tylerprogramming Generate images with ByteDance Seedream through the Kie.ai jobs API. (Seedream is ByteDance's image model — the still-image sibling of Seedance video.)
Setup: needs
KIE_API_KEYin~/.claude/.env.
Models
| Key | Model id | Mode | Notes |
|---|---|---|---|
4.5 (default) |
seedream/4.5-text-to-image |
text-to-image | 2K or 4K |
4.5-edit |
seedream/4.5-edit |
image edit | Up to 14 input images; auto-selected when you pass --image |
5-lite |
seedream/5-lite-text-to-image |
text-to-image | Lighter/cheaper |
Flow
- Parse
$ARGUMENTSfor the concept. If the user gives image(s) to edit/combine → edit mode. - Settings (ask, or defaults on "just do it"):
- Aspect ratio:
16:9(default),1:1,9:16,4:3,3:4,2:3,3:2,21:9. - Quality:
basic(2K, default) orhigh(4K). - Variants: default 3.
- Aspect ratio:
- Craft the prompt (≤3000 chars). Show it before generating.
- Generate:
python3 ~/.claude/skills/kie-seedance-image/generate.py "<prompt>" \
--model 4.5 --aspect-ratio 16:9 --quality basic --count 3 \
[--image <url-or-path> ...] # any --image switches to seedream/4.5-edit
--slug <slug>
- Present: output →
~/images/seedream/<date>-<slug>/. Offer more variants, 4K, a different ratio, or an edit pass on a favorite.
Notes
- Passing
--imageauto-switches to the edit model (combine/restyle up to 14 images). - Local files are auto-uploaded; URLs pass through.
- For YouTube thumbnails specifically,
/yt-thumbnailis still the tuned tool; this skill is the general-purpose Seedream image generator.