ArtSpark — AI Creative Studio for Reverse-Engineering Visual Designs

Most AI art tools require users to already know complex prompting syntax (lighting models, lens types, aesthetic modifiers). When users see an inspiring artwork, they have no systematic way to deconstruct its visual parameters into a reproducible prompt.

Engineered ArtSpark to automate prompt alchemy across three tiers of detail (Simple, Detailed, Ultra-Detailed) and provide multimodal image reverse-engineering using Google Gemini 2.5 Flash, paired with dependable generation endpoints.

Next.js 14 App Router frontend interacting with a FastAPI backend that orchestrates Gemini 2.5 Flash for vision/prompt refinement and delegates generation to Freepik (Flux) with Pollinations.ai fallback.

• Architected prompt expansion pipeline generating 3 granular tiers (Simple, Detailed, Ultra-Detailed) with selectable mood and lighting presets
• Built multimodal vision endpoint in FastAPI utilizing Gemini 2.5 Flash to reverse-engineer prompts from uploaded images
• Implemented automated fallback orchestration between Freepik Flux API and Pollinations.ai to guarantee generation availability
• Designed responsive studio interface with dark theme aesthetics, custom sliders, and smooth Framer Motion transitions

async def expand_prompt(base_prompt: str, tier: str, style: str) -> str:
    system_instruction = f"Transform user ideas into {tier} prompts with {style} aesthetic."
    response = await client.aio.models.generate_content(
        model="gemini-2.5-flash",
        contents=f"Subject: {base_prompt}",
        config=types.GenerateContentConfig(system_instruction=system_instruction, temperature=0.7)
    )
    return response.text