FLUX.2 JSON Prompt Generator

FLUX.2 moved away from the single free-text prompt your SD1.5 habits expect - it wants a structured JSON prompt with camera, scene, subject, style, lighting, and composition as separate fields. Typing that by hand every run is misery. This node is a giant form that assembles it for you, and here's the part that surprises people: it calls no API and needs no key. The name sounds like an LLM wrapper, but it's pure offline string assembly with a big pile of curated presets.

Open it and you'll see a long list of dropdowns - camera_preset, style_preset, lighting_preset, mood_preset, background_preset, composition_preset, focus_preset, lens_preset, f_number_preset, iso_preset, camera angle/distance/model - each with genuinely curated options, from "Kodachrome 1960s" film looks to "DJI Inspire 3 aerial" to "2000s digicam style." The defaults even ship a full sample product shot: a ceramic coffee mug with steam, studio three-point lighting, f/5.6, Sony A7IV.

How it's structured

Every preset has a matching *_text_override field, and the rule is consistent: a non-empty override wins over its preset. So you can lean on presets for 90% of a shot and hand-type the one thing that matters - scene, style_text_override, background, whatever.

• subject_count - number of subjects (0–10), parsed as int. Only subject_1_* is required; subjects 2–5 are optional inputs you add as needed, each with description, position, action, pose, and color palette.
• color_hex_1 through color_hex_5 - exact hex colors for the palette.
• include_shot_on_prefix - whether to prepend "Shot on [camera]" to the natural-language version.
• enable_prompt_expansion - this one is a trap in the fun way: the tooltip itself admits it's a no-op, "reserved for future auto-expansion." Leave it off.

Four outputs, and which you want depends on where the prompt goes:

• json_prompt - the structured JSON, pretty-printed.
• formatted_prompt - a readable natural-language version of the same shot.
• camera_settings - just the camera block, e.g. Sony A7IV | 85mm | f/5.6 | ISO 200.
• json_to_clip_prompt - a flattened version meant for the CLIP/text-encoder input when your workflow expects a plain string.

Wire json_prompt (or json_to_clip_prompt) into your FLUX.2 conditioning/text-encode node. That's the whole loop.

Install

Same as every node in this pack - Manager, search TBG Takeaways, install, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/Ltamann/ComfyUI-TBG-Takeaways

Restart, find it under TBG/Takeaways. No dependencies, no downloads.

Reach for this if you're building FLUX.2 product/editorial shots and want the structured prompt without memorizing the schema. Skip it if you're on FLUX.1 or SD-class models - the JSON prompt schema is FLUX.2's thing. And fair warning: it's a Patreon side-project, so the huge preset list is the polish and there's no doc beyond the tooltips. But for offline, schema-safe FLUX.2 prompting, it does exactly what it says.