Flux 2 Klein Multi Image to Image ComfyUI workflow for Videcool

The Flux 2 Klein Multi Image to Image workflow in Videcool provides a powerful and flexible way to transform multiple images while maintaining structural coherence and visual quality. Designed for speed, clarity, and creative control, this workflow is served by ComfyUI and uses the Flux 2 Klein model together with a Qwen-based text encoder to drive multi-image image-to-image editing.

What can this ComfyUI workflow do?

In short: Multi-image-to-image transformation and editing with Flux 2 Klein.

This workflow takes multiple source images and a text prompt to transform them into new versions while preserving key visual elements or adapting them according to your instructions. It uses Flux 2 Klein as the diffusion backbone, a Flux 2 VAE, and a Qwen 3 8B text encoder to interpret both image content and textual guidance across multiple images simultaneously, producing detailed and coherent outputs that maintain consistency across the image set. The graph chains multiple ReferenceLatent nodes to propagate structure between images and to build positive and negative conditioning that aligns all images with the prompt.

Example usage in Videcool

Download the ComfyUI workflow

Download ComfyUI Workflow file: flux_2_klein_multi-image_to_image-api.json

Image of the ComfyUI workflow

This figure provides a visual overview of the Flux 2 Klein Multi Image to Image workflow layout inside ComfyUI. The nodes are arranged to establish a clean multi-image transformation pipeline: loading two source images, scaling them to a target resolution, encoding each through the VAE, building chains of positive and negative reference latents, configuring the Flux 2 scheduler and sampler, and finally decoding and saving the outputs. The structure makes it easy to see how the Flux 2 Klein diffusion model, Qwen text encoder, VAE, and sampler interact to produce consistent outputs across all images. Users can modify or expand parts of the workflow to create custom variations or integrate multi-image transformation into larger creative pipelines.

Installation steps

Installation video

The workflow requires two source images, a text prompt, and a few basic parameter adjustments to begin transforming images. After loading the JSON file, users can select the input images, enter edit instructions, adjust the CFG value and steps, and then run the pipeline to obtain transformed outputs that maintain consistency between images. Once executed, the sampler produces new images that preserve key structural elements while applying the requested changes, which can be saved and reused across other Videcool tools.

Prerequisites

To run the workflow correctly, download the Flux 2 Klein diffusion model file, the Flux 2 VAE, and the Qwen 3 8B text encoder, then place them into your ComfyUI directory. These files ensure the model can understand input images, interpret text instructions, and produce high-quality transformed outputs that maintain consistency across multiple images. Proper installation into the following locations is essential before running the workflow: {your ComfyUI directory}/models/diffusion_models/, {your ComfyUI directory}/models/vae/, and {your ComfyUI directory}/models/text_encoders/.

ComfyUI\models\diffusion_models\flux-2-klein-9b-kv-fp8.safetensors
https://huggingface.co/black-forest-labs/FLUX.2-klein-9b-kv/resolve/main/flux-2-klein-9b-kv-fp8.safetensors

ComfyUI\models\vae\flux2-vae.safetensors
https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/vae/flux2-vae.safetensors

ComfyUI\models\text_encoders\qwen_3_8b_fp8mixed.safetensors
https://huggingface.co/Comfy-Org/vae-text-encorder-for-flux-klein-9b/resolve/main/split_files/text_encoders/qwen_3_8b_fp8mixed.safetensors

How to use this workflow in Videcool

Videcool integrates seamlessly with ComfyUI, allowing users to load and run the Flux 2 Klein multi-image workflow without manually wiring the underlying node graph. After importing the workflow file into ComfyUI, Videcool can call it to transform two images simultaneously based on the prompt and user preferences. This makes multi-image editing intuitive and accessible, even for users who are not keen on learning ComfyUI internals, while still benefiting from the power of Flux 2 Klein’s guided composition and the Qwen text encoder.

ComfyUI nodes used

This workflow uses the following nodes. Each node performs a specific role, such as loading source images, scaling them, encoding them into latent space, building chained reference latents for positive and negative conditioning, configuring the Flux 2 scheduler, sampling, decoding, and saving the outputs. Together they create a reliable pipeline that can be extended or customized for different multi-image transformation tasks.

• LoadImage (Load Image 1)
• LoadImage (Load Image 2)
• ImageScaleToTotalPixels 1
• ImageScaleToTotalPixels 2
• GetImageSize
• Load Diffusion Model (UNETLoader: flux-2-klein-9b-kv-fp8.safetensors)
• Load VAE (VAELoader: flux2-vae.safetensors)
• VAE Encode 1
• VAE Encode 2
• Load CLIP (qwen_3_8b_fp8mixed.safetensors)
• CLIP Text Encode (Negative Prompt)
• CLIP Text Encode (Positive Prompt)
• ReferenceLatent Negative 2
• ReferenceLatent Positive 2
• ReferenceLatent Chain Negative 2
• ReferenceLatent Chain Positive 2
• ReferenceLatent Negative Final
• ReferenceLatent Positive Final
• ReferenceLatent Chain Final Negative 2
• ReferenceLatent Chain Final Positive 2
• Empty Flux 2 Latent (EmptyFlux2LatentImage)
• Flux2Scheduler
• RandomNoise
• KSamplerSelect
• CFGGuider
• SamplerCustomAdvanced
• VAE Decode
• Save Image

Base AI model

This workflow is built on Flux 2 Klein, a modern image generation and editing model designed for guided image transformation. In this multi-image configuration, Flux 2 Klein works together with the Flux2 VAE and the Qwen 3 8B text encoder to understand visual and textual inputs, then compose elements from one image into another according to the prompt while maintaining coherence. The chained ReferenceLatent nodes help propagate structure between the images so that both the source and target images participate in the final composition.

Image resolution and multi-image behavior

The workflow uses ImageScaleToTotalPixels to scale both input images to a consistent target megapixel value, then derives the working width and height from one of the scaled images with GetImageSize. This ensures the latent representation and Flux2 scheduler are aligned to the same resolution. The multi-image behavior is driven by how the VAE encodings of each image are chained through multiple ReferenceLatent stages, letting Flux 2 Klein inject elements from one image into the other while following the text prompt.

Conclusion

The Flux 2 Klein Multi Image to Image ComfyUI workflow is a robust, powerful, and user-friendly solution for performing flexible two-image transformations in Videcool. With its combination of the Flux 2 Klein diffusion model, a Qwen-based text encoder, chained reference latents, and a modular ComfyUI pipeline, it enables beginners and professionals alike to produce creative and consistent compositions across multiple images with ease. By understanding the workflow components, installation steps, and advantages, users can unlock the full potential of Flux 2 Klein in multi-image transformation scenarios for video production, design projects, and creative workflows.