How to Handle Client Revisions on a Render Without Starting Over

Do you have to regenerate the whole render when a client asks for one change?

Usually not. What a client approves is the whole composition: the camera angle, the proportions of the room and the mood of the light. A note such as “change the sofa” or “make the walls lighter” asks for none of that to move. If you regenerate from scratch, the AI reinterprets the entire scene, and things the client liked can drift along with the thing they wanted changed.

The practical rule is simple: protect the approved image and change only what the note points at. The rest of this article turns that rule into a routine you can repeat on any project:

• Split the client's note into single, concrete changes.
• Apply each change as its own iteration and give every version a name.
• Show the client a clear before and after.
• Know the cases where regenerating is the cleaner move.

A note on terms. BentoMesh is a Turkish-first AI visualization platform for architects and interior designers, and its interface is available in six languages. Tool and mode names are given as they appear in the Turkish interface, with a short English description, because the labels in the English interface may differ. Credits are called işlem hakkı in Turkish. The client and project examples are fictional; they do not describe a real client.

How do you turn a client's note into a list of single changes?

Client notes arrive as sentences, and each sentence can hide several requests. Ask three questions of every line: Which object or surface? What exactly should it become? What must stay the same? The third question is the one people skip, and it tells you what to check when the result comes back.

Words like “lighter,” “warmer” or “more modern” are not changes yet. Ask the client what they mean, or prepare two options. “Lighter” is a good example: it can mean a paler paint color or a brighter room, and those are different edits. The table splits a fictional living-room note:

Number the list (C1, C2, C3) and send it back in writing: “We will make these three changes and keep these things as they are. Correct?” That confirmation catches misunderstandings before you generate, and because every generation uses credits, it also cuts down on trial and error. One more caution: “lighter walls” is a direction, not a paint code. The output will not match a specific manufacturer color, so confirm the final tone with a physical sample.

How can you change only one area of an AI-generated render?

Render Lab is BentoMesh's advanced AI render studio, built for control over lighting, camera, materials and atmosphere. For revisions, the useful part is that you can edit an existing render in the same panel: the edit mode (Render Düzenle) regenerates only a chosen area of an existing output. Adding new camera angles or turning a result into a video also happens in the same panel. For a note about one object or one surface, such as the sofa or the wall color, this is the appropriate route.

In practice the flow is short: switch Render Lab to the edit mode, upload the frame the client approved, and describe the change in a short written instruction, for example “make the wall color anthracite.” The instruction can be written in Turkish or English. Target one object or one surface per instruction.

Three habits make it work:

• Change one parameter at a time. Render Lab's own guidance is to change a single parameter first, for example the atmosphere, and watch its effect; changing many things at once makes the outcome unpredictable.
• Feed it the approved version. The input should be the last image the client signed off, not a version you tried and rejected.
• Compare before you accept. Regenerating an area does not guarantee that everything outside it stays identical. Check fixed points: the number of windows, the shape of the furniture, the lines of the walls. The compare button (Karşılaştır) on the result screen opens a before-and-after slider; use it for your own check before you accept the result.

If you have no render yet, create the first image with the AI render generator and start the revision loop from that frame.

How should you run each iteration and name the versions?

If you apply three changes at once and the result looks wrong, you cannot tell which one caused it, and if the client approves two of the three you have no intermediate version to go back to. So each change is its own iteration, and its input is the previous approved version.

Suggested order: objects first (the sofa), then surfaces (the wall), and light and atmosphere last. Light changes how every surface reads, so adjusting it early makes later comparisons meaningless.

Naming: put the project, room, version and change in the file name, for example living_v03_walls-lighter. Version v01 is what the client saw first. If you made two options for one change, call them v03a and v03b. Names like “final_new_v2” stop telling anyone which file is current after a few rounds.

Keep a simple revision log:

Every iteration is a separate generation and uses credits (işlem hakkı). We do not quote amounts here because they change; current figures are on the pricing page. That is why revisions work well in rounds: collect the notes, agree the list, then generate. If a generation fails, the credits are returned to your account.

How do you show a client the before and after?

The client should never have to open two files and hunt for the difference. In your own presentation software, set it up like this:

• Same crop, same size, side by side. Previous version on the left, new version on the right, version numbers above the pair.
• One sentence under each pair. “v02 to v03b: walls one step lighter. We compared everything else with the previous version and saw no noticeable difference.”
• Say what did not change. “The sofa, rug and painting are as before” saves the client from wondering whether those moved too. Write it only after you have compared the two versions.
• End with a question that closes the choice. “We prepared two tones for the walls: a or b?” gets a faster decision than “What do you think?”

If an unwanted side effect appears, for example the painting changed shape, do not hide it. Either regenerate the same change or tell the client about the difference. If the client spots it first, trust in the rest of the images drops. Before sending, run the image through your usual quality check.

When the client approves a version, get the approval in writing and freeze it: that version is no longer changed against new notes, and a new note starts a new round. If you need presentation resolution, apply the image upscaler (Kalite Yükselt) to the approved version only; there is no point in enlarging every draft. Finally, label the image as an illustrative visualization, not a photograph.

When is it better to regenerate instead of editing?

Editing is for protecting an approved composition. If the note touches the composition itself, forcing an edit wastes time and credits. This table suggests a route by type of note:

If your source is in SketchUp or 3ds Max, the optional BentoMesh plugins send the active view to generation; from other programs you export the view as PNG or JPEG and upload it to the AI render generator. When the source has changed, do not try to edit older versions.

Starting over weakens the earlier versions and approvals, so name the new image something like “v01 new base” and tell the client plainly that it is a fresh start. If windows or furniture do not match your source, simplify the source view first before generating again.

What are the honest limits of AI render revisions?

• No consistency guarantee. BentoMesh does not guarantee that geometry is fully preserved, so even when you regenerate one area, small differences can appear elsewhere. Because every edit is a separate generation, they can accumulate as the chain of edits grows. Compare every version with the previous one and with the source.
• Every generation uses credits. Put the number of revision rounds in the proposal or contract. Current plans are on the subscription page, and you can start with a free account and welcome credits.
• The output is a presentation image. Exact dimensions, product choices and paint codes come from your technical model and physical samples, not from the render.
• Results depend on the source. A blurry or badly framed source gives a weak edit as well.
• AI can make mistakes. Treat the output as something to check, not as a final decision. You make the call.
• It is an illustration. Tell clients and anyone else who sees the image that it is an illustrative visualization, not a photograph.

In short, revision management is less about a tool feature than about routine: split the note, apply one change at a time, name the versions, compare, and get approval in writing. BentoMesh handles the generation step; the decisions and the checking stay with you.

Türkçe sürüm / Read in Turkish