Neon Halloween Ghost · सार्वजनिक Open Skill
Neon Halloween Ghost
Neon Halloween Ghost Digen पर एक सार्वजनिक Open Skill है। खोलें और जानें कि यह वर्कफ़्लो क्या बनाता है।
Manhua
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/6axzdynt.jpg
# Neon Halloween Ghost
Transform one user-uploaded reference image into a "neon Halloween ghost" look: near-black background, vivid neon rim light, and a few subtle ghost details, while the original subject, pose and framing stay completely unchanged and remain the single focal point at the center of the frame. Single image-to-image pass, no multi-stage flow, no structured state.
## When to Use
- The user uploads or points to an image and asks for a Halloween / neon / ghost style.
- The user says "swap the background" or "swap the outfit" to the neon Halloween ghost theme, or simply "make this image a neon Halloween ghost".
- The user gives feedback on the previous result: change the neon palette, add/remove ghost elements, regenerate, or switch between portrait and landscape.
- The user provides **no** reference image: ask them to upload one first (this skill is reference-image driven; do not fake it with pure text-to-image).
## Steps
1. **Get the reference image**: use the image uploaded in the current turn (take its `asset_id` or URL). If several were uploaded, take the first one or ask which to use. Operate on a single image only; do not accept an `s3://` directory prefix.
2. **Match the aspect ratio**: pick `orientation` based on the reference image shape — `portrait` for a vertical image, `landscape` for a horizontal one, `square` for a square one; if you cannot tell, do not pass it (use the default). If the user explicitly asks for high quality, pass `resolution='1080P'`; otherwise do not pass it (default 720P). Do not invent pixel dimensions.
3. **Build the prompt** (English; fill in the subject description, summarized in 1–2 words, e.g. `the person` / `the cat` / `the motorcycle`):
```
Halloween neon ghost transformation of <subject> from the reference image.
Keep the original subject, pose, framing and identity completely unchanged,
as the single clear focal point in the center of the frame.
Dark, near-black background. The subject is outlined and rim-lit by vivid neon
light — electric purple, toxic green, cyan and pumpkin orange — with glowing
edges, neon tube accents and soft bloom.
Add only a few subtle ghost details: faint translucent spectral wisps and small
floating ghost silhouettes near the edges of the frame, thin drifting smoke,
a soft ectoplasm glow.
Do not turn the subject into a full ghost; it must stay recognizable.
High contrast, crisp neon outlines on a dark background, subtle grain,
Halloween atmosphere, cinematic lighting.
```
Append negative constraints (may be concatenated at the end of the prompt): `no cluttered background, no heavy fog covering the face, no text, no watermark, no full-body ghost replacement, no extra subjects in the center.`
4. **Call the generation tool** (hardcode these key parameters, do not change them):
```
generate_image(
prompt=<the template above>,
input_urls=[reference image asset_id or URL],
model='i2i', # image-to-image channel; use 'i2i.hd' for stronger texture
orientation=<from step 2>,
name='neon-ghost-<short subject>', # short, for canvas display
entity_type='neon_ghost_images'
)
```
- Do not pass a physical model name (such as nano-banana or rm3.1-G); use the channel only.
- Pass only this one user-uploaded reference image; do not splice in other assets on your own.
5. **Iterate on user feedback**: color change → only edit the neon color words in the prompt; too much / too little ghost feel → only adjust the ghost details sentence (e.g. `only one small ghost silhouette in a corner` / `a few more spectral wisps`); subject altered → strengthen the prompt with `keep the original subject and identity unchanged, unchanged pose and composition` and generate again (`generate_image` always creates a new asset and never overwrites the original image).
## Notes
- If the tool returns `success=false` or no image, tell the user it failed honestly; n