The tested target–change–keep pattern
Replace the white cup on the desk with a vase of red roses. Keep the person, laptop, plants, composition and other objects.
Edit my photo with text ↗Photo editing guides
Edit photos with text instructionsDescribe the change. Start from your own photo.
Write what you want to change instead of painting an outline. A clear instruction can turn the cup on a desk into a vase of red roses. Upload a photograph and use Text instructions in the editor. Your image is the starting point, not merely a style reference for generating an unrelated picture; the result can nevertheless change details across the photo.
One photo at a time · No installation · Check credits before generating
Original
AI editedFrom your photo to a reviewed edit
- 01
Upload one photo
Choose a JPG, PNG or WebP, or reuse one from recent uploads. Keep a copy of the original.
- 02
Edit photos with text instructions
Tell AI what to change and what to keep. Here, the cup becomes a vase of red roses. No painting is needed; this mode can change details across the photo.
- 03
Review credits and generate
The floating bar shows the estimate and balance. You need to sign in and have sufficient credits. Once submitted, you can submit another task.
- 04
Compare and download
Check the result against the original, inspect edges and faces, and download the version you want. Saved tasks are available in My Works.
A starting instruction you can edit
A shorter, more specific object request
Replace the white cup at the front of the desk with a short glass vase of red roses in the existing window light. Keep the person and computer unchanged.
Edit my photo with text ↗Use words when the change is easy to describe
Text instructions suit a specific object change or an overall visual adjustment you can explain clearly. They avoid the need to trace a shape, but do not give the same region protection as a brush edit. If two nearly identical objects are close together or a face must stay outside the edit, brush replacement offers a clearer spatial boundary. If you want to invent an entirely new image, start with the image generator rather than disguising that task as photo editing.
Identify the target before describing the result
Name the object and its location: the white cup on the desk is more specific than that thing. Follow with the desired action and appearance: replace it with a short glass vase of red roses. Avoid several contradictory changes in one paragraph. Clear words help the model interpret your goal but are not a guarantee of exact obedience. For multiple objects of the same kind, spatial wording can remain ambiguous; switch to a brush region if choosing the exact one matters.
Say what should stay in the photograph
A useful instruction has a target, a change and a keep clause. Mention the person, laptop, plants and composition when those are the scene's anchors. Keep instructions concise enough that they do not contradict the requested change. Saying retain everything is not equivalent to locking pixels: this mode can regenerate details outside the named target. Where factual text, a familiar face or an exact product needs stricter control, preserve the original and choose a more local workflow.
Turn a vague prompt into an actionable request
Make it nicer does not say whether to change the prop, light, colour or setting. Replace the white cup with a small glass vase of red roses, lit by the existing window identifies both the edit and how it should fit. A shorter prompt can be better than a long string of cinematic adjectives. Start with one meaningful request, compare the result, and then refine the part that missed your goal. Repeating the same words is not a reliable way to correct an ambiguous target.
Read the entire output, not only the new flowers
First confirm that the cup became a vase and that the roses fit on the table. Then compare the face, hands, hair, laptop outline, notebook writing and background plants with the source. An appealing replacement can distract from changes elsewhere. Check at both normal viewing size and a closer view. If you need a local change without broader reinterpretation, start again from the original in Brush and replace instead of chaining another global instruction onto an already altered face.
Do not confuse text instructions with adding typography
Writing a prompt tells the editor what to do; it is not a guarantee of exact typesetting. If you request lettering inside an image, compare every character and its layout before using it. This workflow is also different from a deterministic exposure slider or a layered design document: an instruction can create new pixels, not just adjust existing colours. Keep important brand copy, measurements or evidence in a source file that you can validate independently.
Keep alternatives organised
Write down the instruction alongside the original so you know what you asked for. Each submission creates its own task; it does not automatically revise a previous generation for free. Start from the same original when comparing two versions, otherwise differences can accumulate. Saved results can be viewed in My creations. A visually successful attempt is still a single example, not a measured accuracy rate or evidence that another customer's photograph will behave the same way.
Use the edited image for the right purpose
Text-driven edits are useful for personal creative work and exploring an idea. Do not present invented props, altered defects or changed identifying details as an unedited record of reality. Use authorised input photographs and keep an original when the image documents an event, person or product. For an artistic rendering without a content change, compare the photo filters; for a transparent asset or a larger canvas, use the separate background or image-extender tools rather than assuming this mode provides those outputs.
Before you save
- The intended target, not a similar object, changed.
- The requested object fits the perspective and lighting.
- Faces, hands, labels and other scene anchors were compared with the source.
- Any meaningful alteration is disclosed where the photograph documents reality.
A clearer input gives you more control
- Choose a JPG, PNG or WebP, or reuse one from recent uploads. Keep a copy of the original.
- Tell AI what to change and what to keep. Here, the cup becomes a vase of red roses. No painting is needed; this mode can change details across the photo.
Generation requires login and sufficient credits. Check the estimated cost before each submission; extra attempts are separate tasks. Download access follows your plan. Uploaded material is processed by the generation service. Saved tasks can be deleted from your account; see the privacy policy for retention and deletion details.
Privacy and deletion ↗Questions about this editing method
Do I have to brush an area?
No. Text instructions work from the photo and your written request. That convenience comes with a broader editing scope. Use Brush and replace when a specific region should control what can change.
Is this a chat with automatic follow-up edits?
Each submitted instruction is a separate editing task, not an unlimited chat session or an automatically included revision. Check the cost before another submission and choose the starting image deliberately.
Can the prompt lock the person's identity?
A keep clause asks for continuity but cannot promise identical facial features or unchanged pixels. Compare the person against the original. For a local prop change near a face, prefer a brush region that excludes the face rather than relying on a longer prompt.
Describe the change. Start from your own photo.
One photo at a time · No installation · Check credits before generating