Neon Equation Portrait · सार्वजनिक Open Skill

Neon Equation Portrait

Neon Equation Portrait Digen पर एक सार्वजनिक Open Skill है। खोलें और जानें कि यह वर्कफ़्लो क्या बनाता है।

@eggs

Other

0 0 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/wvetbtgc.jpg

# Neon Equation Portrait

Recreates the SeaArt-style app "IMG2.5 Neon Equation Portrait": one hyper-realistic vertical cyber portrait per request, with an optional identity photo. Text is the primary driver (theme, wardrobe, pose, set); the reference photo only supplies the person's identity.

There is no structured project state — settings live in the conversation, products land on the canvas automatically.

## When to use

- The user uploads a photo and asks for a generated image in this neon-cyber-portrait look.
- The user asks for a "neon equations", "glowing formula portrait", "cyber math portrait", "neon glasses portrait" without a photo (text-only variant).
- The user wants a re-run or a single-detail change of a portrait produced earlier (new wardrobe, new theme, fewer/more formulas, different pose or set) — keep the identity and the neon treatment, change only what was asked.

## Inputs and output

- **Text (required):** theme, wardrobe, pose, framing wish, how many formulas, mood. If the user gives only "make one", use the defaults in `references/style-and-prompt.md`.
- **Reference image (optional):** the person to keep. When present it locks identity only — face geometry, eyes, brows, nose, mouth, hair color/length, skin tone, distinguishing marks. Wardrobe, pose and set are replaced by the theme.
- **Output:** one image asset on the canvas.

## Steps

1. Decide the mode:
   - Reference image provided → image-to-image, `model='i2i.hd.lite'`, `input_urls=[<reference image url>]`.
   - No reference image → text-to-image, `model='t2i'`, no `input_urls`; describe the subject yourself (adult woman, opaque dark wardrobe) plus the theme.
2. Before writing the prompt, call `read_skill_file('neon-equation-portrait', 'references/style-and-prompt.md')` and follow its Style / Subject / Layout / Must keep / Must avoid rules and its prompt template + fragment library. Do not improvise the look from memory.
3. Build one English `prompt` from the fragments: subject identity clause (or a described subject in text-only mode) + wardrobe + pose + set + equation overlay + glasses + lighting + background bokeh + framing. Replace the placeholders with the user's theme.
4. Call `generate_image`:
   - `prompt`: assembled English prompt.
   - `model`: `'i2i.hd.lite'` when a reference image is used, `'t2i'` otherwise.
   - `input_urls`: the user's reference image (omit in text-only mode).
   - `orientation='portrait'`. The template is composed natively at 3:4 — pass `aspect_ratio='3:4'` when the user asks for the app's exact framing; otherwise leave it out and let portrait framing stand.
   - `name`: a short title, e.g. `'Neon Equation Portrait'`.
   - `entity_type='characters'`.
5. Reply in one or two natural sentences describing what was generated (identity kept, wardrobe/set changed, formula treatment). Do not paste long URLs. On a finished portrait you may call `set_guidance` in the same turn with 2–4 related next steps (e.g. "more formulas on the jacket", "same person in a night-rain set", "recommend a voice-over companion clip").

Generate one image per request. Only produce several variants if the user asks for them — one call each.

## Notes

- Identity integrity comes first: never change who the person is, their hair color/length, or their facial structure. Wardrobe, pose and scene are what the theme may change.
- Keep clothing opaque and the subject a clearly adult person. Reject sexualized, nude, or minor-looking requests politely without calling the generation tool. The `i2i.hd.lite` channel is an HD channel and does not accept NSFW content.
- Never add title typography, before/after panels, "ORIGINAL INPUT" insets, logos or watermarks — those belong to app marketing art, not to the generated image.
- Do not invent content the app does not have (no extra characters, no props list, no text captions on the image) unless the user explicitly asks.
- Write `model` only as a channel (`i2i.hd.lite` / `t2i`); never a physical model name.