From Floor Plan to 3D Visual: A Workflow Guide for Getting Started Fast

The question isn't really whether AI can turn a drawing into a 3D visual anymore, because most people have seen that it can. The question is which workflow gets you a usable result without burning an afternoon, or a stack of credits, getting there. Here's the actual step by step, plus what to expect at each stage.

Step 1: Know what you're starting from

Planlift works from most of what a practice already has on file: elevations, floor plans, masterplans or site layouts, general drawings or hand sketches, and site boundaries (red line plans). It also accepts photos, so a Google Street View or satellite screenshot can be turned into a line drawing, or a drawing can be superimposed onto a satellite image for context.

The tool auto detects the input type, so there's no separate setup step for each format. Start with whatever you already have.

Step 2: Label the drawing (optional, but worth doing on anything complex)

For masterplans and site layouts especially, it helps to draw areas or drop points and label them: green space, drainage and SuDS, site access, a right of way, housing, parking, a play area, public realm. This tells the AI what each part of the drawing is meant to be, which matters most on cluttered or ambiguous plans. A simple single building elevation usually doesn't need this step; a masterplan with several land uses benefits from it.

Step 3: Choose your output options

This is where the workflow actually replaces prompt writing. Instead of typing out a description, you select from a menu:

Output style: photorealistic CGI, watercolour or artistic illustration, white massing model (SketchUp style), clean architectural visualization, line drawing, or marketing render.

Viewpoint: retained perspective (matching the input), top down or overhead, or oblique and aerial.

Scheme type and architectural style: sets the character of what gets generated.

Materials and level of life and detail: controls how much texture, planting, and activity appears in the scene.

The right combination depends on the audience, not just the drawing. A watercolour illustration reads well for a community facing document; a clean architectural visualization suits a board paper; a photorealistic CGI suits an investor pitch.

Step 4: Generate

The tool builds the prompt in the background, the equivalent of what used to take a manually written brief thousands of words long, and returns a visual in a couple of minutes. One credit per generation.

Step 5: Refine without starting over

Three actions are available on any result:

Regenerate creates a fresh version using the same options you already chose.

Develop takes that result and uses it as the input for a further visual, so a white massing model can be developed into a watercolour, then developed into a CGI.

Tweak is a prompt based edit that changes only the specific thing you ask for, keeping style, layout, camera angle, lighting, and buildings the same otherwise. Useful for making it winter or changing the cladding colour without losing everything else about the image. Tweaks currently use a credit each, so it's worth being specific about what you want changed.

Every generation is saved to project history, so past work, and the exact settings behind it, stays available to revisit later.

Practical tips for getting a good result faster

Start rough, then refine. Rougher hand sketches often produce more interesting first passes than trying to get everything perfect before you generate.

Expect a few iterations, especially flat to 3D. Switching a flat drawing into a 3D viewpoint is the least predictable step in the process right now, so budget two or three generations to land on the angle and character you want rather than expecting the first result to be final.

Use Tweak instead of regenerating from scratch once you're close. It's built specifically to stop small changes from undoing the parts of the image you already liked.

Treat the first result as a feasibility image, not a presentation asset. Current output resolution suits early concept and feasibility work well; if you need print grade resolution for a final presentation, that's on the roadmap but isn't the current strength.

What this doesn't do

Worth saying plainly: this produces images, not editable model files, so there's no .skp or similar output at the end. Camera and cross section position can't be controlled with total precision. And the AI can produce a plausible visual, but it can't understand a place the way a human designer does, so the judgement about whether a scheme is actually right for its context still sits with the architect, planner, or urban designer. The tool speeds up visualization; it doesn't replace design.

Getting started

New accounts get 5 free credits, which is enough to run through this whole workflow on a real drawing and see where it lands. If you want a shortcut past the trial and error, the demos library has proven option combinations for common drawing types, worth checking before you spend your first credits.