FlowInOne

::: { .text-center } Unified image-to-image generation via multimodal flow matching
Seamlessly fuse sketches, text, layouts, and symbols into photorealistic images with a single flow model. :::

:::: {.grid .grid-3} ::: {.card}

🖌️ Multimodal Visual Encoding

Encode freehand sketches, handwritten text, layout primitives, and symbolic instructions into a shared 2D visual latent space—preserving semantics and spatial structure without modality-specific decoders. :::

::: {.card}

🔄 Geometry-Aware Flow Matching

Leverage geometry-preserving flow dynamics to generate high-fidelity images with accurate spatial alignment and structural coherence from fused visual prompts. :::

::: {.card}

🧩 Unified Latent Space

Train a single denoisable latent space that supports diverse input modalities, eliminating the need for alignment losses or separate conditioning pathways. :::

::: {.card}

🚀 End-to-End Generation

Generate photorealistic target images directly from multimodal visual prompts—no cascaded models, no post-processing, no compromise on quality. ::: ::::

::: {.grid .grid-2}

pip install flowinone

::: ::: { .text-center } Get Started → :::