OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Omini Kontext

Omini Kontext is a framework for training and running inference with Flux.1-Kontext-dev using multiple image references. It enables reference-based edits through 3D RoPE embeddings while leaving the model architecture unchanged, and the repository also notes Qwen-Image-Edit support.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Omini Kontext — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Omini Kontext when building or exploring Flux-based image edits in which one or more reference images should guide the generated result, or when training a model for that workflow.

Who it is for

  • Developers creating reference-conditioned image-editing workflows with Flux.
  • Practitioners training image models with multiple reference images.
  • ComfyUI users working with the repository's nodes and patches alongside the Nunchaku extension.

Documented uses

  • Generate an edit using an input image, reference image and text prompt.
  • Train a model for multi-reference image conditioning.
  • Explore character insertion into an existing scene, including approximate spatial placement.
Before you choose

Requirements & limits

The README identifies a Hugging Face account as a prerequisite for model access. A Replicate workflow is also described for trained models uploaded to Hugging Face.

  • The README recommends a CUDA-compatible GPU with 24 GB or more of VRAM.
  • The documented inference stack includes Python, PyTorch, Lightning and Diffusers.
  • Spatial character insertion is experimental; the README notes occasional duplicate characters and approximate placement.
  • ComfyUI integration is provided through nodes and patches described as compatible with the Nunchaku extension.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run a documented two-image inference workflow

Generate and save an image edit conditioned on an input image and a reference image.

Before you start

  • Python 3.8 or newer
  • CUDA-compatible GPU; 24 GB or more VRAM is recommended
  • PyTorch 2.0 or newer and Lightning 2.0 or newer
  • Hugging Face account for model access
  • Repository dependencies, including Diffusers installed from its GitHub repository

Steps

  1. Install the repository dependencies and the documented PyTorch, Lightning and Diffusers requirements.

  2. Load FluxOminiKontextPipeline from the FLUX.1-Kontext-dev model and move the pipeline to CUDA.

  3. Load an input image and a reference image, then load the documented character LoRA weights.

  4. Call the pipeline with the images, reference delta and prompt, then save the first generated image.

Expected outcome

A generated image saved as output.png, conditioned on the input and reference images.

Guide limitations

  • The README's code is an example, not a guarantee of successful execution on a particular setup.
  • The documented inference example uses a particular character LoRA and example prompt; it does not establish that every model or workflow behaves the same.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
Saquib764/omini-kontext · Official homepage title · retained in its source language
How we review sourcesSuggest a correction

Related options

ComfyUI — A separate workflow application for which the repository includes nodes and patches.

Explore project →

Replicate — A hosted route documented for using a trained Omini Kontext model after uploading it to Hugging Face and entering its details on Replicate.