Generate a clear base composition
Begin with the intended subject, environment, framing, and visual hierarchy. Test the composition in the target ratio before spending iterations on surface detail.
5CREDITS
Remaining credits：
Public
Create and Edit Images with GPT Image 2
GPT Image 2 brings text-to-image generation and reference-based image editing into one focused workspace. Start from a written brief or upload as many as 16 source images, choose a composition from square, portrait, landscape, panoramic, and extra-wide ratios, then render at 1K, 2K, or 4K. The model page keeps both workflows together so you can move from an initial concept to a directed revision without changing tools.
GPT Image 2 generation and editing controls
Use only the inputs that support the result you need: a clear prompt, optional reference images, a target aspect ratio, and an output resolution.
Text-to-image generation
Describe the subject, environment, composition, lighting, materials, and intended use. GPT Image 2 turns the brief into a new image while the workspace keeps the selected format and resolution visible.
Editing with up to 16 references
Upload one or more source images when identity, products, layout, or visual direction must come from existing material. Assign a clear purpose to each reference instead of asking every image to control the same detail.
Flexible composition formats
Choose auto sizing or common square, portrait, landscape, wide, and panoramic ratios. Set the final channel first so framing decisions match a product card, poster, thumbnail, banner, or social placement.
1K, 2K, and 4K output
Select resolution according to the review stage and delivery need. Use lower resolution for early composition tests, then move to a larger output after subject, layout, and typography directions are stable.
How to prepare an effective image request
A useful request separates what must remain fixed from what may change. Define the deliverable, organize references, and make the prompt specific enough to review.
1
Set the final use
Choose the placement, aspect ratio, resolution, audience, and visual goal before writing style details. A banner, product listing, and editorial illustration need different framing and information density.
2
Write a structured prompt
State the main subject and action first, then scene, camera or viewpoint, lighting, palette, materials, and constraints. Put required text or product details in exact wording and avoid contradictory style instructions.
3
Give every reference one job
Identify which upload controls identity, product shape, pose, composition, palette, or texture. Remove redundant or conflicting images and explain which source has priority when details disagree.
4
Review before increasing resolution
Check subject accuracy, hands and edges, text, brand details, spatial relationships, and crop safety. Change one variable per revision so the effect of each edit remains clear.
A practical creation and revision workflow
01
02
Refine with selected source images
Switch to the image-editing mode when the result must preserve a person, object, product, or layout. Upload only relevant references and describe both the required change and the details that must stay untouched.
03
Validate the delivery file
Inspect the final crop, resolution, small details, embedded text, and consistency with the original brief. Export the version that matches the actual publishing surface rather than a generic master ratio.
Input requirements and file limits
Prepare files that the current Kuca workspace can accept, and decide which details each input should control before you submit the request. These practical limits help prevent rejected uploads and make a multi-reference edit easier to direct.
01
Up to 16 input images
The editing mode accepts a maximum of 16 reference images. Reaching the limit is not a creative goal: use only sources that add necessary information. A concise set for identity, product shape, composition, and styling is usually easier to explain than many nearly identical uploads. Text-to-image mode does not require a reference image.
02
10 MB maximum per image
Each uploaded image can be no larger than 10 MB. Check the actual file size before starting a long brief, especially when working with camera originals or exported design files. If a file is too large, create a clean web-ready copy that preserves the important subject, edges, texture, and readable details without unnecessary metadata.
03
Choose 1K, 2K, or 4K output
All three output levels are available in the workspace. Use 1K for early visual direction and quick comparisons, then select 2K or 4K when the layout is stable and the destination needs closer inspection. A higher setting provides a larger output; it does not repair an unclear prompt, conflicting references, or an incorrect composition.
04
Use clear, relevant source material
Choose images in which the required person, object, product, or layout is visible enough to guide the edit. Avoid references with incompatible identities, lighting, camera angles, or brand details unless the prompt explicitly resolves the conflict. Do not upload material you lack permission to use, and review the generated result before publishing it.
Choose the right creation mode
The model page offers a text workflow and a reference-based editing workflow. Select the mode from the kind of control your task requires, rather than adding uploads to every request by default.
01
Use text mode for a new visual concept
Start with text-to-image when no existing image must be preserved. It works well for exploring subjects, art direction, environments, product-scene ideas, and composition alternatives. Describe the intended output and visual hierarchy first, then add lighting, palette, materials, camera language, and exclusions that meaningfully affect the result.
02
Use editing mode when source details matter
Choose reference-based editing when a person, object, package, room, pose, or layout should remain recognizable. State the requested change separately from the protected details. For example, identify which background may change while product geometry, label placement, color, and camera position must stay consistent.
03
Compare modes with the same delivery goal
If you are unsure which mode fits, keep the aspect ratio, output goal, and core brief consistent while testing a text-only version and a reference-guided version. Compare instruction adherence, preserved details, composition, and revision effort instead of judging only surface style. This produces a more useful model decision for repeat work.
04
Check credits in the live generator
The generator shows the current credit requirement for the selected mode and settings before submission. Treat that live value as the source of truth because model options and costs can change. Review the displayed credits after changing resolution, variant, or other controls, and submit only when the configuration matches your intended test or final render.
Where this workflow fits into image production
Product concepts
Explore product scenes, packaging directions, materials, and controlled backgrounds while keeping required proportions and brand details explicit.
Campaign visuals
Develop key art and channel-specific crops from one visual brief, then adapt the selected direction to banners, social placements, and promotional layouts.
Reference-based revisions
Change a background, styling, lighting, pose, or composition while telling the model which source details must remain recognizable.
Editorial illustration
Turn abstract topics, articles, and reports into directed illustrations with a defined point of view, visual metaphor, and publication format.
Story and concept frames
Create consistent visual checkpoints for characters, locations, props, and key moments before committing to a larger production sequence.
Large-format drafts
Move an approved composition to 2K or 4K for closer review and delivery when the chosen placement needs more source resolution.
Frequently asked questions about image generation and editing
GPT Image 2 supports text-to-image generation and image-to-image editing. You can start with a prompt or upload reference images, select an aspect ratio, and choose 1K, 2K, or 4K output.
The GPT Image 2 editing workflow accepts up to 16 input images. Use fewer, purposeful references when possible, and describe the role of each source so identity, layout, and style instructions do not compete.
The workspace offers auto sizing plus square, portrait, landscape, wide, panoramic, and extra-wide formats, including 1:1, 3:2, 2:3, 16:9, 9:16, 2:1, 1:2, 3:1, 1:3, 21:9, and 9:21.
Yes. Choose the editing variant, upload the source images, describe the requested change, and state what must remain unchanged. This is useful for controlled revisions rather than recreating the entire scene from scratch.
Not always. Early composition tests are easier to compare at a smaller resolution. Move to 2K or 4K after the subject, crop, layout, and visual direction are approved.
Lead with the subject and intended result, then specify environment, composition, viewpoint, lighting, palette, materials, required text, and constraints. For editing, also list what must change and what must be preserved.
Start creating in the image workspace
Start from a prompt or reference images, then refine the format and resolution in the Kuca image workspace.