Eric Hiss
513b975fb8
- Add DeepSeek Prompt Rewriter node for fast API-based rewriting - Add Load Image with Prompt node for metadata extraction - Add config.json.example with API backend templates - Add flux2/schema.yaml - Update .gitignore to exclude dev/ and config.json (contains API keys)
121 lines
5.1 KiB
YAML
121 lines
5.1 KiB
YAML
meta:
|
|
model: FLUX.2
|
|
description: >-
|
|
Structured prompting schema distilled from Black Forest Labs' official FLUX.2
|
|
guide. Use this for JSON-mode prompts or as reference when expanding
|
|
natural-language prompts with matching details.
|
|
references:
|
|
- https://huggingface.co/blog/flux-2#advanced-prompting
|
|
- https://docs.bfl.ai/guides/prompting_guide_flux2
|
|
|
|
prompt_formats:
|
|
natural_language:
|
|
summary: >-
|
|
Cinematic, sensory-rich prose (150-220 words) with photographic and
|
|
artistic terminology. Include hex colors, camera specs, and multi-reference
|
|
cues inline. Negative prompts are ignored by the base API but may be routed
|
|
downstream if the graph expects them.
|
|
reminders:
|
|
- No negative prompt support at the API level; only emit negatives if the
|
|
workflow explicitly consumes them.
|
|
- Mention reference directives exactly once each.
|
|
- Quote literal text ("OPEN") and hex codes (#FFAA33) when critical.
|
|
- Call out camera hardware ("shot on Hasselblad X2D", "85mm lens") for
|
|
photoreal scenes.
|
|
|
|
structured_json:
|
|
summary: >-
|
|
Emit a single JSON object (no markdown fences) that follows the schema
|
|
below. Prioritize clarity over brevity; the model accepts rich descriptions
|
|
inside every field.
|
|
required_keys:
|
|
- scene
|
|
- subjects
|
|
- style
|
|
- lighting
|
|
- mood
|
|
- background
|
|
- composition
|
|
- camera
|
|
schema:
|
|
scene: Overall scene description string.
|
|
subjects:
|
|
- description: Detailed subject description.
|
|
position: Where it sits within the frame.
|
|
action: What the subject is doing.
|
|
pose: Optional pose/gesture field.
|
|
color_palette: Optional array of hex codes or descriptive colors.
|
|
wardrobe: Optional wardrobe/material specifics.
|
|
accessories: Optional supporting props tied to the subject.
|
|
- "Repeat object for every subject in frame."
|
|
style: Artistic style or medium ("ultra-realistic product photography").
|
|
color_palette: Optional array describing the global palette.
|
|
lighting: Light quality, direction, setup ("three-point softbox").
|
|
mood: Emotional tone ("mysterious, high-tension").
|
|
background: Environment description.
|
|
composition: Framing guidance ("rule of thirds", "symmetrical").
|
|
camera:
|
|
angle: Camera angle ("low angle", "bird's-eye").
|
|
distance: Shot distance ("medium shot", "macro close-up").
|
|
focus: Focus behavior ("shallow depth on subject", "infinite").
|
|
lens: Lens descriptor ("prime portrait lens").
|
|
lens_mm: Numeric focal length when relevant.
|
|
depth_of_field: Qualitative DOF notes.
|
|
f_number: Aperture value string.
|
|
iso: ISO value.
|
|
shutter_speed: Optional shutter/exposure reference.
|
|
color_grading: Optional note about grade/film stock/processing.
|
|
props: Optional array of secondary objects with instructions.
|
|
post_processing: Optional description ("cross-processed Ektachrome").
|
|
text_overlays: Optional array with literal copy, placement, style.
|
|
references: Optional mapping describing how each image reference should
|
|
influence the output when available.
|
|
instructions: Optional free-form block for extra constraints.
|
|
validation:
|
|
- Output must be minified or neatly spaced JSON, never YAML or prose.
|
|
- Do not include trailing comments or explanations.
|
|
- Hex colors must include the leading '#'.
|
|
|
|
examples:
|
|
json_minimal: |
|
|
{
|
|
"scene": "Professional studio product photography setup with polished concrete surface",
|
|
"subjects": [
|
|
{
|
|
"description": "Minimalist ceramic coffee mug with steam rising",
|
|
"position": "Center foreground",
|
|
"action": "Stationary",
|
|
"pose": "upright",
|
|
"color_palette": ["matte black", "#FF2A2A"],
|
|
"wardrobe": "glossy ceramic glaze",
|
|
"accessories": "small chrome spoon to the right"
|
|
}
|
|
],
|
|
"style": "Ultra-realistic product photography",
|
|
"color_palette": ["matte black", "concrete gray", "soft white highlights"],
|
|
"lighting": "Three-point softbox with diffused highlights",
|
|
"mood": "Clean, professional, minimalist",
|
|
"background": "Polished concrete surface with blurred studio backdrop",
|
|
"composition": "rule of thirds",
|
|
"camera": {
|
|
"angle": "high angle",
|
|
"distance": "medium shot",
|
|
"focus": "Sharp on mug details",
|
|
"lens": "macro prime",
|
|
"lens_mm": 85,
|
|
"depth_of_field": "shallow",
|
|
"f_number": "f/5.6",
|
|
"iso": 200,
|
|
"shutter_speed": "1/125"
|
|
},
|
|
"color_grading": "Neutral cinematic grade with gentle contrast",
|
|
"post_processing": "Subtle clarity boost, no grain"
|
|
}
|
|
|
|
notes:
|
|
- JSON prompts are ideal for automation, multi-reference breakdowns, and
|
|
precise color control. Natural-language prompts remain better for quick
|
|
explorations or when downstream systems expect plain text.
|
|
- Feel free to add extra top-level keys if a workflow benefits from them; the
|
|
model silently ignores unknown fields.
|