--- title: "Edit Images with Text: Write Requests That Work" url: https://www.imagequarry.com/edit-images-with-text/ last_updated: 2026-10-20 --- # Edit images with text, one clear sentence at a time To edit images with text, you write the change as a sentence and the tool applies it to the picture: "make the coat dark green and keep the face, hands and crop." How well that works depends mostly on the sentence. In ImageQuarry you select a version, type the request, and get a new version, with the earlier one kept, so you can rewrite a request that missed and send it again from the same starting point. A good request names the target, the change and what must stay. Write your first one that way. Name the target. Name the change. Name what stays. ## What editing an image with text means Editing an image with text means changing a picture by writing an instruction instead of operating tools. The instruction is often called a prompt, though an edit prompt differs from a generation prompt. A generation prompt describes a whole picture from nothing. An edit prompt describes a difference: it assumes an image already exists and says what about it should change. That makes edit prompts shorter but more exacting. They need a target the tool can find, a change it can act on, and often a boundary that tells it what not to disturb. The text has to carry everything a hand on a mouse would otherwise do, including where. Phrases like "the left sleeve", "behind the bottle" and "only the sky" do the job a selection tool used to. In ImageQuarry the request stays attached to the version it produced, so you can always see which wording made which result. _Editing images with text is not adding a caption or overlay text to a picture. If you want lettering on an image, that is a separate job, and AI tools vary in how cleanly they draw letters._ ## Weak requests and stronger rewrites - **From vague to directional**: "Make it better" gives the tool no direction. Say which quality to move and which way. - **From ambiguous to targeted**: "Change the chair" fails when there are three chairs. Pick one out by position or color. - **From a list of fixes to one fix**: Five changes in one sentence make it hard to tell what worked. Send the first one alone and queue the rest. - **From negative to positive wording**: "No harsh light" names what you dislike. Describing the light you want gives the tool a picture to aim for. - **Adding a keep clause**: When the image is mostly right, say which parts must survive the edit. ## How to edit with text in ImageQuarry 1. **Start from an image** Upload a picture or describe one to generate. Either becomes the version you edit. 2. **Select a version and write the request** Pick the version, then type the change: target, direction and, where it matters, what to keep. 3. **Read the result against the wording** A new version appears with your request beside it. Check whether each part of the sentence was followed. 4. **Rewrite or continue** If a part was missed, reselect the earlier version and send a sharper request. If it landed, build on the new one, and download when you are done. ## What supports text editing in ImageQuarry - **Plain language requests**: Write the change the way you would explain it to a colleague: new lighting, a new background, an added object, a moved element or a new color direction. - **Keep and change in one sentence**: A request can carry both halves, such as "swap the sofa fabric to linen and keep the cushions and rug." Review the version to see how closely it held. - **Every wording on record**: The request that produced a version stays with it, so you can trace which phrasing gave which result. - **A safe place to rephrase**: Because each request makes a new version and earlier ones remain, a weak sentence costs a retry, not your image. - **Pictures for what words miss**: When a material or palette is hard to describe, add a reference image instead. References guide the result; review each version to see how closely it follows them. ## How to write an edit request **Use the three-part shape.** Most good requests have a target ("the background wall"), a change ("to warm terracotta plaster") and a boundary ("leave the shelf and plants alone"). You can skip the boundary when the image has nothing to lose, but on a picture you mostly like, it is the most useful clause you can write. **Borrow the words of the trade.** Photographers and designers have short, precise names for things: rim light, overcast, high key, shallow depth of field, negative space, rule of thirds. They pack a lot of direction into a few words and tend to read clearly to an AI editor. **Give relative amounts.** "Slightly warmer", "much darker", "twice as much space above the head" are easier to act on than an adjective alone. If the first result overshoots, ask for half of that from the previous version. **Rephrase rather than repeat.** Sending the same sentence again after a miss often gives a similar miss. Change the wording: name the target differently, move the keep clause to the front, or split the request in two. **Keep a log of phrasing that works.** Requests that work on your kind of image are worth reusing. Since each request is stored next to its version, you can look back at a good result and copy its wording into the next project. **Check the parts you did not mention.** A keep clause is a request, not a lock. After each edit, look at the areas you asked to keep and confirm they are still right before you build on that version. ## Typed instructions compared with manual editing tools | Task | Manual tools | Text instruction | |---|---|---| | Choosing what to change | Draw a selection or a mask | Describe the target by position, color or name | | Setting how much | Drag a slider to a number | Use relative words such as slightly or much | | Protecting an area | Lock a layer or mask it out | Add a keep clause, then review the version | | Repeating an edit later | Save a preset or action | Reuse the wording that worked | | Fine, exact placement | Direct and precise | Approximate; refine with a follow-up request | ## Limits of editing with words - **Words are approximate**: Some changes are easier to point at than describe. If a request keeps missing, try a reference image or a smaller, more specific change. - **Keep clauses are not guarantees**: Saying what to keep makes it more likely to survive, but you still review every version for drift in faces, text and fine detail. - **Lettering varies**: If the edit involves words inside the image, such as a sign or a label, expect to check spelling and spacing carefully, or add type afterwards in a design tool. - **Results vary between runs**: The same sentence can produce different versions. Choosing a different model for the same request is one way to see another result. ## Plans for editing with text Each request uses credits from a paid plan: Starter at $9 per month with 100 image credits a month, or Pro at $29 per month with 300 image credits a month if you edit often. Rewriting a request counts as a new request. - Starter: Starter at $9 per month with 100 image credits a month - Pro: Pro at $29 per month with 300 image credits a month ## Editing images with text: questions ### Can I edit images with text for free? Not with ImageQuarry, where every plan is paid; Starter costs $9 per month. Each text request you send draws on the plan's monthly image credits. ### Can I edit an image with a text prompt online? Yes. ImageQuarry runs in your browser at app.imagequarry.com. You upload or generate an image there, type the edit prompt, and the new version shows up in the same workspace. ### Do I need an account to edit with text? Yes. Text editing happens inside your ImageQuarry account, so you create one and choose a plan first, then start sending requests. ### What are good AI photo edit prompts? Good edit prompts are short and specific. Name the target, the change and what to keep, for example "Put the bottle on folded fabric. Warmer light." or "Add a sharp beam of sunlight from above." Avoid stacking unrelated changes into one prompt. ### How long should an edit request be? Usually one or two sentences. Long requests tend to mix several changes, and it becomes hard to tell which part caused a problem. If you need many changes, send them as a series. ### Should I say what not to change? Yes, when the image is mostly right. A keep clause such as "keep the face, hands, crop and background" tells the edit what matters to you. Review the new version to confirm those parts held. ### What if the edit ignores part of my sentence? Select the version you started from and rewrite the request rather than resending it. Try putting the missed part first, naming it more precisely, or giving it a request of its own. ### Does editing with text replace text in an image? It can try. A request such as "change the shop sign to read Bakery" asks for new lettering, but AI tools vary in how accurately they draw letters, so check the spelling closely. ## Related guides - [AI image editor](https://www.imagequarry.com/ai-image-editor/): The editing category in full - [Conversational AI photo editor](https://www.imagequarry.com/conversational-ai-photo-editor/): Chaining requests into a session - [AI inpainting](https://www.imagequarry.com/ai-inpainting/): Changing one region of an image - [Reference image AI generator](https://www.imagequarry.com/reference-image-ai-generator/): Show a direction with a picture - [How the process works](https://www.imagequarry.com/#st-1): Describe, edit and refine - [AI image guides](https://www.imagequarry.com/guides/): Every ImageQuarry guide in one place ## Write the sentence you have been meaning to. Name the target, the change and what to keep, then send it and read the new version against your words.