j ddlmZddlmZeejdz dz Zeddee e ffdZ e d Z d Z dS) ) lru_cache)Pathmagic_prompt_system_promptszv1.txtN)maxsizereturnchi}d}g}tdD]}|}|drv|drad|vr]|*d|||<|dd}g}|||*d|||<d |vrttd |S) Nzutf-8)encoding[]  systemz has no [SYSTEM] section) _SYSTEM_PROMPT_FILE read_text splitlinesstrip startswithendswithjoinlowerappend ValueError)sectionscurrentlineslinestrippeds HC:\Users\johnn\OneDrive\Documents\AI\models\ideogram4\prompt_enhancer.py_load_sectionsr!s8HG E#--w-??JJLL::<<   s # # (9(9#(>(> 3hCVCV"$(IIe$4$4$:$:$<$<!qtn**,,2244GEE LL     IIe,,2244x/IIIJJJ Oraz # Ideogram 4 Prompt Notes Ideogram 4 was trained on structured JSON captions. Supply JSON directly, or optionally use **Magic Prompt** to turn a short idea into JSON. Direct plain text is not the upstream-supported prompting workflow. ## Recommended JSON Shape Use these top-level keys in this order: ```json { "high_level_description": "One or two sentences describing the whole image.", "style_description": { "aesthetics": "mood and visual keywords", "lighting": "lighting setup", "photo": "camera and lens details", "medium": "photograph", "color_palette": ["#FFFFFF", "#111111"] }, "compositional_deconstruction": { "background": "Environment and background description.", "elements": [ {"type": "obj", "bbox": [120, 260, 760, 760], "desc": "A detailed subject description."}, {"type": "text", "bbox": [80, 100, 180, 900], "text": "EXACT WORDS", "desc": "Typography and placement description."} ] } } ``` ## Layout And Bboxes - `compositional_deconstruction` is the most important section for spatial control. - `bbox` uses normalized coordinates from `0` to `1000`, ordered as `[y_min, x_min, y_max, x_max]`. - Keep each bbox aligned with its description: if an object is on the left, its `x` values should be low; if it is near the bottom, its `y` values should be high. - Text elements should use `type: "text"` and include both the exact `text` to render and a `desc` describing font, color, size, and placement. ## Style - In `style_description`, use either `photo` for photographic prompts or `art_style` for non-photo prompts, not both. - Include `aesthetics`, `lighting`, and `medium` when you include `style_description`. - Preserve style key order: `aesthetics`, `lighting`, `photo`, `medium` for photos; `aesthetics`, `lighting`, `medium`, `art_style` otherwise. Optional `color_palette` comes last. - Use uppercase hex colors like `#1B1B2F`. Up to 16 colors can guide the whole image; up to 5 can guide one element. ## Practical Tips - `compositional_deconstruction` is required, with `background` before `elements`. `high_level_description` is recommended; `style_description` is optional. - Keep object descriptions positive and concrete. - Serialize compactly when possible, with no comments or markdown fences in the prompt itself. - If writing JSON by hand is tedious, enable **Magic Prompt** and start from a short natural-language idea. ) functoolsrpathlibr__file__resolveparentrdictstrr!IDEOGRAM4_PROMPT_ENHANCERIDEOGRAM4_PROMPT_INFOSr"r r-sd8nn,,..58UUX`` 4S#X(+N,,X61r"