How to use GPT Image 2 in CreateForge

Use text-to-image when the composition can be invented from a written brief. This is appropriate for concept art, editorial scenes, product mood images, and campaign directions that do not need an existing subject to remain recognizable. Start with the subject and action, then add environment, composition, light, material, color, and finish in that order.

Use image-to-image when the task begins with an existing object, layout, character, or visual language. Upload only references that have a clear purpose and state that purpose in the prompt. Tell GPT Image 2 which reference defines identity, which defines style, and what may change. Unexplained references can create conflicting instructions instead of more control.

• Text-to-image is the cleaner starting point for a new composition or visual concept.
• Reference editing is better when recognizable structure or art direction must carry forward.
• Use the fewest references needed to communicate the constraint clearly.

Open with the primary subject and the action or state that matters. Follow with the camera position, framing, environment, lighting, surfaces, color relationships, and final medium. If the image must leave room for copy, describe where that negative space belongs. If text appears inside the image, provide the exact words and request a legibility check during review.

Avoid stacking contradictory styles or asking for several unrelated scenes in one image. GPT Image 2 can interpret detailed language, but detail is useful only when it establishes priority. Put non-negotiable constraints early, group related instructions, and remove adjectives that do not change a reviewer’s decision about whether the output is usable.

Choose the aspect ratio for the final placement rather than generating a generic square and hoping to crop it later. A landscape banner needs horizontal movement or distributed subjects, while a vertical placement needs a composition that survives narrow framing. Mention that composition in the prompt as well as selecting it in the control.

Resolution changes both the intended delivery and the current quote, so select it after deciding how the asset will be reviewed or published. A concept round may not need the same setting as a final campaign image. The current model facts panel on this page is read from the production Catalog; the quote in the generator remains authoritative for the exact request.

Check the result against the original acceptance list: subject identity, number of objects, spatial relationships, hands and faces where relevant, text, material behavior, and crop. If one area fails, revise that instruction while keeping successful parts stable. Replacing the whole prompt after every attempt makes it difficult to learn what the model actually misunderstood.

A successful CreateForge result is saved to the private Library after reconciliation, so use the saved asset as the comparison source. Record the prompt change that produced a material improvement. That small practice turns GPT Image 2 from a sequence of disconnected experiments into a repeatable image production workflow.

A provider success is the beginning of review, not automatic approval. Inspect the image at its intended display size and then at full resolution. Check spelling, small symbols, repeated objects, fingers, product geometry, reflections, edge transitions, and any area that a reference image was supposed to protect. Record one concrete defect at a time. A revision such as keep the approved composition and correct only the label alignment gives the next attempt a clearer boundary than asking for a generally better result.

For client or team work, save the accepted Library asset together with the prompt, selected resolution, source permissions, and a short approval note. Export in the format required by the downstream layout rather than repeatedly regenerating for a file conversion. This handoff record makes it possible to explain why one version was chosen, recreate the working direction later, and separate model quality from changes introduced during crop, compression, typography, or final color preparation.