A LookX course should teach the second image.

Open a sketch. Generate a handsome building. Everyone nods. Now ask for the same camera, the same mass and a quieter facade. This is where a LookX AI course either becomes architectural education or collapses into a prompt parade.

LookX publishes introductory material, but its public learning offer is fragmented. The company advertises online video tutorials and separate team training with interactive questions. Its official YouTube series covers the interface and learning from community images, while the current product has moved into real-time generation, editing and video. The right response is a compact course tied to today's tasks.

Is there an official LookX AI course?

There is official training, though not one obvious public syllabus that carries every architect from first login to office adoption. LookX's first official tutorial introduces Creations Plaza, Start Creations, Models Agora and model training. A later official episode shows how to learn from community images and begin from Models Agora.

Those videos date from 2023. That does not make them useless, but it makes interface memory a poor learning objective. LookX's current product page lists RealTime, video generation, custom model training, Style Adapter, editing, upscaling and plugins for SketchUp and Rhino. A present-day course should teach the relationship among those controls, not insist that an old menu still sits in the same corner.

LookX also sells education support to institutions. Its official pricing page describes an education plan with faculty and student training courses, technical support and course-aligned customization. That is a contact-sales offer, not a self-serve public class. An individual learner still needs a sensible practice sequence.

Before lesson one, fix the assignment

Use one input throughout the course: a simple exterior perspective with readable openings, edges and ground contact. Keep a copy outside LookX. Record its dimensions and the date. If the source contains a client project, clear its upload under the firm's policy first; our cloud upload gate covers the decision.

Write a preservation brief before generating. Name what cannot move: camera, primary mass, floor count, opening rhythm and site boundary. Then name one variable allowed to change. “Make it better” is not a brief. It is a request for plausible trespass.

Create a contact sheet for every exercise. Put the source beside the outputs and label the settings or route used. LookX's community can be useful for inspiration, but a professional course needs evidence of cause and effect.

Exercise one: make a boring baseline

Begin with the source image and a restrained description of use, material and light. Do not ask for a famous architect's style. Do not add fog because the interface looks lonely. Generate a small set and select the output that best preserves the source, even if another image is prettier.

This teaches the first difficult habit: retention outranks spectacle. The official product page says LookX can transform text and sketches through RealTime and can expand prompts with its Prompt Assistant. Those features may improve descriptive richness. They do not know which parapet carries the design argument unless the input and review criteria make it clear.

Mark every geometry change on the selected output. Use the defect categories from our geometry hallucination checklist: silhouette, opening count, structure, circulation and site edge. The baseline passes when the learner can explain what LookX preserved and what it invented.

Exercise two: separate content from composition

LookX's Gen-2 release introduced separate “Control By Content” and “Control By Composition” routes according to the company's 30 April 2024 release note. The vendor said Gen-2 also added 16:9 and 9:16 output and increased output resolution to 1.5K square. Treat those as documented product claims and confirm the controls available in the current account before teaching them by name.

Run two versions from the same source. In the first, preserve content and change atmosphere. In the second, preserve the broad subject while changing compositional emphasis. Place both beside the source. Ask which architectural facts moved with each control.

The point is not to crown a better image. It is to build a mental model of the control surface. If both routes redraw the facade with equal freedom, the learner has discovered a limit worth more than another hour of adjective tuning.

Exercise three: edit one decision and defend the boundary

Select a local change that a reviewer might actually request: replace one cladding field, reduce planting density or warm an interior light zone. Mask only the necessary region. Keep the source and previous output visible.

LookX's Gen-2 release note says Edit V2 was designed for more precise edits and more natural transitions. That is the vendor's claim. The exercise measures a narrower question: did the requested region change while neighboring joints, openings and shadows remained credible?

Use a difference view if the image editor provides one, or alternate the before and after frames. Our architectural inpainting guide recommends judging the mask boundary as carefully as its center. AI likes to spend beyond the marked budget.

Reject an attractive edit when it alters an unrequested element. That rejection is the lesson. A course that only saves successes trains taste; a course that records failures trains control.

Exercise four: build a review set, not a gallery

Produce a small set in which every image tests one named option. Keep camera and mass fixed. Label the variable in ordinary project language: darker brick, deeper reveal, fewer trees. Avoid labels such as “version 7 final final,” the natural dialect of unmanaged generation.

LookX's current page offers Style Adapter and one-click duplication from Creations Plaza. These can speed a starting point according to the vendor. In the course, use them only when the borrowed reference and its role are recorded. A style source is project information, not decorative weather.

End with a review sheet containing the original, the chosen baseline and two controlled revisions. Add a note under each frame stating what changed and what must be ignored. This turns the image from an atmospheric suggestion into a discussable design artifact.

How much should the LookX course cost?

The course itself can be completed on whatever current plan exposes the required controls. LookX's official pricing page currently lists a free Basic tier, an individual monthly subscription at $20 and an annual subscription at $199. It describes limited credits for advanced functions on Basic and unlimited image generation on paid individual plans. Check current checkout terms before paying because product prices and credit rules can change.

Do not buy a long subscription to finish lesson one. Complete the baseline, inspect the available edit controls and estimate how many revisions the real project would require. The relevant cost is not images per month. It is usable decisions per review.

A paid external course should earn its fee through current interface guidance, source files, critique and a method for comparing revisions. Hours of narrated clicking have little value when the product moves faster than the recording.

What should an architect know after the course?

The learner should be able to choose an input route, state a preservation brief, make one controlled revision and package the result for review. They should also know when LookX is guessing. That last skill is the expensive one.

Do not grade the course on the best frame. Grade it on whether another person can read the source, settings and change note, then understand why the chosen image survived. This is the difference between software familiarity and an office method.

The first image gets applause. The second image gets hired.

Sources checked

Product and training claims were checked on 1 November 2026 against LookX's current product, team and pricing pages, its official tutorial videos and its Gen-2 release note. LookX performance statements remain labeled as vendor claims, and this course makes no claim of first-hand product testing.