ChatGPT, Claude, Codex + any MCP client
Connect Dream Pixel Forge to the conversation you already use. Your agent turns one brief into on-brand images and video, checks the work, and brings the strongest results back to you.
Connect the studio once. After that, every creative job starts with a brief and ends with work you can review in the same conversation.
01 · CONNECT ONCE
Ask Claude, Codex, ChatGPT, or another MCP client to connect Dream Pixel Forge. You approve the machine in your browser, with no API key pasted into the chat. Your brands, moodboards, and personas are ready when the conversation resumes.
02 · BRIEF, THEN REVIEW
Ask for a spring launch pack and move on. The agent plans the set, generates stills and clips against your selected brand, checks the results, and brings back the strongest work with the run budget still visible. You stay the art director without managing each generation.
Yes. Copy this prompt into your agent and it will install the CLI, ask you to approve the machine, and learn the workflow on its own.
Set yourself up with Dream Pixel Forge so you can generate on-brand images and video for me. 1. Install the CLI: npm install -g dreampixel-cli (or use npx dreampixel-cli). The command it installs is dpf. 2. Run dpf whoami. If no key is configured, ask me to run dpf login in my terminal: it opens my browser to approve this machine. Never ask me to paste an API key into this chat. 3. Run dpf skill install in the project root so you have the full workflow on disk: create a run, plan concepts, generate, validate with vision, and render video in the background. 4. Confirm the connection with dpf tools, then ask me for my first brief. If you connect over MCP instead of a terminal, the endpoint is https://www.dreampixelforge.com/api/mcp with my API key as an Authorization: Bearer header.
Not because it is clever. Because it is convenient: the workflow you already have picks up a studio.
No new app to learn. Ask for visuals in the middle of a Claude, Codex, or ChatGPT session and the results come back to the same conversation.
Your brand profile, moodboards, and AI personas ride along as context, so you never re-describe your style to get consistent output.
Runs carry a 20-credit image budget, and vision validation scores every result against the brief before it reaches you.
The same run produces stills and short clips. Video renders in the background while the agent moves on to the next concept.
The same studio output as every other Dream Pixel Forge surface: product scenes, key art, portraits, posters, interiors, and short video.
Instead of typing prompts into a web form, an AI agent runs the whole creative job: it plans concepts, generates images and video, and checks the results, while you approve the brief and review the output. Dream Pixel Forge ships its own creative director agent in the studio, and the CLI and MCP endpoint let you bring your own.
Any MCP client: the Claude desktop app, claude.ai connectors, Claude Code, Codex, and ChatGPT all work. Any terminal-based coding agent can drive the dpf CLI directly, and dpf skill install teaches it the whole workflow in one command.
No. dpf login opens your browser so you can approve the machine: nothing to copy or paste. Headless environments and CI read the DPF_API_KEY environment variable instead, and connectors that cannot send headers can carry the key in the endpoint URL path.
The endpoint serves your brand profile, moodboards, and AI personas to the agent as context. Output matches your style without re-describing it in every request, the same way the built-in creative director works.
Generations spend the same credits as any other studio work, and every run carries a 20-credit image budget so an over-eager agent cannot run up your bill. See the pricing page for current credit packs and plans.
Yes. dpf video renders clips in the background from the same run, and the agent polls dpf run status with the watch flag until the URL is ready. Video works over MCP as well.
Yes. dpf vision in validate mode scores a generation against the brief and your brand profile, so the agent fixes near-misses before you ever see them. You review finished results, not raw attempts.
Install the CLI or add the MCP endpoint, approve the machine in your browser, and hand your next visual brief to the agent you already trust.