Prompt guide for the new Leonardo Vision model

I've been curious about this model too, especially for creating concept art for our sales pages. Have you found it helpful to completely avoid certain words or phrases from other platforms? Like, terms I use a lot in Midjourney seem to get ignored or produce weird results here.

For character poses, I got the best results when I described the camera angle and the character's action separately. Instead of "a woman confidently pointing at a graph," I had to write "low-angle shot, a woman in business attire, her arm extended with index finger pointing directly at viewer, a large transparent data visualization floats in front of her." It's more like writing a shot list. Does that match your experience so far?

Also, for backgrounds, are you using negative prompts? I saw a setting for that but I'm not sure what to put in there to avoid muddy details.

Your observation about prompt structure is correct. The shot list approach is effective because the model seems to parse spatial and compositional elements separately from style descriptors. I've found it also responds well to explicit layering terms, like "foreground: subject pointing, midground: data viz layer, background: blurred office environment with large windows."

Regarding your question about avoiding terms from other platforms, yes, there is definite lexicon bleed that causes issues. Terms like "cinematic," "hyperrealistic," or "octane render" from Midjourney don't map cleanly here. They often introduce unwanted stylistic artifacts or are simply ignored. I maintain a simple exclusion list for this model that includes those common Midjourney modifiers.

For negative prompts, they are critical for background clarity. Use them to remove elements that cause visual noise. For marketing concepts, I consistently use: `blurry background, cluttered background, text overlays, watermark, distorted faces, extra limbs`. This forces the model to allocate its detail budget to the primary subject and key background objects you specify.

The exclusion list is a great idea. I've been keeping a similar note in a text file. I'd add "volumetric lighting" and "Unreal Engine 5" to that list of problematic terms - they tend to make everything look oddly shiny.

Your negative prompt list is solid for general use. For data viz concepts specifically, I've found I need to add `floating numbers, random charts` to avoid those weird, nonsensical graph elements popping up behind the main subject.

Have you noticed any particular style terms that *do* work reliably? I've had decent luck with "flat illustration style" and "corporate photography" as anchors.