Realistic 3D modelling is a difficult skill to master. Those who take it on board to learn the esoteric skills of trueSpace, 3D Studio Max, and LightWave without tutoring or lessons, often find the going more complex and more time consuming than they first expected. Even texturing a box or positioning a light, or just plotting a camera move can take hours of endless fine-tuning and attention to detail. And the modelling side of 3D art and realisation is the most difficult, the most exacting of all the skills you have to master.

To deal with this horrible brick wall facing those users that want to create and work with 3D, MetaCreations have come up with Canoma. It sets out to create photo-realistic 3D models from 2D photos and illustrations, with the barest of tool sets and minimum user input.

Focusing heavily on buildings, architecture, interiors, and still life, Canoma uses a system called perspective projection. By super-imposing various wireframe shapes on the flat image, and pinning their vertices to the corners visible on the image, you gradually build up a full 3D realisation of your scene. In addition, by stealing parts of the image to use as textures, and duplicating where necessary, the end results look almost indistinguishable from the original - bar the fact that you can rotate it, zoom in, or export the whole thing in Web-ready VRML.

You start with a basic 2D image, centred in the main window. Seventeen basic 3D shapes or primitives are lined up along the bottom. On the simple side: cube, rectangle, pyramid and trapezium. On the more complex: staircase, curtain, archway and table. After a quick survey of your scene, you simply click on a shape to bring it into the workspace. You use the mouse to drag it out to scale, and then pull the corners of your object into alignment with its 2D equivalent for pinning. Once complete, you can choose to view the results as a 3D model, flat shaded or textured. Textures are built from instant snapshots of the image underneath each 3D shape. The results with Quick rendering are undetailed but functional. For a performance toll, the Quality setting adjust the output to compensate for brightness differences, as well as anti-aliasing and effectively minimising the seams on objects.

Multiple angles
One picture may suffice for a very simple scene, but ideally you'll have multiple shots of your subject, taken from various angles, probably with a digital camera. This not only allows you to correctly texture map the rear of objects and deal with half-concealed shapes, but also makes the scene more realistic and accurate as Canoma uses the extra information to adjust its maths and establish the correct perspective.
When you bring in an extra picture of the scene, you're asked to roughly realign the current 3D models to its perspective. This not only helps you visualise the new angle, but means less work is necessary to configure the models. Canoma also uses this extra information to make corrections to the earlier structure.
Sometimes, however, especially on close-up shots, details are out of the frame and cannot be accurately pinned down. This is where beads' come in. They're not based on corners, but on vertices, which appear along a line and can be used to nudge out-of-shot objects into shape.

Multiple shots can also be used to increase the quality of the textures. You can shoot from a variety of general distant views and then zoom in and take close-up shots of details. These can then be merged into the Canoma scene to increase the resolution of the textures, and to allow zooming into your image without too much degradation. For further detail, individual textures can be loaded into a 2D package such as Photoshop and tweaked even more. Saved changes are instantly updated back in Canoma.

The interface is typical MetaCreations, all graphic-designed and welcoming pastel colours. A suite of trackballs and directional arrows control the camera views. You can pan left and right, track in and out, and control the depth of field. A floating palette acts as a library of all your current images, while a toolbar holds a basic but uncluttered array of workflow tools.

Solo mode is on hand to single out models in complex scenes. The Glue tool makes for very precise alignment of objects. Stacking allows you to pile object on top of object. The Duplicate command is useful for saving time rendering similar landscapes, while the Mirror Textures effect cleverly duplicates material on to out-of-shot sides of objects to save you having to. Each item has in Information dialog with shortcut commands (rotate 90 degree around axis, and so on) for geometrically-precise manoeuvring.

When you aren't creating, you can animate. In a basic motion studio you can plan and direct flybys of your scene. You shift the camera (rotate, bank, zoom, or dolly) and create keyframes. Canoma fills in the motion in between.

While animations can be rendered out as QuickTime files, Canoma has a spread of export options for the 3D models it creates. VRML2 and MetaStream - MetaCreations own answer to VRML - are high up on the list, given the Web potential of this product, but users of Poser, Bryce 3D, and LightWave are catered for by the OJB format. Importing finished models into Poser, however, proves complicated (much copying of individual files into specific directories is required). The trueSpace SCN format is very well catered for - Canoma scenes are perfectly preserved in the Caligari 3D package. MetaCreations' mandate is to make every piece of its software as approachable, intuitive and as easy to use as possible. Sometimes it fails. Sometimes its enthusiasm for features clutters the interface. With Canoma however it has succeeded.

Quick and easy
Canoma, as you suspect from its wafer-thin manual, is monumentally easy to use. In fact, its simplicity, pared down design and quick-fire results actually make it enjoyable to use. There's something intensely creative about wringing a 3D model from a 2D image. Breaking an image into primitives and trying to isolate the best combination of shapes to sum up a structure is taxing but very satisfying, exercising every last neuron of spatial reasoning in your brain.

It's also quick. Arranging a townscape or standard interior shot into 3D can take minutes. However, using Canoma is a real skill. You develop your own techniques and, after time, you find yourself quickly reducing images into primitives and negative space, and you find your pinning and beading skills rapidly improving. At the same time, Canoma learns more and more about your scene. You have to use less and less pins as the program leaps ahead and guesses accurately where your shape will lie.

However, there are things it can't do. The absence of anything close to a curve or a sphere is irritating and left unexplained. Anything organic or naturalistic is practically out of the question. There are objects in the primitives selection with editable polyline contours - essentially allowing you to add points to their surface to create objects like cars, or signs or even people - but the detail you can achieve is disappointing. If you think back to 3D action games about three years ago, you should have a good idea of the quality.

Also, for a 3D package, the animation features are very basic. Disappointingly, you can't add acceleration or deceleration, or any kind of dynamic features to your motion. Neither are there provisions for scaling keyframes. Having said that, there's some scope for unique visual effects, such as changing the field or view and warping a still image with bizarre fish-eye lens distortion.

The rendering algorithm uses a lot of fuzzing, blurring and clever edge detection to line up textures, but this can give a lot of the output a kind of Canoma sheen'. And if you&q