A Better Image Edit Starts with What You Keep

A Better Image Edit Starts with What You Keep A practical briefing guide for creators working with AI image tools Why a clear boundary matters An image can be almost right: the subject is useful, the composition works, but the setting distracts from the message. Starting again may discard the strongest part of the picture. Before writing an editing prompt, ask two questions: what needs to change, and what would make the result unacceptable if it changed? Those answers form a more useful brief than a long string of adjectives. Write the preservation instruction first Identify the details that carry meaning. For a product photograph, these might include its silhouette, label wording, material and color. For a portrait, they might include the person's expression, clothing and pose. An instruction to preserve these details is a request, not a guarantee: review the output against the original rather than assuming the model followed every condition. Turn a vague request into a usable brief Instead of "make this look professional," try a specific, limited instruction. Illustrative prompt: "Replace the busy background with a pale, softly lit studio wall. Keep the mug's shape, handle, printed lettering and position unchanged. Keep a natural contact shadow. Do not add props or text." This example is proposed wording, not a tested result. If the lettering changes, reject that version or restore it through a suitable manual workflow. Where an editing interface fits The public workspace at Nano Banana AI supports uploading an image and describing an edit in text. That makes it a relevant place to try a clearly bounded modification. Keep the original file, save the exact instruction, and compare each result at a size where small details are visible. Avoid submitting confidential source material until you have reviewed the service's privacy and usage terms. Publisher disclosure: Prepared by xzmowei on behalf of the operators of the three websites discussed. This is a practical guide, not an independent review, peer-reviewed paper, or report of measured product performance. xzmowei | Practical creative notes | September 2026 1

Putting the brief into practice Change one variable before changing five If the first attempt is unsatisfactory, diagnose the mismatch. Did the background change correctly but the object deform? Did the lighting improve but the crop remove important space? Adjust the part of the instruction connected to the failure. A new style, angle, color palette and background in every attempt make it harder to understand which change helped. Know when editing is the wrong starting point Sometimes there is no suitable source image to preserve. In that case, an image-creation brief is more appropriate. Whisk AI provides separate subject, scene and style reference inputs alongside text instructions. Use those roles to clarify what each reference contributes. A scene reference should describe the environment, for example, rather than accidentally becoming the new subject. Use only reference material you have permission to upload. Move to motion only after approving the still A polished still can become the starting point for a short motion concept. Dola AI offers an image-to-video route as well as text-to-video. Before using the approved image as input, decide on one movement and review the resulting clip from beginning to end. A good opening frame does not establish that the subject, lettering or background remain consistent throughout. A short review that earns its place Compare the original and edited versions side by side. Check identity, edges, lettering, shadows and the intended crop. For a real product listing, do not let an attractive variation misrepresent the item being sold. Keep a small record of the source, prompt and rejected changes. The goal is not simply a more attractive image; it is an image that still communicates the right facts. Resources and scope The following links identify the websites discussed, not a ranking or a claim of affiliation with similarly named services. Features, availability and usage rights can change. Consult the current workspace and terms before using outputs publicly. Whisk AI - reference-based image creation https://whisk-ai.co Nano Banana AI - prompt-based image editing https://nanobanana-ai.org Dola AI - text-to-video and image-to-video https://dolai.video xzmowei | Practical creative notes | September 2026 2

• https://whisk-ai.co
• https://nanobanana-ai.org
• https://dolai.video