I used to treat AI image tools like a slot machine. Type something. Get four results. Pick one. Type again. Twenty minutes later I had a Downloads folder full of final_v3_really_final.png and no idea which prompt produced the version I actually liked.
That annoyance is what pushed me to build DREAMMMER differently. Not another "better prompt box." A workspace where the chat is on the left and an infinite canvas on the right maps every request, prompt, render, and revision as cards wired to what they came from.
The actual pain
When I'm making a product shot or a social thumbnail, I rarely get it in one pass. I change the lighting. Then the crop. Then I decide the bottle should sit on stone instead of a table. Each of those is a new generation, and most tools treat them as disconnected jobs.
I wanted lineage. If I hover an earlier render, I should be able to branch from that image — edit it, remove the background, upscale it, or hand it back to the agent with the image attached — without losing the path that got me there.
So that's what the canvas does. Searches connect to prompts, prompts connect to renders, and a revision hangs off the image it changes. Ten rounds in, you can still see how each image was made.
How a session usually goes for me
- I describe the idea in plain language. I don't write a Midjourney novel.
- The agent turns that into a detailed English prompt and picks a model (Auto), or I lock a model myself.
- I keep talking: cooler light, less clutter, keep the same product label.
- When a still looks right, I sometimes ask for a short video from the same conversation instead of exporting and hopping to another app.
That last part surprised me more than the canvas. Image → small motion → done, without rebuilding context elsewhere.
If you want to see the kind of outputs people leave public, the showcase is here: https://dreammmer.com/showcase
One-click tools still matter
Not everything needs a conversation. Sometimes I just want the background gone or an old photo cleaned up. Those live under https://dreammmer.com/tools — upload, pick a tool, get a result. The useful bit is that the result still opens in chat, so I can refine it with the agent instead of starting over.
Credits, because people always ask
New accounts get 50 free credits. Cost depends on the model and settings (resolution, video length, etc.). If a generation fails, the credits for that task come back. I keep the rates on https://dreammmer.com/pricing so you're not guessing mid-session.
What I'm still figuring out
Canvas UX is hard. Too many nodes and it looks like spaghetti. Too aggressive about auto-layout and people feel like they lost control. I'm iterating on snap-to-grid, edge routing, and the chat/canvas split so you can drag the divider or go full-screen on the graph.
Also: Auto model selection is good until it isn't. Some tasks want a specific look that only one model nails. Giving people both Auto and a manual model menu was the compromise.
If you try it
Site: https://dreammmer.com — Sign in with Google. Browse templates and public showcases without an account; creating needs login so conversations and credits stick to you.
I'm mostly looking for feedback from people who already bounce between Midjourney / Flux / video tools and hate re-explaining the same brief every time. What would make a "conversation + history canvas" feel trustworthy enough that you'd actually finish a client job in it?
Top comments (0)