Start with one source image, describe one change and name what should stay. Learn to choose a ratio, compare results and revise without carrying unwanted changes forward.
LumiYingUpdated
Good image editing prompts separate what must stay fixed from what should change. This makes each generation easier to evaluate and revise.
Begin with one clear original. If you add a second image, explain which image supplies the subject and which supplies a style or setting.
Specify the face, product outline, exact label text, fabric or camera view that matters. These are instructions to check against the result, not a guarantee that every detail stays identical.
Separate background replacement, recoloring and lighting into different trials. A failed result is easier to diagnose when only one change was requested.
Save the source separately. Use it again to compare alternative directions; use an edited result as a new reference only when its existing changes are worth keeping.
Prepare the input, set the output and compare the result before starting another generation.
Upload a source with readable details and enough space around the subject. Decide on one change, such as a blue background or burgundy fabric. A blurred label or hidden edge cannot be reliably recovered by asking the model to preserve it.
Name the target, the change and the details to keep. For the bottle example below: change only the background and surface to pale blue; keep the cap, bottle outline and exact label text. Avoid adding cleanup, relighting and restyling to the same first trial.
Select the output ratio in the controls; writing 16:9 or 9:16 in a prompt does not change that setting. Start with one output at an available draft resolution. Check the displayed credit quote before generating, and keep the same settings when comparing prompt revisions.
First check whether the requested change happened. Then inspect the details you named: label letters, face, hands, silhouette, edges, shadows and crop. Reject an attractive result if a critical detail changed. Preservation rules do not create a pixel-locked selection.
For a different background or color, return to the original so earlier changes do not accumulate. For a small correction to an accepted edit, use that output as the next reference and name the correction. Save an approved still before using it as a video start frame.
These AI-generated examples show the edit and the details to inspect. Load a source image, prompt and suggested ratio, then review the settings before generating your own version. Results vary by image, prompt and model.
Change the teal fabric to burgundy. Compare the face, hands, boots and embroidery as well as the new color: small brightness or material changes still need review.
Keep the face, pose, hands, embroidery, accessories and grey background unchanged. Change only the tunic and cloak fabric from teal to deep burgundy.
Start again from the original teal-outfit image, not the burgundy result. Compare the character outline and the contact shadow after replacing the grey background with a courtyard.
Keep the character, teal outfit, pose and lighting unchanged. Replace only the plain grey background with a softly blurred stone castle courtyard.
Change only the background and supporting surface. Read LUMI SERUM 30ml letter by letter, then compare the cap, glass edges and shadow; a correct blue backdrop alone is not enough.
Keep the bottle, cap, label text "LUMI SERUM 30ml" and lighting unchanged. Change only the background and surface to pale powder blue.
Replace the bracketed details with visible objects in your own image. Choose the matching output ratio in the controls, then check the result against your preservation rules.
Use the uploaded image as the main subject. Keep the subject shape, color, and edges unchanged. Replace only the background with [new scene], using [lighting style] and a clean commercial composition.
Best for product listings, campaign concepts, thumbnails, and cleaner social visuals.
Keep the product shape, label text, color, lighting and framing unchanged. Remove only [specific distraction] and reconstruct its immediate background. Do not move, relight or restyle the product.
Keep cleanup separate from a new background or lighting pass so you can tell which change caused a problem.
Use the uploaded image as the composition reference. Preserve the subject and layout, but apply [visual style], [color palette], and [mood]. Do not add text or extra logos.
Useful for moodboard variants, social campaigns, and art direction tests.
Turn this image into a clean first frame for a 9:16 AI video. Keep the main subject clear, add [cinematic lighting], leave room for motion, and avoid text overlays.
Select 9:16 in the ratio control as well. Check the generated still first, then use the accepted image as a start frame in image-to-video; this prompt does not generate a video.
Inspect the failure before changing the prompt or spending credits on another pass.
Use these links to move from prompt planning into image editing, model choice, and video continuation.
Upload a reference image and generate prompt-based edits in LumiYing.
Learn model-specific prompt and reference-image workflows.
Turn an approved edited still into a short AI video clip.
Create consistent scene references after editing image assets.
State what the uploaded image is, what must stay unchanged, and what should change. A focused prompt is easier to review than a long prompt with unrelated edits.
They can help with background replacement, lighting changes, cleanup, scene staging, and creative variants. For ecommerce images, explicitly preserve product shape, label, and color.
Keep one primary edit per generation, include preservation rules, and revise one variable at a time. If the background failed, change the background instruction without rewriting the whole prompt.
Yes. Once the edited still looks clean, use it as a start frame, storyboard panel, product keyframe, or thumbnail before moving into image-to-video generation.
No. The editor generates a new image from references and instructions; it has no brush, mask or region selection. Faces, text, materials and areas outside the intended edit may change, so compare them with the original.
Loading fills the source image, prompt and suggested ratio. Review the model, settings and displayed quote before generating; loading alone does not submit a generation.
Use the original to try an alternative direction. Use an accepted result when you want to keep its changes and adjust one detail. Repeatedly editing a flawed result can carry its errors into later versions.
Open the AI Image Editor, upload a reference image, and write the first focused edit prompt.
