GPT Images 2.5 Prompts and Output Examples

A Japanese creator's digital illustration made with ChatGPT Images 2.5 marks the final day of the jet-black series.

The author shares a GPT Image 2.5 prompt template that simulates nine consecutive shots of a phone falling from its rear camera's view, with random rotation, defocus and motion blur.

The author argues that closing the Sora website was a bad decision for image generation, losing both a community for browsing others' work and the preference data that could have improved GPT-Image 2.5.

A general workflow: upload a selfie to Gemini, Grok, ChatGPT or GPT Images 2.5, paste the author's prompt and generate an image.

The author shares images made with GPT-Image-2.5 Sunburst that were left out of an MV during selection, noting they show no obvious flaws.

The open-source Ming-Image-0.1 model splits a finished design into transparent RGBA layers so only the layer you want changed goes to GPT Image 2.5, avoiding full-image redraws.

The author generated a song with Suno, created images with GPT Image 2.5 and animated the shots with Seedance 2.5 to build a music video from scratch.

A long Chinese prompt for a 24mm wide-angle, water-level, high-speed photo of a woman breaking the pool surface, with negative constraints against plastic skin.

The 780th piece in Japanese creator Maruko's digital art series, a jet-black neon chibi character made with ChatGPT Images 2.5.

A test of Images 2.5 with eight visually distinct characters failed even when simply placing them side by side, suggesting seven is the limit for these characters without specific instructions.

A full Chinese long-form prompt for a high-angle, full-body fine-art portrait with pink chiffon, white roses and golden afternoon light, ready to copy and use.

The author shares a way to save API quota: in Claude Code, ask it to pass image generation tasks to Codex, which calls GPT Image 2.5 to produce the images.