Keep Editing Until It Is Perfect

Chat-based editing

Describe what you want changed in plain language. "Make the walls warmer." "Add more natural light from the left window." "Change the flooring to dark hardwood." The AI interprets your instruction and generates a refined version while preserving everything else.

Your original render is pinned at the top of the conversation for easy comparison. Each message produces a new version you can continue editing. The chat panel sits alongside your render, so you can see the result immediately and fire off the next instruction.

When to use this: when you know what you want changed and can describe it in words. "Warmer light," "darker floor," "remove the plant on the left." Fast and intuitive for broad changes.

Visual annotations

When words are not enough, draw directly on the render. Circle the lamp you want replaced. Draw a rectangle around the wall that needs a different color. Add a text label saying "wood texture here." The AI sees both your marks and your text and knows exactly where and what to change.

Available tools

• Pencil — freehand drawing for circling elements or marking areas
• Rectangle — the AI reads this as "change what is inside this box"
• Circle — highlights a specific element to modify
• Text — labels to describe what you want in specific areas
• X mark — tells the AI to remove the marked element

When to use this: when you need to point at something specific. "That lamp" is ambiguous in chat — circling it is not.

Area-select editing

When 90% of your render is perfect and only one part needs work, select just that area and re-render it. The surrounding image stays exactly the same. Only the selected region gets regenerated based on your instructions.

This is faster than re-rendering the entire image and prevents introducing unwanted changes to parts that already look right. Draw a selection around the problem area, type what you want changed (or leave it blank for a fresh generation), and the AI regenerates only that region.

When to use this: when one corner of the image has an artifact, or you want to change a single material without risking the rest of the scene. Surgical precision.

Reference images

Paste a photo of the exact material, style, or look you want. A photo of herringbone parquet flooring. A swatch of the fabric for the sofa. A mood board image for the overall atmosphere. The AI uses your reference to match textures, colors, and style in the re-render.

Reference images can be pasted directly onto the annotation canvas or attached alongside a chat message. The AI compares the reference with the current render and applies the visual qualities you are pointing to — color palette, texture pattern, material finish, lighting mood.

Iteration history

Every edit creates a new version. Eler tracks your full editing history in a filmstrip view so you can scroll through all iterations, compare any two versions side by side, and go back to an earlier point if a later edit went in the wrong direction.

Each version links back to the edit that produced it — you can see exactly what chat instruction, annotation, or area-select produced each result. Nothing is lost. You can branch off from any earlier version and take the design in a different direction.

Combining methods

All four editing methods work together. A typical refinement session might look like:

1. Start with a chat instruction: "Make the overall lighting warmer and more afternoon-like"
2. Annotate the result: circle a wall and add a text label "exposed brick here"
3. Paste a reference image of the specific brick texture you want
4. Use area-select on a corner where an artifact appeared to clean it up

Each step is tracked in the iteration history. The combination of text, visual marks, and reference images gives you maximum precision over the final result. For the initial 3D setup before rendering, see material assignment which controls how surfaces appear in the first render, and 3D models and images for the full upload-to-render workflow.