Were these examples tested in ChatGPT?
Codex’s built-in image generation tool made the ten images. Each detail gives the original prompt. The same instructions were not tested in the ChatGPT interface.
Examples with complete prompts
Compare ten finished Christmas images, from portraits and pets to baking, a snowy porch and a glass ornament.
Codex’s built-in image tool generated the ten festive scenes from text. Each detail includes the original English prompt, a Chinese translation and notes on the actual image.
Curated prompts
Christmas preparations and details
Give the hands distinct jobs in the gingerbread scene. Balance blue snow and warm lights on the porch, and use narrow reflections to show the glass ornament’s texture.
Choose by visual goal
The window portrait balances cool daylight with warm tree lights; the couple scene mixes blue hour and shop windows; the family scene gives four people distinct positions. State the count and arrangement before adding decorations, then inspect faces, hands and overlapping limbs.
Choose by visual goal
The pet example uses a low viewpoint, a red bandana and soft tree lights. The actual dog lies down although the original instruction requests sitting; the detail records that difference. The interior balances dusk outside with warm practical lights. Inspect verticals and highlights rather than filling every surface with decorations. The tabby example uses a low viewpoint to show a paw reaching toward a ribbon.
Choose by visual goal
Cluster decorations at lower left and keep the upper and right areas open. Generate the background first, then add names, dates or a greeting in a layout tool. The example contains no lettering and is not a preflighted print file.
From selection to revision
Each detail retains the complete prompt, input requirements and example context. Decide on a goal and revise one part at a time; an example does not guarantee identical output across models.
All six examples create new scenes from text. To retain a real person, use a clear authorized reference and specify which surroundings or clothes may change. That editing workflow was not tested by these examples.
Use a window or blue hour for cool light and a tree or shop window for warmth. Specify what lights the face. If skin looks unnatural, reduce colored lighting before adding more sources.
Check person counts, fingers and gaze in portraits, pose and silhouette for pets, and usable empty space for cards. Keep the original prompt version and revise one concrete problem at a time.
Codex’s built-in image generation tool made the ten images. Each detail gives the original prompt. The same instructions were not tested in the ChatGPT interface.
You can try a separate reference-photo edit with suitable inputs and inspect the result. This collection includes no real-person inputs, likeness tests or paired before-and-after comparisons.
Find more examples and complete prompts for your next project.