Anyone else having issues with image generation bot outputs being... wildly inconsistent?

Hey folks, Bob here! 👋 I've been deep-diving into automating some social media content creation using Poe's array of image generation bots, and I've hit a real head-scratcher. While the potential is massive for hooking these into an iPaaS workflow, the actual output quality feels like a rollercoaster I didn't buy a ticket for.

I'm trying to generate a consistent style of "tech-enthusiast digital art" for a weekly blog header. Using the same, meticulously crafted prompt across bots like DALL-E 3, StableDiffusionXL, and Midjourney (via Poe), I'm getting bafflingly different results each time, even with the same bot! It's not just about artistic variation—it's like the core instructions are being interpreted from a different dictionary each Tuesday.

Here's a concrete example of my prompt structure. I've been feeding them via API calls from Make (formerly Integromat), thinking I could pipeline this:

```json
{
"bot": "DALL-E-3",
"query": "A futuristic server rack glowing with blue neon lights, placed in a minimalist, sunlit studio. Style: clean vector art with a slight grain texture. Aspect ratio: 16:9."
}
```

One day, I get a stunning, crisp vector graphic. The next, with the exact same payload, it outputs a photorealistic server room with zero neon. It's as if the 'style' directive carries a different weight each time. This inconsistency throws a major wrench into any automated workflow. How can you build a reliable Zapier alternative or webhook chain if the core content is unpredictable?

I'm particularly curious about:
* **Prompt Weighting:** Are certain bots on Poe ignoring parts of the prompt after a certain length? I've tried using "::" or "—" for emphasis with mixed results.
* **Rate Limiting & Quirks:** Could hitting different rate limits across these models cause a fallback to a less-capable version, altering output?
* **Context Window:** Is Poe's wrapper potentially truncating or reformatting our detailed prompts before sending them to the underlying model?

This feels like a bigger issue for anyone wanting to use these bots in an event-driven, automated setup. The manual "try until you get a good one" approach kills the whole efficiency gain. Has anyone else run into this? Found any workarounds or specific prompt engineering tricks that *actually* create consistency on Poe's platform?

Happy integrating,
Bob