DALL·E Flow is an interactive, human-in-the-loop workflow for creating high-definition images from text prompts, built by Jina AI and published on GitHub with about 2,800 stars. It chains together several early generative image models into one guided creation pipeline.
The workflow first generates a grid of image candidates using DALL·E-Mega, GLID-3 XL and Stable Diffusion, then ranks them against the prompt with CLIP-as-service so the user can pick the most faithful result. The chosen candidate can be upscaled and refined through diffusion-based post-processing to produce a final high-resolution image. Everything runs as Jina flows exposed through a notebook-friendly client, making the pipeline easy to experiment with interactively. The project was a popular showcase of multi-model generative pipelines; the codebase was last updated in 2023. The repository ships with a Gradio-style interactive client and detailed setup guides, and it remains a frequently cited example of chaining multiple generative models behind one API. Its modular flow design also made it a teaching example for building multi-model AI services with the Jina ecosystem.
DALL·E Flow is free and open source. There are no paid tiers; you run the workflow on your own hardware or cloud GPU and supply your own model weights.
DALL·E Flow is for generative art enthusiasts and ML engineers interested in multi-stage, human-guided image generation pipelines rather than one-shot generation.