In the AI era, generating images is convenient, but fine-tuning details, comparing options, or iterating on a single image often requires tedious back-and-forth. This article explores the capabilities of 'Infinite Canvas' and shows how one small idea can expand into a full suite of assets.
A few days ago, I wrote "🎬 Don't Let Assets Gather Dust! A Step-by-Step Guide to Building a 'Viral Video Creative Production Line'", sharing how to turn saved videos into your own finished clips.
https://x.com/Lonely__MH/status/2102716023463911749
If you can make videos, what about the accompanying product shots, packaging, and promotional posters? Do we have to go back to the chat box and generate them one by one?
To be honest, without a professional design background, my biggest headache with AI image generation is changing details: regenerating often changes parts I was already happy with.
Recently, I tried 'Infinite Canvas'. I just wanted to test it out, but as I kept working, inspiration flowed. Starting from one poster, I ended up creating a complete brand identity set, and even made a 15-second ad along the way.
01 | See the Results First
This was an unplanned success. Here are the results 👇🏻
The project is for a fictional pomegranate sparkling drink brand called "Sui Ren Shi He" (Harvest Time Harmony).
A tiny idea gradually took shape into something complete.
0:00 / 0:15
02 | What Can Infinite Canvas Do?
Type /Create Canvas in the Doubao Work chat box to start a canvas. To see how to open it and what tools are inside, watch this demo 👇🏻
0:00 / 0:40
This creative design upgrade includes not only Infinite Canvas but also editable designs and Seedream 5.0 Flash.
The Design Agent can help plan solutions first, then generate images, videos, and sets of materials.
In Infinite Canvas, selecting an image pops up a row of tools. Their general uses are:
- Reference Generation: Use this image as a reference to continue generating new images.
- Edit: Box the area that needs modification and state how to change it.
- Layer Splitting: Break the image into independent elements to adjust position and size.
- Cutout: Extract the main subject separately.
- Erase: Remove unwanted objects from the scene.
- Outpainting: Extend the background beyond the original image to add space and whitespace.
- Enhance Clarity: Improve image sharpness.
At the bottom of the canvas, there are tools for uploading materials, image generation, video generation, and marking. If you want to create a design set or a video, you can also call 'Creative Design' or 'Creative Video' from the left-side chat box.
Reference images and different versions are laid out on the same canvas, making it easy to compare them side by side. Pick a satisfying one and use it as a reference to continue; if there's an issue, bring the specific image and keep editing.
03 | Generate an Image First
Open 'Creative Canvas' in the Doubao Work client. If you have a reference image, upload it; if not, enter a prompt or just tell Doubao what you want to generate.
My inspiration this time came from this pomegranate poster 👇🏻
Looking at this image, I suddenly thought: The name is already there. What would an autumn pomegranate sparkling drink called "Sui Ren Shi He" look like? Let's try it.
Select the poster and click 'Reference Generation'. You don't need a long prompt; just say what you want to do, for example:
1Refer to this image and help me create a product poster for a pomegranate sparkling drink.
This round used the newly added Seedream 5.0 Flash. It generates quickly, suitable for trying several directions to see which vibe is worth pursuing.
💡 Use Flash for finding inspiration and testing directions. Once the direction is set, try Pro for fine-tuning. Don't obsess over every small detail from the start.
04 | Create a Brand Material Set Next
With the product image ready, I had a new idea: What if I designed the packaging and promotional materials together?
I found the 'Creative Design' feature inside, brought in the product image, and asked it to design a unified brand visual system.
It provided a plan first. After I confirmed, it created a brand showcase board. Then I asked it to generate corresponding images based on the plan, resulting in gift boxes, tote bags, social media graphics, and display setups.
0:00 / 0:36
Next, I asked it to create a set of promotional posters. However, when making a set, I encountered an issue: some images had products inconsistent with the initial design.
You can't have a different product in every image of the same set.
I referenced the problematic images, pointed out the differences, and included the correct image as a reference to let it continue adjusting.
🔥 Original and new versions are both on the canvas, so discrepancies are easy to spot side by side. When modifying, just refer to the image directly without re-explaining everything.
05 | Make the Product Move
With packaging and posters done, I had another thought: Since it's the same product, can I make an ad using this design set?
So I called 'Creative Video', brought in the product images, and asked it to generate a 15-second e-commerce ad.
It planned the shots first, then generated the final clip. The previously selected product images became references for the video, so no need to start from scratch.
0:00 / 0:15
However, the Chinese text in the final clip was somewhat unclear, and details needed polishing. Before actual publication, careful checking is still required.
06 | Good Images Can Still Be Improved
Previously, I mainly continued generating and modifying based on images. For images you're already happy with but just want to tweak, Layer Splitting and Outpainting are very useful.
Layer Splitting breaks down elements in the scene for adjustment. If you don't want to change content but just move the title or adjust the subject size, try this. Which parts can be adjusted depends on the split result. Demo below 👇🏻
Outpainting gives the scene more space. If you want room for copy or want to change the aspect ratio, let it extend the background outward. I expanded a square product image to vertical format this time.
It's like pulling the camera back slightly, revealing space that wasn't visible before. Sometimes new ideas emerge from this.
What if you run out of credits while frequently making images/videos? Doubao has separate 'Creation Credit Packs' to replenish image and video generation quotas. These credits are not subject to rolling usage limits. If you want to focus on producing a batch of materials or try more directions, you can reduce credit anxiety. Flexible spending 😄
07 | FAQ
Q: Can I use it without design experience or prompt writing skills?
Yes. Reference an image and say "Help me make a poster" or "Help me make an ad video" to start. Add requirements as needed, look at the results, and keep chatting. No need to learn a prompt syntax first.
Q: Can I use my own product photos, logos, and materials?
Yes, upload your own image materials and reference them to continue creating. Start with product images if you have existing products; tell it to preserve brand names, logos, etc.
Q: If I'm unhappy with edits, can I revert to previous images?
Original images remain during reference generation and conversational editing. If new results are unsatisfactory, select the original image to continue; the canvas also has undo and redo functions.
Q: How to maintain style consistency across a set of images?
Keep a satisfying image as a reference and continue working with it. If things drift off-style, reference the correct material and point out specific differences. Reference images help sustain style, but review them together after generation.
Q: After layer splitting, can text in the image be edited directly?
Depends on the split result. If text remains an image element, it can be moved or scaled, but not directly typed into like a text box. To change copy, continue requesting modifications and check the result.
08 | Conclusion
Today I mainly experienced Doubao Work's Infinite Canvas, along with 'Creative Design' and 'Creative Video'. Without a professional design background, I just went through a simple workflow, hoping to inspire everyone.
For me, the most interesting part is: You don't need to think of everything upfront. Have an idea, try it, modify if unsatisfied, and all previous attempts stay on the canvas. Many new inspirations emerge bit by bit this way.
Such tools truly lower the barrier to creation. Beginners can start with a sentence or an image, trying to realize ideas in their heads; experienced designers can quickly try different solutions, saving energy for judgment and polishing.
Tools can do more and more for us, but what to express, what to like, and what is worth continuing are still decided by ourselves.
Some ideas used to stop at "just thinking." Now, you can start trying hands-on.
📚 Past Articles
- What's it like to fit all terminal Agents on one phone?
- Complete Guide to Local LLM Deployment
- Hermes Agent Practical Guide: From X Anxiety to Automatic Sedimentation
- Programmer Hair Loss Prevention Guide
- Hermes Connects to iMessage
- Hermes Connects to X Premium
- Hermes Agent Complete Guide
- Hermes Agent Beginner Guide: Auxiliary Models
- Hermes Agent Beginner Guide
- Hermes Agent Advanced Guide
- Hermes Agent Incomplete Guide
- Nigeria Region Claude Pro Subscription Complete Guide
- Nigeria Region Apple ID Registration Tutorial
- Turkey Region Half-Price ChatGPT Plus Subscription
- US Region Apple ID Registration
- Claude/ChatGPT/Gemini Alipay Subscription
- Mac Local LLM Deployment Complete Tutorial
- IP Quality Check
- Why Doubao Doesn't Recommend Your Brand
- How to Explain to Your Grandma That Doubao Isn't Real
If this tutorial helped you, feel free to Follow + Like + Repost 👏🏻
Follow @Lonely__MH for beginner-friendly tutorials and AI tool insights.
