Here at Global Symbols we are using the FLUX.1-schnell model.

We use custom LoRA techniques on top of the FLUX.1-shnell output to draw the symbol in a style that matches existing symbol sets. These are unique to Global Symbols. 

Its weakness is similar to other models: vague prompts often produce decorative or busy images. This is why prompts need to be precise.

Also, Flux.1 Schnell does not work with negative prompts. You need to describe what you want to appear in the image, and not what you want to omit.

Example

• Prompt with negative items: child eating, no background, no extra objects, no text
• Prompt with positive items: child eating at a plain table, simple white background, single bowl