GPT Image 1.5 Text To Image: High-Fidelity Text-to-Image Generation on playground and API | RunComfy

What makes GPT Image 1.5 Text To Image stand out

GPT Image 1.5 Text To Image is a premier generative engine designed to construct high-fidelity visuals entirely from natural language. Unlike image editing tools that modify existing pixels, this model excels at "creation from scratch," translating complex semantic instructions into coherent, photorealistic, or stylized images. It leverages advanced language understanding to accurately render specific layouts, typography, and object relationships defined purely by your text.

Prompting guide for GPT Image 1.5 Text To Image

Since there is no reference image, your prompt is the only blueprint. Start by defining the Subject, Medium (e.g., photo, 3D render, oil painting), and Environment. In GPT Image 1.5 Text To Image, be specific about the composition and lighting, as the model will build the scene exactly according to your adjectives. Set image_size to match your intended composition (e.g., wide for landscapes). Use quality='hd' when fine details like text or textures are critical.

Examples:

• Photorealism: "A futuristic eco-city, biophilic skyscrapers with vertical gardens, golden hour lighting, shot on 35mm lens, high detail." (Set quality=hd)
• Marketing Asset: "A minimalist 3D icon of a rocket ship launching, isometric view, gradient blue background, high gloss finish." (Set background=transparent)
• Typography/Logo: "A neon sign on a brick wall that says 'OPEN LATE' in pink script font, cinematic lighting, dark atmosphere."
• Layout Control: "Split screen composition: left side shows a chaotic city, right side shows a peaceful forest, high contrast style."

Pro tips:

• Be the Director: Since you are generating from scratch, explicitly describe camera angles (e.g., "drone view", "macro shot", "eye level").
• Describe the "Vibe": Use mood keywords like "melancholic", "vibrant", "sterile", or "cozy" to guide the color palette.
• No "Preservation" Needed: Do not use terms like "keep the face" or "change the background"; simply describe the entire scene you want to see.
• Iterate with Variations: If the composition isn't right, change the prompt (e.g., move "cat" from "left" to "center") rather than trying to edit the previous seed.

Note: If you need to modify an existing image (e.g., changing the background of a photo you already have), do not use this model. Instead, switch to the GPT 1.5 Image Edit model which is specialized for instruction-based manipulation.