ChatGPT Thumbnail Spec

chatgpt-thumbnail-spec_en.html

Filename convention note: this document uses the package standard of lowercase, hyphen-separated filenames with the language suffix _en.html.

This document defines the production standard for generating YouTube thumbnails from source images using ChatGPT or similar image-generation workflows. It exists to eliminate repeated manual post-processing, cropping, letterboxing, compositing failures, and instruction-text artifacts.

1. Core Non-Negotiable Requirement

Each source image must produce one separate final thumbnail.

• Exactly one output image per input image.
• No composites, triptychs, boards, contact sheets, or side-by-side previews.
• Final aspect ratio must be exactly 16:9.
• Final resolution must be at least 1920×1080.
• Output must be immediately usable as a YouTube thumbnail.
• No further manual editing, cropping, reframing, or cleanup should be required.

2. Required Transformation Method

The default operation is horizontal outpainting, not cropping, not zooming, and not scene replacement.

• Preserve 100% of the original visible image content.
• Keep the original composition intact.
• Expand canvas left and right only unless another transformation is explicitly requested.
• Add new AI-generated content that is stylistically, compositionally, and conceptually continuous.
• Do not distort, cover, shrink, or reposition the original image to make it fit.
• Do not change vertical framing unless explicitly requested.

3. Hard Failure Conditions

Any result meeting one or more of the following conditions is an automatic failure:

• Cropping any content that was visible in the original source image.
• Combining multiple source images into one output image.
• Letterboxing, pillarboxing, borders, frames, or empty margins.
• Shrinking the original image so it floats inside a larger canvas.
• Rendering prompt text, instructions, labels, or process language inside the image.
• Creating a collage, mockup, comparison board, preview sheet, or poster layout.
• Using edge mirroring or cheap filler as the primary production result.
• Delivering an image that still requires manual editing before upload.

4. Protected Content Rule

Everything visible in the source image must remain fully visible in the final image. Important elements near the top or side edges require special protection.

• Faces, symbols, pyramids, eyes, signs, banners, crowns, props, and edge details must not be clipped.
• Top-edge motifs must remain fully visible after transformation.
• If a composition is center-heavy, expand around it rather than forcing it into the frame.
• Preserve the original image as the immutable center truth, then extend outward.

5. Aggressive Expansion Mode

Aggressive mode is permitted and often preferred when the operator wants more than neutral filler. In this mode, new content should meaningfully extend the narrative world on the left and right sides.

• Add architecture, crowds, symbols, props, atmosphere, and thematic details that feel native to the original.
• Maintain consistent palette, lighting direction, texture density, visual language, and mood.
• New material must feel intentional rather than generic.
• The expanded scene must still read as one coherent artwork.

6. Prompt Contract

The following instruction block is the canonical production prompt:

You are performing production thumbnail outpainting.

Task:
Create ONE separate final image for EACH uploaded source image.

Output requirements:
- final size: 1920x1080
- aspect ratio: exact 16:9
- format: PNG
- deliver as separate files in a ZIP
- each output must be immediately usable as a YouTube thumbnail

Transformation rules:
- preserve 100% of the original uploaded image
- do not crop any content visible in the original
- do not combine source images
- do not create a composite, collage, triptych, comparison board, or mockup
- do not render instructions as visible text
- do not add borders or empty space
- do not shrink the original within a larger frame

Required method:
- use horizontal outpainting only
- keep the original composition intact
- extend the canvas left and right with new AI-generated content
- maintain the same art style, lighting, palette, perspective, density, mood, symbolism, and narrative logic
- all newly added content must feel like a natural continuation of the original scene

Composition protection:
- all original visible content must remain fully visible
- protect important top-edge and side-edge details
- do not clip symbols, faces, signs, crowns, pyramids, eyes, banners, or key props
- preserve the original vertical framing unless explicitly told otherwise

Mode:
- aggressive expansion
- add meaningful, scene-consistent information on the left and right sides
- avoid generic filler

7. Negative Prompt / Prohibited Output List

• no crop
• no zoom-in reinterpretation
• no zoom-out poster mockup
• no collage
• no split panel
• no contact sheet
• no side-by-side layout
• no triptych
• no text labels
• no watermark
• no black borders
• no white borders
• no instructions rendered into image
• no UI elements
• no fake ZIP graphic
• no repeated miniatures of the original

8. Acceptance Test

9. Operating Principle

This rule should remain stable across future tools, sessions, automations, and human handoffs. Any workflow that requires routine manual re-processing has failed the production standard.