Imagine Computer Review (All-in-One Creative Workflows)
What Imagine Computer gets right
Imagine Computer is a new creative suite from Imagine Art. It pulls together the latest multimodal models like MiniMax H3, Cance 2.5, GPT Image 2, and frontier LLMs like Fable 5 and GPT-5.6 Sol. The pitch is simple: one platform for everything from cinematic films to marketing ads to fashion shoots, with persistent project memory so you don't have to reexplain your stack every time.
We connected YouTube and GitHub through Composio MCP. The API is solid. We set fine-grained permissions: read-only for GitHub, no delete on YouTube, ask before uploads. Then we built a knowledge base with our dog HK website's brand guidelines and stack details. The supported file types are text-based only right now—MD, JSON, PDF—so for images you have to stuff them into a PDF. Video support would be huge for visual work; we expect it soon.
The real test was a landing page redesign for dog HK. The agent read our context files, connected to the GitHub repo, and produced a cleaner, storytelling-driven page featuring Missy and Muffin, Michael's dogs. It understood our stack without hand-holding, and we only gave technical constraints like outputting HTML first. The first attempt looked good. The harness is clearly tuned for marketing and website design. We also had it create an ad for dog HK, and it surprised us by pulling art assets from the repo and scraping new images from other sites. It had a good marketing sense without explicit scene instructions.
The sites tool is the standout. We picked a goal, like CRM in sales, and in about five minutes it spun up a subscription and billing tracker with our spending data. You can click and edit specific elements inline. If your repo describes your stack, the agent gets it right away.
The rough edges
Model selection is confusing. The default model tiers—Flash, Pro, Ultra—don't tell you which LLM you're actually using. You can only pick a specific model from a new chat bubble, not from the project folder. We defaulted to Nano Banana 2 for images because we didn't specify GPT Image 2 in chat; the tier matters more than you'd expect.
There's no plan mode. The agent executes tasks immediately. We'd rather have a preview step so we can control the direction.
The biggest issue is agent drift. After two or three prompts, the agent lost context. In one session, it said "no prior tool result in my context" and started reexplaining to itself. It chased its tail, so we had to stop and start a new session. Short sessions work well; long ones fall apart.
Pricing and verdict
We're on the Ultimate plan at about $34 per month. It offers up to 3,000 image generations and 530 video generations per month. That's competitive with other AI creative suites, but the beta limitations mean you'll be working in short bursts.
Imagine Computer has promise. The website design and marketing flows are strong on the first or second run. But the agent drift, opaque model tiers, and missing plan mode make it a tool for early adopters who don't mind restarting sessions. For now, it's a fast prototyping suite, not a long-haul production partner.