Supported models for styled slide decks in v0.8?

[Affenjimbo]

Hi there! I installed version 0.8 yesterday and have been testing it out. I initially wanted to try it with Gemini, but I'm getting a 'Base URL' error.

Today, I tried switching to Deepseek. The API seems to connect and I do receive text answers, but as soon as I select a slide deck and choose a style, I stop getting responses.

Before I add more funds to different API accounts for testing, could you let me know which model currently works best for generating a slide deck with a specific style? Thanks so much for your help!

[richardchen874-sys]

This is a good question to ask before adding funds to multiple API accounts.

For styled slide-deck generation, I’d test models with a very small deck first instead of comparing them only by normal chat responses. The important checks are usually:

• whether the model can follow a structured slide schema
• whether it keeps the requested style consistent across slides
• whether long deck context causes timeout or empty responses
• whether the provider supports the endpoint / payload format expected by the app
• cost per successful deck, not just cost per token

Since DeepSeek is returning normal text but stops after selecting a deck + style, the issue may be related to structured output size, context length, timeout, or provider compatibility rather than the model being unusable.

For independent builders testing AI design tools, I’d start with a 2–3 slide deck, one simple style, and logging enabled, then compare DeepSeek / Gemini by success rate, latency, and cost per completed deck.

I’m working around lower-cost OpenAI-compatible API access for independent AI builders, so this kind of model compatibility and cost-per-output testing is exactly the area I’m interested in.