DreamLayer AI is an AI agent for image and video production. You describe what you want, and the agent orchestrates the best image and video models to generate, edit, and refine your visuals, from logos and product photos to video content.

DreamLayer routes every request to the best-performing model for that task, selected using its own open benchmark results. Image generation, photo editing, and video generation each use different frontier models, and the routing updates as the leaderboard changes.

Single-model tools use one model for every job. DreamLayer benchmarks models continuously and picks the strongest one per task, adds region-precise editing so you can change just one part of an image, and covers video production in the same conversation.

Yes. The select-area tool lets you paint over a region with a brush and either describe what it should become or remove it entirely. Only the selected pixels change; the rest of the image is preserved exactly.

No. DreamLayer is a hosted product that runs in your browser. You get free credits when you sign up, and the agent handles all model access behind the scenes.

DreamLayer starts free with credits included on signup. After that, you can buy credit packs or a monthly plan inside the app. Every generation and edit shows its credit cost before you run it.

The open-source DreamLayer project is benchmarking infrastructure for image and video diffusion models. It automates prompts, seeds, configs, metric scoring, and reproducible run logging so researchers can compare models consistently. The code is available on GitHub.

Built-in evaluation metrics include CLIP Score, FID, precision, recall, F1, LPIPS, SSIM, PSNR, and temporal consistency for video, all logged automatically with the prompts, seeds, and configs that produced them.

Yes. Every run is logged with its prompts, seeds, configs, outputs, and metric scores, and can be exported as CSV, JSON, or a complete benchmark bundle, so results stay traceable and repeatable.

The open-source tool measures how models actually perform; the hosted agent uses those results to route each request to the best model. The leaderboard on this page comes directly from that research.