クロードAgents · Claude Code

Let Claude draw for your project.

A skill is a folder with a SKILL.md that tells Claude when and how to do something. This one lets Claude Code generate anime images through AnimeAPI and save them where your project needs them: README art, game sprites, placeholder characters. It knows the video API too, for short animated clips.

Install the AnimeAPI skill for Claude Code. Create these two files exactly, then run chmod +x on the script:

~/.claude/skills/animeapi/SKILL.md
```markdown
---
name: animeapi
description: Generate anime-style images (characters, scenes, sprites, backgrounds, key art) and short anime video clips with AnimeAPI and save them to disk. Use when the user asks for anime or manga style images, animations or visual assets.
allowed-tools: Bash(${CLAUDE_SKILL_DIR}/scripts/generate.sh *)
---

# AnimeAPI: anime image and video generation

Generate anime-style images (and, when asked, short anime video clips) through
AnimeAPI and save them into the project.

## Setup

`ANIMEAPI_KEY` must be set in the environment the agent runs in (for example
`export ANIMEAPI_KEY=ank_...` in your shell profile). A `.env` file is not
read automatically. Keys: https://animeapi.ai/dashboard/api-keys

## How to generate

Run:

```bash
${CLAUDE_SKILL_DIR}/scripts/generate.sh "<prompt>" <output-file> [model] [orientation]
```

The script needs `curl` and `jq`.

- `model`: `anime` (default, $0.003, best for characters, tag-based),
  `ultra` ($0.01, follows long sentences, renders legible text) or
  `ultra-max` ($0.02, highest detail).
- `orientation`: `portrait` (default), `square` or `landscape`.
- Use a `.webp` output path for `anime` and `.png` for the ultra tiers.
- The script prints `enhanced_prompt` (what the model actually read) and the
  remaining `balance_usd`.

## Writing prompts

- anime: describe the character first (who, hair, outfit, pose), then the
  place and light. Your prompt is rewritten into Danbooru tags automatically.
  It is strongly biased toward drawing a person; for animals or objects as the
  subject, use ultra.
- ultra / ultra-max: write a full sentence; an anime style is added
  automatically. Put any text that must appear in the image in quotes.

## Errors

- 401 INVALID_KEY: ANIMEAPI_KEY is missing or revoked.
- 402 INSUFFICIENT_BALANCE: tell the user to top up at
  https://animeapi.ai/dashboard/billing. Do not retry.
- 400 MODERATION_BLOCKED (ultra tiers): rephrase the prompt.
- 500 GENERATION_FAILED: the charge was refunded; one retry is fine.

Each image costs money. Generate one image, show it, and ask before making
batches.

## Videos

Only when the user asks for a video or animation. Videos are MP4 (24 fps, with
sound), 5 to 15 seconds, billed per second up front: $0.016/s at 480p,
$0.04/s at 768p (a 5-second 768p clip is $0.20). State the price and ask
before anything longer than 5 seconds or more than one clip.

1. Start a job (build the JSON with `jq -n` if the prompt has quotes):

   ```bash
   curl -sS https://animeapi.ai/api/videos -H "Authorization: Bearer $ANIMEAPI_KEY" \
     -H "Content-Type: application/json" \
     -d '{"prompt": "<what happens>", "resolution": "768p", "duration": 5, "orientation": "landscape"}'
   ```

   It returns 202 with an `id`. Optional `first_frame` / `last_frame`: an
   image URL or a `data:image/png;base64,...` data URI (up to 10 MB) that the
   clip starts / ends on; the first frame sets the shape. For a local image,
   pipe it through jq into a file (too long for a command-line argument):

   ```bash
   base64 < in.webp | tr -d '\n' | jq -Rs --arg p "<what happens>" \
     '{prompt: $p, duration: 5, first_frame: ("data:image/webp;base64," + .)}' > /tmp/video.json
   curl -sS https://animeapi.ai/api/videos -H "Authorization: Bearer $ANIMEAPI_KEY" \
     -H "Content-Type: application/json" -d @/tmp/video.json
   ```

2. Poll every 5 seconds until `status` is `succeeded` or `failed`:

   ```bash
   curl -sS https://animeapi.ai/api/videos/<id> -H "Authorization: Bearer $ANIMEAPI_KEY"
   ```

3. On `succeeded`, download `video_url` into the project (`curl -sS -o out.mp4 <video_url>`).
   On `failed`, the charge was refunded and `error` says why.
```

~/.claude/skills/animeapi/scripts/generate.sh
```bash
#!/usr/bin/env bash
# Usage: generate.sh "<prompt>" <output-file> [model] [orientation]
set -euo pipefail

PROMPT="${1:?usage: generate.sh \"<prompt>\" <output-file> [model] [orientation]}"
OUT="${2:?usage: generate.sh \"<prompt>\" <output-file> [model] [orientation]}"
MODEL="${3:-anime}"
ORIENTATION="${4:-portrait}"
: "${ANIMEAPI_KEY:?Set ANIMEAPI_KEY first: https://animeapi.ai/dashboard/api-keys}"

# jq builds the JSON so quotes and newlines in the prompt are escaped safely.
BODY=$(jq -n --arg p "$PROMPT" --arg m "$MODEL" --arg o "$ORIENTATION" \
  '{prompt: $p, model: $m, orientation: $o, allowNSFW: false}')

RESPONSE=$(curl -sS https://animeapi.ai/api/generate \
  -H "Authorization: Bearer $ANIMEAPI_KEY" \
  -H "Content-Type: application/json" \
  -d "$BODY" \
  -w $'\n%{http_code}')
STATUS=$(tail -n1 <<<"$RESPONSE")
JSON=$(sed '$d' <<<"$RESPONSE")

if [ "$STATUS" != "200" ]; then
  echo "AnimeAPI error $STATUS $(jq -r '.code // ""' <<<"$JSON"): $(jq -r '.error' <<<"$JSON")" >&2
  exit 1
fi

mkdir -p "$(dirname "$OUT")"
curl -sS -o "$OUT" "$(jq -r '.image_url' <<<"$JSON")"
echo "saved: $OUT"
echo "enhanced_prompt: $(jq -r '.enhanced_prompt' <<<"$JSON")"
echo "balance_usd: $(jq -r '.balance_usd' <<<"$JSON")"
```

→Or install it by hand

1. Put your key in the environment

   Claude runs the script in your shell, so the key must be exported there. Add it to ~/.zshrc or ~/.bashrc and restart Claude Code.

   export ANIMEAPI_KEY="ank_your_key_here"

2. Create the skill folder

   Personal skills live in ~/.claude/skills/ and work in every project. Use .claude/skills/ inside a repo to share it with your team instead.

   mkdir -p ~/.claude/skills/animeapi/scripts

3. Add SKILL.md

   ---
   name: animeapi
   description: Generate anime-style images (characters, scenes, sprites, backgrounds, key art) and short anime video clips with AnimeAPI and save them to disk. Use when the user asks for anime or manga style images, animations or visual assets.
   allowed-tools: Bash(${CLAUDE_SKILL_DIR}/scripts/generate.sh *)
   ---
   
   # AnimeAPI: anime image and video generation
   
   Generate anime-style images (and, when asked, short anime video clips) through
   AnimeAPI and save them into the project.
   
   ## Setup
   
   `ANIMEAPI_KEY` must be set in the environment the agent runs in (for example
   `export ANIMEAPI_KEY=ank_...` in your shell profile). A `.env` file is not
   read automatically. Keys: https://animeapi.ai/dashboard/api-keys
   
   ## How to generate
   
   Run:
   
   ```bash
   ${CLAUDE_SKILL_DIR}/scripts/generate.sh "<prompt>" <output-file> [model] [orientation]
   ```
   
   The script needs `curl` and `jq`.
   
   - `model`: `anime` (default, $0.003, best for characters, tag-based),
     `ultra` ($0.01, follows long sentences, renders legible text) or
     `ultra-max` ($0.02, highest detail).
   - `orientation`: `portrait` (default), `square` or `landscape`.
   - Use a `.webp` output path for `anime` and `.png` for the ultra tiers.
   - The script prints `enhanced_prompt` (what the model actually read) and the
     remaining `balance_usd`.
   
   ## Writing prompts
   
   - anime: describe the character first (who, hair, outfit, pose), then the
     place and light. Your prompt is rewritten into Danbooru tags automatically.
     It is strongly biased toward drawing a person; for animals or objects as the
     subject, use ultra.
   - ultra / ultra-max: write a full sentence; an anime style is added
     automatically. Put any text that must appear in the image in quotes.
   
   ## Errors
   
   - 401 INVALID_KEY: ANIMEAPI_KEY is missing or revoked.
   - 402 INSUFFICIENT_BALANCE: tell the user to top up at
     https://animeapi.ai/dashboard/billing. Do not retry.
   - 400 MODERATION_BLOCKED (ultra tiers): rephrase the prompt.
   - 500 GENERATION_FAILED: the charge was refunded; one retry is fine.
   
   Each image costs money. Generate one image, show it, and ask before making
   batches.
   
   ## Videos
   
   Only when the user asks for a video or animation. Videos are MP4 (24 fps, with
   sound), 5 to 15 seconds, billed per second up front: $0.016/s at 480p,
   $0.04/s at 768p (a 5-second 768p clip is $0.20). State the price and ask
   before anything longer than 5 seconds or more than one clip.
   
   1. Start a job (build the JSON with `jq -n` if the prompt has quotes):
   
      ```bash
      curl -sS https://animeapi.ai/api/videos -H "Authorization: Bearer $ANIMEAPI_KEY" \
        -H "Content-Type: application/json" \
        -d '{"prompt": "<what happens>", "resolution": "768p", "duration": 5, "orientation": "landscape"}'
      ```
   
      It returns 202 with an `id`. Optional `first_frame` / `last_frame`: an
      image URL or a `data:image/png;base64,...` data URI (up to 10 MB) that the
      clip starts / ends on; the first frame sets the shape. For a local image,
      pipe it through jq into a file (too long for a command-line argument):
   
      ```bash
      base64 < in.webp | tr -d '\n' | jq -Rs --arg p "<what happens>" \
        '{prompt: $p, duration: 5, first_frame: ("data:image/webp;base64," + .)}' > /tmp/video.json
      curl -sS https://animeapi.ai/api/videos -H "Authorization: Bearer $ANIMEAPI_KEY" \
        -H "Content-Type: application/json" -d @/tmp/video.json
      ```
   
   2. Poll every 5 seconds until `status` is `succeeded` or `failed`:
   
      ```bash
      curl -sS https://animeapi.ai/api/videos/<id> -H "Authorization: Bearer $ANIMEAPI_KEY"
      ```
   
   3. On `succeeded`, download `video_url` into the project (`curl -sS -o out.mp4 <video_url>`).
      On `failed`, the charge was refunded and `error` says why.

   allowed-tools lets Claude run this one script without asking each time. ${CLAUDE_SKILL_DIR} points at the skill folder wherever it’s installed.

4. Add the script

   #!/usr/bin/env bash
   # Usage: generate.sh "<prompt>" <output-file> [model] [orientation]
   set -euo pipefail
   
   PROMPT="${1:?usage: generate.sh \"<prompt>\" <output-file> [model] [orientation]}"
   OUT="${2:?usage: generate.sh \"<prompt>\" <output-file> [model] [orientation]}"
   MODEL="${3:-anime}"
   ORIENTATION="${4:-portrait}"
   : "${ANIMEAPI_KEY:?Set ANIMEAPI_KEY first: https://animeapi.ai/dashboard/api-keys}"
   
   # jq builds the JSON so quotes and newlines in the prompt are escaped safely.
   BODY=$(jq -n --arg p "$PROMPT" --arg m "$MODEL" --arg o "$ORIENTATION" \
     '{prompt: $p, model: $m, orientation: $o, allowNSFW: false}')
   
   RESPONSE=$(curl -sS https://animeapi.ai/api/generate \
     -H "Authorization: Bearer $ANIMEAPI_KEY" \
     -H "Content-Type: application/json" \
     -d "$BODY" \
     -w $'\n%{http_code}')
   STATUS=$(tail -n1 <<<"$RESPONSE")
   JSON=$(sed '$d' <<<"$RESPONSE")
   
   if [ "$STATUS" != "200" ]; then
     echo "AnimeAPI error $STATUS $(jq -r '.code // ""' <<<"$JSON"): $(jq -r '.error' <<<"$JSON")" >&2
     exit 1
   fi
   
   mkdir -p "$(dirname "$OUT")"
   curl -sS -o "$OUT" "$(jq -r '.image_url' <<<"$JSON")"
   echo "saved: $OUT"
   echo "enhanced_prompt: $(jq -r '.enhanced_prompt' <<<"$JSON")"
   echo "balance_usd: $(jq -r '.balance_usd' <<<"$JSON")"

   chmod +x ~/.claude/skills/animeapi/scripts/generate.sh

   It needs curl and jq. jq builds the JSON, so prompts with quotes or apostrophes are safe.

→Use it

Ask for what you want. Claude picks the skill up from its description, or you can call it directly with /animeapi.

• “Make a hero image for the README, a girl coding at night by a rainy window, landscape, save to docs/hero.webp”
• “Generate three enemy portraits for the RPG in square format into assets/enemies/”
• “Use ultra to make a cafe sign that says OPEN LATE, save to public/sign.png”
• “Animate docs/hero.webp into a 5-second clip where she looks up from the laptop, save to docs/hero.mp4”

Videos go through curl rather than the script (start a job, poll it, download the MP4), so Claude Code asks before running those commands.