Thumbnails

What it asks you for

Video topicrequired

What the video is about.

Headline

Short text to burn in (3–5 words).

Options

How many to produce.

What it tells the AI to do

The full instructions, exactly as they run.

Produce {{ count | default(3) }} thumbnail options at 1280×720.

- Topic: {{ topic }}
- Headline: {{ headline }}

Two rules that decide whether a thumbnail works:

1. **Compose the text in code, not in the model.** Generate the background/subject
   with `inference image`, then burn the headline in with ImageMagick or
   ffmpeg `drawtext` on a computer. Diffusion models cannot spell reliably, and
   a misspelled thumbnail is unusable.
2. **Leave the bottom-right corner clear** — the duration badge covers it, and
   anything important there is lost.

Aim for one focal point and text large enough to read at 168px wide. Deliver as
a `media.gallery` with `selection` so the user can pick their A/B pair.
## Delivery

Write the finished artifact into Drive. Never upload a user's content to a
third-party file host (`0x0.st`, `catbox.moe`, `transfer.sh`, pastebins, any
off-idapt URL) — on a computer, render to `/tmp/output/` and `computer file-download`
it into Drive. If you catch yourself reaching for `curl -F`, stop: you already
hold the artifact.

• image
• thumbnail
• youtube

More images skills

• Create an imageGenerate an image from a text prompt with a chosen model.
• Edit an imageEdit an existing image using a natural-language instruction.
• Remove the background