ChatGPT image brief
Upload a reference and get a clear, editable brief for ChatGPT Images. The result states what to create or edit, which details must stay fixed, and what the finished image should deliver.
Create a clean vertical product photograph based on the reference. Preserve the overhead composition, white watch, cool gray surface, and soft silver reflections. Remove all branding and keep the watch face blank.
Purpose
catalog product image
Preserve
layout, object, light
Change
remove branding
Format
vertical 4:5 image
ChatGPT image brief
Generate the six outputs, open ChatGPT Image, and review the instruction as the first turn of a conversation: what to create or edit, what stays fixed, and what changes.
No image selected
Add an image to create all six prompt formats.General · Midjourney · FLUX · Stable Diffusion · ChatGPT Image · Nano Banana
A production request
OpenAI guidance favors clear requests with a purpose, subject, action, setting, and visual style. Add framing, lighting, materials, and constraints only when they affect the result. A straightforward request may need just one to three sentences.
This generator writes the model-specific brief from the visual choices in your reference. It identifies what the image depicts and how the composition works, then expresses the result as instructions a conversational image system can follow. The wording can say “preserve,” “replace,” “remove,” or “keep unchanged” because ChatGPT Images supports iterative creation and editing through natural language.
The generated brief is not the original source prompt. The image cannot reveal the conversation that preceded it, uploaded references, selection masks, system behavior, model version, seed, policy decisions, or later edits. The service creates a new brief from what is visible. You should review every inferred detail before treating it as intentional.
One request on this site produces six prompt versions. This version differs from the keyword-focused phrasing used for Midjourney, the relationship-heavy FLUX description, and the conditioning fields used in Stable Diffusion. It explains the task, the desired output, and the boundaries of the change in language that fits a conversational editing loop.
The first-turn brief
Purpose resolves visual decisions. Product images need label space, story illustrations need continuity, and social posts need a defined canvas and readable hierarchy.
Open with a direct verb: create, edit, replace, remove, extend, restyle, or generate. Say what the image is for when purpose affects the design. “Create a vertical catalog product image” gives the system a clearer job than “make this look better.”
Identify the visible subject, action, pose, and defining features without guessing private identity or unsupported facts. The brief should distinguish what must remain recognizable from details that can vary. Use relationship language when several people or objects appear.
Name the setting, medium, composition, camera angle, lighting, palette, material, and mood that control the result. Connect style to a usable visual treatment. “Editorial photograph with soft side light and muted plaster tones” guides more decisions than a broad label such as “professional.”
For editing, make the boundary explicit. The instruction can preserve the person, pose, product shape, composition, or color palette while changing the background, clothing, season, text, or format. Do not ask to preserve everything and overhaul the entire image in the same request.
Specify aspect ratio, transparent background, exact visible text, negative space, or destination format when required. OpenAI notes that ChatGPT Images can add text and produce transparent backgrounds. Check spelling, layout, and selection boundaries after generation because edits can extend beyond the selected area.
Preserve / change boundary
An editing brief should separate fixed elements from requested changes. “Keep the subject, pose, camera position, and lighting unchanged; replace the wall with pale green tile” creates a testable boundary. If the output changes the face or crop, you can point to a broken preservation rule instead of deciding vaguely that the edit feels wrong.
Give each turn one primary edit. ChatGPT Images supports conversation, so you do not need to solve restoration, background replacement, wardrobe, typography, relighting, and reframing in one enormous message. Start with the structural change, inspect the result, then refine text or surface detail in a follow-up instruction.
Text inside images deserves exact wording and layout context. Put the required copy in quotation marks, identify where it belongs, describe hierarchy and type treatment, and keep the amount reasonable. OpenAI guidance says clarity helps details such as layout, texture, materials, and light. Generated lettering still needs human review before publication.
Safety and authenticity boundaries remain part of the workflow. Do not use the generated text to misrepresent evidence, impersonate a person, or make a provenance claim. Follow platform policies and use content you have permission to process. The tool describes visible creative direction; it does not certify what happened in the source image.
Conversational editing
The first prompt establishes the task and its boundaries. Copy the generated ChatGPT image prompt into ChatGPT with the reference attached, inspect the result, and make later turns respond to visible output instead of restating the entire brief with new synonyms.
Decide whether the reference supplies inspiration for a new image or should be directly edited. Write that choice into the ChatGPT image prompt. A generation can borrow composition and mood. An edit should name what remains fixed and what changes on the uploaded source.
Compare the generated brief with the reference. Remove guessed identity, unsupported camera metadata, or invented text. Verify the main subject, action, framing, light, color, material, and intended format. Correct a wrong preserve rule before adding decorative description.
Use one instruction for the main operation, such as replacing a background while keeping the subject. Review the result at full size. If the system changed something protected, repeat the preservation requirement and narrow the requested edit.
Write the next turn about the actual result: reduce the headline size, restore the original crop, soften the shadow, or remove an added object. Keep the working version in History so the final sequence records what changed instead of hiding every revision inside one block of text.
Over-edit diagnosis
Conversational systems follow the instruction they receive, but broad requests leave many design choices open. Clear task boundaries reduce unintended edits and make corrections shorter.
The request has no measurable direction. Replace it with the change you can see: soften hard facial shadows, remove background clutter, create more headline space, or match the reference camera position. A ChatGPT image prompt should define the decision, not outsource taste.
A major restyle cannot keep every pixel unchanged. Name the identity, pose, product shape, crop, or layout that matters most. The ChatGPT image prompt can then allow texture, environment, palette, or medium to change without creating a contradiction.
Long compound requests make it hard to tell which instruction caused a miss. Divide the workflow into structural edit, visual polish, and text correction turns. Each ChatGPT image prompt should have one primary outcome and a short set of preservation rules.
A generated description may infer a softbox or lens feel, but it cannot verify the capture settings or original generation process. Use those details as creative direction. Do not present the ChatGPT image prompt as forensic evidence or provenance metadata.
Editing questions
Yes. A vision-capable workflow can inspect the visible subject, scene, composition, light, color, material, style, and mood, then write a new ChatGPT image prompt from that evidence. It cannot recover the exact original prompt, hidden references, model settings, or editing history.
Yes. Attach the reference, ask for a description grounded in visible evidence, and turn that description into a clear creation or editing brief. This page handles that conversion in one analysis and supplies a ChatGPT-specific version alongside the General and four other model prompts. Review inferred camera, text, and identity details before using the result.
Upload the reference here, generate the six outputs, and select the ChatGPT Image version. Check the subject, purpose, preserve rules, requested changes, visible text, and format. Then copy the brief into ChatGPT with the reference image attached and begin with one primary creation or editing task.
Include a direct task, the subject and action, setting, style, composition, lighting, important materials, output format, and exact visible text when needed. For editing, separate what must remain unchanged from what should change. One to three clear sentences can be enough for a simple request.
No. A finished image does not reveal the original ChatGPT image prompt, conversation, uploaded references, model state, selection masks, or editing history. The tool writes a new brief based on visible evidence.
Direct constraints such as “remove the logo” or “keep the background empty” are useful. Pair them with the desired result and avoid a long generic ban list. For edits, preservation and change instructions are usually easier to evaluate.
Visitors receive three daily credits. One analyzed image uses one credit and creates all six model versions. Accounts add editable History, and batch analysis is available to Pro subscribers.
Start the first editing turn
Upload the reference, choose whether you are creating or editing, and review the generated ChatGPT image prompt for task, subject, preserve rules, changes, and output format. A ChatGPT image prompt works best when one turn has one primary outcome. Keep each ChatGPT image prompt specific enough that you can tell whether the requested edit succeeded, then refine from what the image actually did.