Why Are My API-Generated DALL-E Images Lower Quality Than ChatGPT's?

Below is a likely explanation along with some steps for troubleshooting:

Why the Quality Difference?

It appears that the tool you were using via Pipedream was not actually invoking DALL‑E 3—even though it was advertised or labeled as such. Instead, it was using DALL‑E 2, which generally produces lower-quality images compared to DALL‑E 3 (especially when comparing the standard output versus the enhanced ChatGPT output).

What to Check in Your API Request

1. Model Parameter:
   Make sure your API request explicitly sets the model to the DALL‑E 3 version. For example, you might need to include:

   JSONCopied
   {
     "model": "dall-e-3",
     "prompt": "...",
     "size": "1024x1792"  // or your desired option among 1024x1024, 1024x1792, 1792x1024
   }

   Note that if you omit the parameter or if it’s being overridden, you might default to DALL‑E 2.

2. Quality Settings:
   DALL‑E 3 supports an optional quality parameter. If you want enhanced details, specify this parameter as:

   JSONCopied
   {
     "quality": "hd"
   }

   This can help you achieve the closer image consistency to what ChatGPT shows.

3. Endpoint and Parameters:
   Verify that your endpoint and parameters match the latest documentation. For DALL‑E 3, the endpoint is the same, but the supported sizes and quality options differ from DALL‑E 2:

   • Supported sizes for DALL‑E 3: 1024x1024, 1024x1792, or 1792x1024
   • Only one image (n: 1) per request is allowed.

Conclusion

The discrepancy you are seeing is most likely because the API call being made via Pipedream is inadvertently using DALL‑E 2. When you ran the same prompt via Zapier—which correctly routed your request to DALL‑E 3—the generated image quality improved.

Double-check your request configuration (especially the model parameter) to ensure you’re using DALL‑E 3, and include any optional parameters like quality: "hd" if needed. This should help align your results with what you see from ChatGPT.