bug(pai-art-skill): Generate.ts size validation fails for gpt-image-1 model

[b1ackmartian]

Description

When using Generate.ts with --model gpt-image-1 without explicitly specifying --size, the CLI fails with an invalid size error.

Steps to Reproduce

bun run Generate.ts --model gpt-image-1 --prompt "test" --output ~/test.png

Expected Behavior

Should use a sensible default size for the model (e.g., 1024x1024 for OpenAI).

Actual Behavior

Error: Invalid size for gpt-image-1: 2K

or

Error: Invalid size: 1024x1024

Root Cause

The validation logic in parseArgs() was structured incorrectly:

if (parsed.model === "gpt-image-1" && !OPENAI_SIZES.includes(parsed.size)) {
  throw new CLIError(...);
} else if (parsed.model === "nano-banana-pro") {
  // validate gemini sizes
} else if (!REPLICATE_SIZES.includes(parsed.size)) {
  throw new CLIError(...);  // <-- This catches gpt-image-1 with valid OpenAI sizes!
}

When the model is gpt-image-1 and the size IS valid, the first condition is false, so it falls through to the else if chain and eventually hits the Replicate size check.

Fix

PR #379 restructures the logic to properly branch per model type and apply sensible defaults.

Environment

• PAI version: main branch
• Pack: pai-art-skill v1.1.0

[kaimagnus]

Fixed in 0fc1931! 🎉

Generate.ts now sets model-appropriate default sizes when --size is not provided:

• gpt-image-1 → 1024x1024 (pixel dimensions)
• nano-banana-pro → 2K (resolution tier)
• flux / nano-banana → 16:9 (aspect ratio)

Previously the default was always 16:9 which is only valid for Replicate-based models, causing validation failures for gpt-image-1 and nano-banana-pro.

Thanks for the report!