Hi! I’m building a Piko that acts as a branding assistant for psychologists. It uses gpt-4o
with a detailed prompt and is supposed to generate three images during the conversation (a conceptual illustration, a logo, and a Canva-style template).
The problem is that the images generated are low quality and look like they’re from DALL·E 2, not DALL·E 3.
I have already disabled the “Use default image generation (Flux)” option as recommended, and confirmed that the image is being generated inside the Piko’s flow — but the results don’t resemble the clean, sharp DALL·E 3 images I get using ChatGPT.
Additional details:
- I’m using the
gpt-4o
model. - The prompt is identical to what I use in my working GPT on chat.openai.com.
- When I use the same image prompts directly on ChatGPT (e.g., the surreal Magritte-style illustration), it generates stunning results with clear detail and text handling — but not inside Pickaxe.
- I’ve tried switching models, toggling Flux, etc., with no improvement.
- I need these images to be downloadable and good enough to be embedded into a final PDF branding document.
Questions:
- How can I make sure my Piko is using real DALL·E 3 (like in ChatGPT)?
- Is there a way to force it to use the actual
v1/images/generations
DALL·E 3 API instead of the internal Flux renderer?
Thanks in advance — any guidance would be really appreciated!