Hey folks, a while back I posted this request asking for help replicating the Flux-Kontext Portrait Series app from Replicate output in ComfyUI.
Runtime profile
Source description
Hey folks, a while back I posted this request asking for help replicating the Flux-Kontext Portrait Series app from Replicate output in ComfyUI.
Well... I ended up getting it thanks to a lot of help in the community. I posted on Reddit first but I forgot I should have posted one here..
This is a work-in-progress, not a polished solution, but it should get you 12 varied portraits using the FLUX-Kontext model—complete with pose variation, styling prompts, and dynamic switches for RAM flexibility.
🛠 What It Does:
Generates a set of 12 portrait variations using dynamic prompt injection
Rotates through pose strings via iTools Line Loader + LayerUtility: TextJoinV2
AI-generated commentary
AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.
Use this ComfyUI workflow to turn an input image, styling prompts, rotating pose strings, and a chosen seed into 12 varied FLUX-Kontext portrait images, with optional face swap and background removal.
Dynamic prompt injection rotates through pose strings using iTools Line Loader and LayerUtility: TextJoinV2.
Any Switch (rgthree) switches between model, CLIP, and VAE options for low- or normal-RAM setups.
Batch text injection and seed control are included.
The output is a set of 12 portrait variations with pose preservation and face consistency across the images.
Optional face swap and background removal tools are included, but the final images are not stitched into a grid.
The base model is Flux.1 Kontext.
For background removal, download model.pth and place it in ComfyUI/custom_nodes/ComfyUI_LayerStyle/RMBG-1.4/.
Required node packs include comfyui_essentials, comfyui_layerstyle, ComfyUI_LayerStyle_Advance, and comfyui-custom-scripts.
Other named node packs are comfyui-itools, comfyui-multigpu, rgthree-comfy, comfy-core, and ReActorFaceSwap; ReActorFaceSwap is optional.
Required model files include clip_l.safetensors, flux1-dev-kontext_fp8_scaled.safetensors, flux1-kontext-dev-Q5_K_S.gguf, and t5xxl_fp8_e4m3fn.safetensors.
To cycle through the prompts, queue 12 runs and set the text number to zero.
Suggestion · not verified
You can change the prompts and background color, adjust the seed, and remove the optional face-swap block.
Suggestion · not verified
This workflow is described as a work in progress rather than a polished solution; prompt logic may need refinement and pose diversity may need manual tweaks.
Does this need to be edited?
Sign in to send an edit request.
Sources
1 sourceSource excerpts
1 excerptSource context: 847 downloads · Type Workflows · Base model Flux.1 Kontext
Allows model/clip/VAE switching for low vs normal RAM setups using Any Switch (rgthree)
Includes pose preservation and face consistency across all outputs
Batch text injection + seed control
Optional face swap and background removal tools included
Que up 12 and make sure the text number is at zero (see screen shots) it will cycle through the prompts. You of course can make better prompts if you wish. The image makes a black background but you can change that to whatever color you wish.
lastly there is a faceswap to improve on the end results. You can delete it if you are not into that.
This is all thanks you BGRated who requested such a workflow. who did this for me on Matteo's Discord server. Thank you zGenMedia you rock.
📦 Node Packs Used:
rgthree-comfy (for switches & group toggles)
comfyui_layerstyle (for dynamic text & image blending) NOTE: Download this model (model.pth) and place it ComfyUI/custom_nodes/ComfyUI_LayerStyle/RMBG-1.4/
comfyui-itools (for pose string rotation)
comfyui-multigpu (for Flux-Kontext compatibility)
comfy-core (standard utilities)
ReActorFaceSwap (optional FaceSwap block)
ComfyUI_LayerStyle_Advance (for PersonMaskUltra V2)
⚠️ Heads Up:
This isn’t the most elegant setup—prompt logic can still be refined, and pose diversity may need manual tweaks. But it’s usable out the box and should give you a working foundation to tweak further.
📁 Download & Screenshots:
[ Workflow: https://pastebin.com/v8aN8MJd ] Just remove the txt at the end of the file if you download it.
Grid sample and pose output previews attached below are stitched by me the program does not stitch the final results together.
Any help or support can be done here . It literally puts food on my table. If you need Runpod credits any new joins get $500 for $10 install.
Estimated VRAM requirement
32+ GB estimated
23.6 GB of model files. Model file total + 25% loading overhead + 2 GB execution buffer, rounded up.
Requirements
11 requirementsText encoder · 235 MB · SAFETENSORS · Unknown
Checkpoint · 11.1 GB · SAFETENSORS · Unknown
Checkpoint · 7.71 GB · GGUF · Unknown
Text encoder · 4.56 GB · SAFETENSOR · fp8 · Unknown
Node pack · Registry
Node pack · Registry
Node pack · Registry
Node pack · Registry
Node pack · Registry
Node pack · Registry
Node pack · Registry