ComfyUI Flux Kontext Dev Grouped Workflow: Image Combo Generator

Because sometimes, one image just isn’t enough.

This documentation covers a ComfyUI Flux Kontext Dev Grouped workflow that fuses two images into a beautifully composed single output using ImageStitch, FluxKontextImageScale, and a dash of AI-enhanced creativity with FLUX.1 Kontext Image Edit. Whether you're nesting penguins on pillows or just merging concepts side-by-side like a true visual alchemist, this setup has you covered.

📷 Example Workflow Layout

🐧 Example Input Images

Image A:

Image B:

🛠️ What This Workflow Does

This workflow takes two input images, lines them up side-by-side, and gives them a cozy makeover using the FLUX.1 Kontext Image Edit node — not once, but twice. The first pass refines the stitched image with prompt-based guidance. The second adds enhancements (like stylized text or flair) without blowing up the original composition.

This allows for:

• Prompt-based layout blending

• Multi-step generation from source images

• Artistic augmentation with minimal loss of visual coherence

🧠 Key Components

Node	Purpose
LoadImage (x2)	Loads both input images (black-penguin.png and purple-penguin.png)
ImageStitch	Combines the two images into one, placed side-by-side
FluxKontextImageScale	Normalizes dimensions for encoding
VAELoader + VAEEncode	Prepares the image for latent-based editing
FLUX.1 Kontext Image Edit (1st pass)	Adds prompt-based composition using the stitched latent image
FLUX.1 Kontext Image Edit (2nd pass)	Adds optional enhancements (like text overlay) on top of previous results
PreviewImage	Lets you preview the stitched output
SaveImage (x2)	Saves both versions: the first with prompt-based fusion, the second with additional text or tweaks

⚙️ Parameters That Matter

🧵 ImageStitch Settings

Setting	Value	Description
direction	right	Stitch image2 to the right of image1
match_image_size	true	Resizes images to match dimensions
spacing_width	0	No extra space between images
spacing_color	white	Not used here due to zero spacing

🧬 FLUX.1 Kontext Image Edit Settings

Setting	Pass 1	Pass 2
prompt	Place both cute penguins...	Add a big purple 3D style "Double Naplin" text...
seed	386100423184929	979175714064968
sampler	euler	euler
scheduler	simple	simple
denoise	1	1
cfg	1	1
guidance	2.5	2.5
unet	flux1-dev-kontext_fp8_scaled.safetensors	(same)
clip1	t5xxl_fp8_e4m3fn.safetensors	(same)
clip2	clip_l.safetensors	(same)

✅ Benefits

• Multi-pass refinement: Clean stitching + customizable augmentation.

• Prompt controlled: Direct how your final image looks — colors, composition, context.

• Fully automated: No Photoshop-fu required.

💡 Usage Tips

• Stick to images with similar dimensions for better results.

• Keep spacing_width at 0 unless you want a border or divider.

• First pass = layout/prompt, second pass = enhancement/prompt (e.g., text overlays, stylization).

• For fancy setups, consider using masks or ControlNet for alignment — though not needed here.

🧯 What-Not-To-Do-Unless-You-Want-a-Fire

• ❌ Don’t mismatch your VAE — this workflow uses ae.safetensors. Mixing in something random might cause encoding artifacts.

• ❌ Don’t skip VAEEncode — the entire Flux pipeline expects latents, not raw pixels.

• ❌ Don’t pipe FluxKontextImageScale output directly into ImageStitch — it needs the stitched output, not individual scaled images.

• ❌ Don’t forget your prompts — or you’ll just get a very boring stitched image.

📍 Setup Instructions (ComfyUI)

1. Place all required model files:

   • flux1-dev-kontext_fp8_scaled.safetensors → models/diffusion_models

   • t5xxl_fp8_e4m3fn.safetensors and clip_l.safetensors → models/text_encoders

   • ae.safetensors → models/vae

2. Load the JSON into ComfyUI using the ⚙️ Load Workflow button.

3. Add your two input images in the LoadImage nodes.

4. Tweak prompts, sampler, and denoise as needed.

5. Click Queue Prompt and watch the penguins get comfy.

📚 Additional Resources

• 🔗 Download Workflow JSON

• 🧠 FLUX.1 Kontext Documentation

• 🎓 ComfyUI Basics

📎 Example Node Configuration

json

"widgets_values": [   "right", true, 0, "white" // <- ImageStitch ]

json

"widgets_values": [   386100423184929,   "fixed",   20,   1,   "euler",   "simple",   1,   2.5,   "ComfyUI",   "Place both cute penguins laying next to each other on a pile of pillows and ensure that the colors are maintained.",   "",   "flux1-dev-kontext_fp8_scaled.safetensors",   "default",   "t5xxl_fp8_e4m3fn.safetensors",   "clip_l.safetensors",   "flux",   "default",   "ae.safetensors" ]

📝 Notes

• Yes, you can swap the penguins for anything else. No, we won’t judge.

• If your final output looks wonky, blame your prompts (or your seed).

• Grouped nodes are handled in workflow>FLUX.1 Kontext Image Edit, so don’t panic when you see multiple passes — it’s intentional.