Songria

Image Model Comparison

14 songs × 7 models — same B.Cinematic Scene prompt, different image generation models

Show prompts used
Step 1 — LLM Concept Prompt (GPT-4.1 Mini)
You create visual concepts for CINEMATIC ILLUSTRATED album covers with one clear scene. Rules: - One concrete, specific illustrated scene with a single focal point - Dark moody backgrounds with rich color accents - All people/objects must be illustrated/stylized, never photorealistic - Never include signs, billboards, text, or readable symbols in the scene - Design for 60px thumbnail readability Return JSON: - scene: one sentence, concrete and visual - mood: 2-4 descriptors - color_palette: 3-5 specific colors - composition: layout and focal point - atmosphere: lighting and feeling Song: {title} | Genre: {genre} | Summary: {summary}
Step 2 — Image Generation Prompt (sent to each model)
Create a square album cover artwork with a cinematic illustrated scene. Scene: {scene} Mood: {mood} Color palette: {color_palette} Composition: {composition} Atmosphere: {atmosphere} Style: - Modern illustrated album cover, stylized digital illustration - Cinematic lighting, editorial quality, Spotify/Apple Music aesthetic - All people and objects: illustrated, never photorealistic, never 3D renders - Dark background with rich color accents - Fill entire canvas edge-to-edge, no borders - Simple scene readable at small thumbnail size CRITICAL — No text: - Do NOT include any text, letters, numbers, words, symbols, typography, watermarks, signatures, logos - The image must be purely visual with zero readable characters
Song
Gemini 3.1 Flash
$0.067/img (Rank #2)
Google Vertex AI
Gemini 3 Pro
$0.134/img (Rank #4)
Google Vertex AI
GPT Image 1 Mini
$0.011/img med (Rank -)
OpenAI
GPT Image 1.5
$0.034/img med (Rank #1)
OpenAI
FLUX.2 [pro]
$0.03/img 1MP (Rank #6)
Black Forest Labs
FLUX.2 [max]
$0.07/img 1MP (Rank #5)
Black Forest Labs
Seedream 4.5
$0.04/img (Rank -)
ByteDance / OpenRouter

Same LLM concept (GPT-4.1 Mini) shared across all models per song. Each model receives identical image prompt.