What to build#

1. Bridge script: scripts/runpod-gen.sh#

./scripts/runpod-gen.sh \
  --endpoint <endpoint-id> \
  --prompt "..." \
  --width 1024 \
  --height 1024 \
  --output /tmp/runpod-output.png

The script:

1. Reads RUNPOD_API_KEY from env
2. Builds the JSON payload with prompt + params
3. POST to https://api.runpod.ai/v2/{endpoint_id}/runsync
4. Extracts image URL from response
5. Downloads to output path
6. Returns path on success, exit 1 on failure

2. Integration#

• Callable from any agent session via exec
• Same pattern as civitai-gen.sh for consistency

Acceptance criteria#

• Works with ComfyUI serverless endpoint — built-in workflow JSON for FLUX/FLUX1-dev/SDXL + custom workflow file support
• Environment variables: RUNPOD_API_KEY, RUNPOD_ENDPOINT_ID — both handled with CLI flag override
• Error handling (401, timeout, failed inference) — 401→clear error, polling→timeout→exit 2, failed inference→exit 1 with debug output
• Script callable from any agent session — same pattern as civitai-gen.sh, set -euo pipefail, bash syntax validated
• Manual smoke test with a test image — BLOCKED: Maxime hasn't provided endpoint ID. API key is set as RUNPOD_API_KEY in env. Once endpoint ID is known, run:

  ./scripts/runpod-gen.sh --endpoint <id> --prompt "test" --model flux --output /tmp/runpod-smoke.png
  

• All existing tests pass — syntax OK, help text OK, error handling OK

Blocked on#

t-3456 — Maxime needs to sign up, add credits, deploy endpoint

Reference#

projects/image-gen/runpod-setup-spec.md RunPod API docs: https://docs.runpod.ai/serverless/

Maxime's Answer#

The issue you failed previously should be over. Please try again to process this task. Thank you