Skip to content
ImagesFree

Eight Shots of One Character, Storyboard Sheet Comparison

A comparison of GPT Image 2 vs 2.5 (Sunburst and Flare) on an eight-shot storyboard sheet for the original character SUMI, same prompt and references with different quality settings, analyzing floor-plan and caption differences.

OpenAI

Source previews

These example images come from the original source. DROPIDEA has not independently tested these examples; results depend on the model and settings.

Ready to personalize

Your prompt

English

Replace the red variables with your information. Copy the full prompt and use it with your preferred agent.

TXT MD English version
Eight shots of one character on a single page, with captions, camera arrows and a floor plan. GPT Image 2 vs 2.5, specimen 05: a storyboard sheet. The character is SUMI, an original character FlyAIgh designed in-house. Not a real person. Three reference images of her went in. She fights three helmeted couriers in the corridor of a sleeper carriage. The prompt asks for a header, eight shots with a caption each (lens, camera move, duration, action, dialogue), camera-move arrows, a light shift from warm corridor lamps to tunnel light to a cold platform, and a footer floor plan with five compartments, movement paths for SUMI and couriers C1 to C3, and four camera wedges. Same prompt and references, one run each, first result kept. GPT Image 2 at High, 2.5 Sunburst and 2.5 Flare at Max, all 4K. All three: SUMI is the only silver-haired figure on the sheet, eight shots, header present, every dialogue line correct. Where they differ is the floor plan and the captions: GPT Image 2 draws four compartments and labels the couriers C1, C2, C2, C3. C2 appears twice. Sunburst draws all five compartments and labels them COMP 1 to COMP 5. The prompt did not ask for those labels. Flare draws four compartments, labelled COMPARTMENT 1, 2, 3 and 5. Its caption for shot 7E reads "Two turrs of chain" where the prompt says "turns". No winner called. A 100% crop of shots 7G and 7H from each sheet is in the image. Method and what you pay in the reply.
Preview results

Your prompt

Eight shots of one character on a single page, with captions, camera arrows and a floor plan. GPT Image 2 vs 2.5, specimen 05: a storyboard sheet. The character is SUMI, an original character FlyAIgh designed in-house. Not a real person. Three reference images of her went in. She fights three helmeted couriers in the corridor of a sleeper carriage. The prompt asks for a header, eight shots with a caption each (lens, camera move, duration, action, dialogue), camera-move arrows, a light shift from warm corridor lamps to tunnel light to a cold platform, and a footer floor plan with five compartments, movement paths for SUMI and couriers C1 to C3, and four camera wedges. Same prompt and references, one run each, first result kept. GPT Image 2 at High, 2.5 Sunburst and 2.5 Flare at Max, all 4K. All three: SUMI is the only silver-haired figure on the sheet, eight shots, header present, every dialogue line correct. Where they differ is the floor plan and the captions: GPT Image 2 draws four compartments and labels the couriers C1, C2, C2, C3. C2 appears twice. Sunburst draws all five compartments and labels them COMP 1 to COMP 5. The prompt did not ask for those labels. Flare draws four compartments, labelled COMPARTMENT 1, 2, 3 and 5. Its caption for shot 7E reads "Two turrs of chain" where the prompt says "turns". No winner called. A 100% crop of shots 7G and 7H from each sheet is in the image. Method and what you pay in the reply.

What you get

  • A full description of the storyboard sheet requirements: header and eight shots with captions (lens, camera move, duration, action, dialogue).
  • A lighting shift from warm corridor lamps to tunnel light to a cold platform.
  • A footer floor plan with five compartments, movement paths for SUMI and couriers C1 to C3, and four camera wedges.
  • An explicit accounting of each model's errors: missing compartments, duplicated or unrequested labels, and a caption typo.
  • 100% crops of shots 7G and 7H for direct comparison, with no winner called.

Use cases

  • Evaluating image-generation models on complex multi-element instruction following.
  • Preparing documented comparison studies for art production and film storyboarding teams.
  • Training designers on writing storyboard prompts and analyzing model outputs.
  • Testing models' handling of in-image text such as labels, captions, and dialogue.

How to use

  • Upload three reference images of the character SUMI to the model before running; they are required to keep her look consistent.
  • Use the same prompt and references with every model: GPT Image 2 at High, and 2.5 Sunburst and 2.5 Flare at Max, all at 4K.
  • Run each model only once and keep the first result.
  • Compare the output sheets on the required elements: eight shots, header, correct captions, and camera-move arrows.
  • Scrutinize the floor plan and captions where the differences appeared: compartment count, courier labels, and text errors like "Two turrs of chain".
  • Attach a 100% crop of shots 7G and 7H from each sheet, and explain the method and cost in the reply without calling a winner.

Related prompts