zai-image-generation-skill

What I do

I provide the exact recipe for an agent to generate an image from a text prompt via the Z.AI GLM-Image API and save it to a local file. OpenCode's provider layer is chat-only, so image generation (a separate /images/generations endpoint) is reachable only through a direct HTTP call. The API returns a temporary URL (expires in ~30 days) on mfile.z.ai, so the recipe always downloads the result to a persistent file on disk.

Why a skill (not a provider)

• OpenCode providers (@ai-sdk/openai-compatible) speak chat completions — they cannot hit the /images/generations endpoint or return binary files.
• GLM-Image is a dedicated generation model (glm-image, cogview-4-250304), separate from the chat/vision lineup. Calling it requires a targeted POST {base}/images/generations.
• The result is a URL, not text — a file must be produced and its path returned to the caller.

Prerequisite — API key resolution

The recipe resolves the key robustly (env first, then opencode's credential store):

KEY="${ZAI_API_KEY:-$(jq -r '.["zai-coding-plan"].key // .["zai"].key // empty' \
     ~/.local/share/opencode/auth.json 2>/dev/null)}"
[ -z "$KEY" ] && { echo "ZAI_API_KEY not found — set it (export ZAI_API_KEY=...) or run \`opencode auth login\` (Z.AI)."; exit 1; }

If neither source has the key, stop and report — do not proceed.

Recipe

Set the options, then run generate → extract URL → download → verify:

# --- options ---
PROMPT="A cute kitten on a sunny windowsill, blue sky and white clouds"  # REQUIRED
OUT="${OUT:-./glm-image-$(date +%s).png}"          # output path (default: ./glm-image-<ts>.png)
SIZE="${SIZE:-1280x1280}"                           # glm-image: 1280x1280 | 1568x1056 | 1056x1568 | 1472x1088 | 1088x1472 | 1728x960 | 960x1728
QUALITY="${QUALITY:-standard}"                      # standard (~5-10s) | hd (~20s, richer detail)
MODEL="${MODEL:-glm-image}"                         # glm-image (default) | cogview-4-250304
ENDPOINT="${ZAI_IMAGE_ENDPOINT:-https://api.z.ai/api/coding/paas/v4}"  # coding plan (subscription). Alt: https://api.z.ai/api/paas/v4 (pay-as-you-go)

# --- 1. generate ---
RESP=$(curl -sS --max-time 180 -X POST "$ENDPOINT/images/generations" \
  -H "Authorization: Bearer $KEY" \
  -H 'Content-Type: application/json' \
  -d "$(python3 -c 'import json,sys
print(json.dumps({"model":sys.argv[1],"prompt":sys.argv[2],"size":sys.argv[3],"quality":sys.argv[4]}))' \
      "$MODEL" "$PROMPT" "$SIZE" "$QUALITY")")

# --- 2. extract URL (or surface the error) ---
URL=$(printf '%s' "$RESP" | python3 -c '
import sys, json
d = json.load(sys.stdin)
if d.get("error"):
    sys.stderr.write("API error: " + json.dumps(d["error"]) + "\n"); sys.exit(1)
print(d.get("data",[{}])[0].get("url",""))') || { echo "Generation failed. Response: $RESP"; exit 1; }
[ -z "$URL" ] && { echo "No image URL in response: $RESP"; exit 1; }

# --- 3. download to file (follow redirects; URL contains '?') ---
curl -L -sS --max-time 120 -o "$OUT" "$URL" || { echo "Download failed for $URL"; exit 1; }

# --- 4. verify it's a real image ---
file "$OUT" | grep -qiE 'image|png|jpeg' && echo "SAVED: $OUT ($(wc -c <"$OUT") bytes)" \
  || { echo "Downloaded file is not an image: $(file "$OUT")"; exit 1; }