Test Style Image · publiczny skill Open Skill
Test Style Image
Test Style Image
News
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/hpjkkxyv.jpg
# Test Style Image (image-to-image style redraw)
Redraw the single image the user uploaded into one unified "Testing" look: **the whole frame's color, material and lighting all converge on the single "Testing" style, and the frame serves that one theme only**; the subject in the original image is rewritten as it would look under the "Testing" theme.
- Input: **a reference image** (user upload / an image already on the canvas) + optional **text** notes
- Output: **one image** (added to the canvas automatically)
## When to use
- The user uploads or points at an image and asks for it to be turned into the "Testing" style, e.g. "turn this into a Testing image", "redraw this one in that style".
- The user wants another version of **the same image**, or only wants one thing changed (lighting, material, color, a subject detail).
- The user adds a short text note (what the subject is, wanting a portrait frame, a detail that should stand out more) — still use this skill.
Not applicable: pure text-to-image with no reference image at all (see the fallback under "Notes").
## Steps
1. **Get the reference image first**
Take this image's `asset_id` or `uri` (`s3://` / https) from the user's attachment / a canvas asset. For a user-uploaded image, pass `asset_id`; **do not** build a URL yourself and do not invent an input that does not exist. If it is unclear which image the user means, ask one question before starting.
2. **Write the prompt (English; the style word is fixed, and all four "unified" clauses must be present)**
Follow the template below, replacing the placeholders with this image's actual content:
```
{subject description from the original image, reimagined in the Testing theme},
unified Testing color palette — one single color scheme across the whole frame,
unified Testing material and surface finish,
unified Testing lighting,
one single theme, everything in frame serves that theme, no other style mixed in,
keep the original composition, framing, subject placement and layout of the reference image
```
- **Name the style with the one fixed term "Testing"**: keep the same term throughout the prompt — do not swap in synonyms and do not mix in any other style word. Color, material and lighting must each be written out explicitly as "unified".
- Make the subject description specific: first say what the subject in the original image is, then how it looks under the "Testing" theme; do not just write "make a Testing-style image".
- Translate any Chinese notes from the user (composition, details, mood) into English and fold them into the prompt — do not drop them.
- Follow the original layout: keep the original composition, subject placement and layout relationships, and do not add subjects that are not in the original unless the user explicitly asks for them.
3. **Call the generation tool (image-to-image)**
```python
generate_image(
prompt=<the prompt from the previous step>,
input_urls=[<asset_id or uri of the reference image>], # required, this is what makes it image-to-image
model='i2i', # image-to-image channel; use 'i2i.hd' for higher quality
orientation=<follow the original: portrait / landscape / square>,
name=<short title, e.g. "Testing style image" or the subject name>,
entity_type='test_images',
)
```
- **`input_urls` is required**; without it this degrades to text-to-image and is not this skill's output.
- `model` only takes a **channel**: with a reference image always use `i2i`, or `i2i.hd` for higher quality (`.hd*` does not support NSFW). **Do not** write a specific checkpoint or physical model name — if the tool rejects it, switch back to the channel and retry once.
- `orientation` follows the original image's direction; leave `resolution` unset by default (720P), and only pass it when the user explicitly asks for 1080P/2K. Do not write pixel width/height — express size only