MNeMoNiCuZ-ComfyUI-mnemic-nodes/web/docs/MNeMiC_Ideogram4RandomPrompter.md

🎲 Ideogram 4 Random Prompter

Experimental. Generates a complete random Ideogram 4 caption: a scattered set of regions, each described with words pulled live from the wonderwords dictionary, plus random style, lighting, medium, palette and arrangement.

Nothing here is curated — every content word is a dictionary word. It is a generator to explore with, not a prompt assistant. Everything stays visible on the node for that reason; nothing is hidden behind Advanced.

Requires the wonderwords package.

Inputs

Canvas:

• seed — Same seed and settings give the exact same caption, preview and boxes.
• width / height — Canvas size in pixels; Ideogram 4 prefers multiples of 16.

How many regions:

• region_count_min / region_count_max — The count is rolled between them, inclusive. Set both the same for a fixed count.
• background_weight / large_weight / medium_weight / small_weight — Relative likelihood of each size tier. Background blocks cover 70%+ of the canvas; large is ~34-62% per axis, medium ~17-40%, small ~5-17%. All four are summed and each region picks a tier proportionally; all zero falls back to equal weighting.

Wording:

• word_length_bias — Preferred word length in characters. 0 means no preference.
• word_length_randomness — Spread around that length. Ignored when bias is 0.
• scene_framing — pure gives each region a bare word list, which Ideogram tends to render as a collage of separate items. scene weaves the same words into one sentence with articles and spatial connectors, which reads as a single coherent scene — this is what makes photograph actually look like a photograph. Connector words do not count toward the word budget.
• region_word_min / region_word_max — Content words per region.

Regions:

• freeform_chance — Chance a region drops its bounding box and blends softly into the scene instead of sitting in a hard rectangle. Freeform regions are excluded from the bbox output.
• text_region_min / text_region_max — How many regions are rendered as in-image text instead of objects. An exact count, clamped to the total region count.
• element_palette_chance — Chance a region carries its own colour sub-set instead of the image palette. No effect when color_palette is none.

Style:

• medium — photograph writes a photo style key with focal length and aperture; anything else writes an art_style key.
• color_palette — Palette family, mirroring 🎨 Colorful Starting Image. none emits no palette at all and lets Ideogram choose.
• color_harmony — Relationship between the palette colours. Ignored when the palette is none.
• positioning_bias — Where regions cluster. Ignored unless arrangement is none.
• arrangement — spiral, burst or grid structured placement; overrides positioning_bias.

Description:

• description_length — Target word count for the auto-generated high-level description.
• description_override — Use this text verbatim as the high-level description; the prefix and length settings are then ignored.
• description_prefix — Prepended to the generated high-level description.
• description_background_prefix — Prepended to the generated background description.

Outputs

Same shape as 🧩 Ideogram 4 Prompt Builder:

• prompt — The caption as JSON text.
• preview — Rendered preview of the regions.
• bboxes — Region boxes in pixels, freeform regions excluded.
• width / height — Passed through.

Tip

For something that looks like a real photo rather than an asset sheet: scene_framing on, medium = photograph, a low region count, and high background/large weights.