Last week I helped a friend fix an event landing page. The AI artwork looked excellent, but the uploaded file was nearly 6 MB. On a phone, visitors saw an empty block before the image appeared, and the card crop removed half of the subject's face. AI had reduced the time needed to create the image, but it had not removed the work required to deliver it properly.
Canva introduced its first in-house image model, Canva Craft, on October 22, 2026. According to Canva's official announcement, Craft can generate images at up to 2K resolution, supports more than 30 aspect ratios and can render text in 55 languages. Canva positions it for posters, social posts, ads and other design-heavy work.
The useful story is not simply that another image model has launched. It is that more people can now produce something close to a finished design. A Dream Lab output still is not automatically a production-ready web asset, though. This workflow covers the gap between generation and publishing.
TL;DR: Choose the destination and aspect ratio before generating in Canva Craft. After export, crop or remove the background first, then convert to WebP or JPG, compress for the actual display size and inspect metadata before publishing. PocketKit's aspect-ratio, conversion, compression, background-removal and EXIF tools cover that post-generation workflow.
What actually changed with Canva Craft?
Three parts of the announcement matter for day-to-day production:
- Composition is planned before rendering. Canva describes a two-stage architecture that first reasons about elements, spatial relationships, scale, hierarchy and text, then renders the image.
- Aspect ratios are not an afterthought. Craft supports more than 30 ratios plus custom sizing, so a creator does not have to generate a square and force it into a banner later.
- Typography is a first-class use case. Prompts can be written in multiple languages and the model can generate text in 55 languages. Small print, prices, dates and brand names should still be checked character by character.
Canva's rankings and quality claims come from its release materials and evaluations. They do not guarantee that every prompt will work on the first attempt. Hands, exact product geometry, fine text and spelling remain sensible pre-publication checks.
Step 1: decide where the image will live
The same “autumn coffee launch” needs a different composition for a website hero, a square feed post and a vertical story. Define the container first so the prompt can place the subject intentionally.
| Destination | Common ratio | Composition note |
|---|---|---|
| Blog or landing-page hero | 16:9 | Keep the subject away from the headline area |
| Product or profile card | 1:1 | Center the subject with crop tolerance on every edge |
| Story or short-video cover | 9:16 | Keep faces and text clear of top and bottom interface areas |
| Social landscape post | 1.91:1 or current platform recommendation | Put critical content inside a safe area |
Use the AI Image Aspect Ratio Calculator when you need matching pixel dimensions. Ratio controls composition; resolution controls detail. A 1600×900 image and a 1920×1080 image are both 16:9, but they do not have the same file size or pixel count.
Platform specifications change, so check the destination's current guidance before a campaign instead of treating one size chart as permanent.
Step 2: write the prompt like a small design brief
“Make a premium coffee poster” leaves nearly every decision to the model. A useful image prompt answers five questions:
- Purpose: website banner, social post or product image;
- Subject: object, person, setting and action;
- Composition: placement, empty space and camera distance;
- Visual treatment: lighting, materials and color, not just “premium”;
- Constraints: no watermark, no extra copy, no cropped hands, and so on.
For example:
Create a 16:9 website hero for an independent coffee shop. Place a dark green ceramic cup on the right third, lit by soft morning window light, on a wooden table with a few coffee beans. Leave clean dark negative space on the left for an HTML headline. Realistic commercial photography. No text, logo, watermark or extra tableware.
If requirements keep disappearing from your prompts, use the AI Prompt Generator to organize the goal, constraints and output requirements, then trim the result for Craft. A prompt is not better because it is longer; each sentence should change a visible decision.
Step 3: make structural edits before reducing file size
Use this order after generation:
1. Crop and recompose
Confirm that the destination container does not remove the subject. Export separate landscape, square and vertical variants when a campaign spans several placements. CSS object-fit: cover is not an art director.
2. Remove the background when the asset needs it
Product shots, portraits and sticker-like graphics can go through the AI Background Remover. Inspect hair, glass and semi-transparent shadows before accepting the transparent PNG.
3. Pick a delivery format
- WebP: a strong default for most website images;
- JPG: useful for photography and older production workflows;
- PNG: appropriate for transparency and lossless hard edges, usually at a larger size.
Prepare variants with the Image Converter. Keep a high-quality master separately from the optimized web delivery file.
4. Compress last
Use the Image Compressor after dimensions and format are settled. Watch skin texture, small product labels and smooth sky gradients while adjusting quality. There is no universal quality setting: the goal is the smallest file that still looks clean at its real display size.
Step 4: do a ten-second privacy check
A fully generated image may not contain camera GPS data. However, a file that has passed through a phone library, a photo composite or another editor can still carry device, time, location or software metadata. This is easy to miss when a team combines an AI background with a real product photo.
Inspect public assets with the EXIF Cleaner. Metadata removal does not hide a street number, employee badge, license plate or screen visible inside the picture; those require a visual review.
A repeatable Canva Craft publishing checklist
- Choose the platform, ratio and safe area first
- State purpose, subject, composition, treatment and constraints in the prompt
- Zoom in on text, hands, product geometry and brand spelling
- Make a deliberate crop for each placement
- Test WebP for web delivery and keep PNG where transparency matters
- Compress at the real display dimensions and preview on a phone
- Review EXIF data and visible private information before publishing
- Keep an uncompressed master so you never have to upscale a compressed copy
The model creates the draft; people still own delivery quality
Craft's combination of composition planning, typography and multiple aspect ratios is more relevant than a simple “prettier image” claim. Search pages, ads and landing pages ultimately load a real file. The model's ranking cannot compensate for the wrong crop, an oversized asset or a misspelled price.
The practical workflow is short: choose ratio → write the brief → inspect structure → convert → compress → clean metadata → preview on multiple devices. Those extra minutes are often faster than regenerating the image ten times.
Source note: Craft resolution, ratios, language support, architecture and availability in this article come from Canva's October 22, 2026 announcement. Features and plan limits may change, so verify the current Canva documentation. PocketKit is not affiliated with or endorsed by Canva.