematic_effects (37 - bokeh, anamorphic, chromatic aberration, cinestill).
• genre (179), color_modifiers (57), effects (24) and the enormous aesthetics (1,027) list are there for mood and micro-genre exploration.
• seed - drives Random / Multiple Random.
• attributes (off by default) - adds the pack's extra descriptor words for each entry.

Outputs are prompt and random, both strings, into a CLIP Text Encode.

How to install it

ComfyUI Manager: search comfyui_dagthomas, install, restart. Manual:

cd ComfyUI/custom_nodes && git clone https://github.com/dagthomas/comfyui_dagthomas
cd comfyui_dagthomas && pip install -r requirements.txt

then restart. That requirements.txt installs the whole pack - openai, anthropic, google-generativeai, transformers, decord, scipy - because the repo also carries cloud-LLM and video nodes. This node needs none of them; decord is the usual build failure, and the builders generally still load if it does.

Where people get burned

• The aesthetics list is a rabbit hole. 1,027 entries is a lot of very specific micro-aesthetics, most of which the model has no clean handle for. Use it for discovery, not as a reliable lever. Real film stocks and grades do far more.
• Contradictory grades. A color_grading pick and a color_modifiers pick can pull opposite directions and average out. Change one at a time, fixed seed, and A/B.
• Director names are the softest lever here. "Christopher Nolan" evokes less than "70mm film, cool color grading, wide shot" does. Prefer the concrete photographic terms; use director/movie picks as a seasoning on top.
• Reference films drift. "Blade Runner (1982)" biases atmosphere and palette, not an exact frame. Treat it as a mood, not a lookup.