🚀 Ultimate Pose & Outfit Workflow for Krea2 Identity
I created this workflow with nodes using Gemini. The nodes contain plenty of influencer poses, model poses, and hundreds of outfits. I also included backgrounds. You can create prompts by randomly selecting each one individually. No need heavy local LLMs and slow generation times. This workflow is built to give you maximum control over poses, influencer aesthetics, outfits, and backgrounds with zero friction.
👤 Need a Unique Influencer Face?
If you need a completely unique face for your AI model, check out my other workflow:
👉 Create Faces From 3-Images (Unique Face Creator) — grab it, generate a brand new face, and use it right here with this pose workflow!
✨ Key Features
No Heavy LLMs Required: Instant prompt generation and pose switching without loading massive language models. Fast, efficient, and lightweight.
Built-in Gemini Pose Node: Hundreds of preset poses, spicy options, influencer styles, outfits, and tailored backgrounds.
Two-Pass Refinement: Uses a 2-sampler setup for high-detail skin texture and realism.
Auto Background Removal: Integrated REMBG setup for clean subject isolation and grounded encoding.
🛠️ How to Install the Custom Node
Copy the
.pyfile into your mainComfyUI/custom_nodes/folder.Restart ComfyUI and load this JSON workflow!
📦 Recommended Models & Encoders
Diffusion Model: https://civarchive.com/models/2741166/muse-by-stable-yogi-krea2?modelVersionId=3147272
CLIP Encoder: https://huggingface.co/DreamFast/Qwen3-VL-4b-Heretic-ComfyUI/blob/main/qwen3-vl-4b-heretic_fp8_e4m3fn.safetensors
VAE: https://huggingface.co/martineux/altkreas/blob/main/krea2RealVae_v10.safetensors
☕ Support the Project:
If you enjoy this workflow and want to support future node releases and updates, consider buying me a coffee:
Description
FAQ
Comments (20)
really awesome, fun to play with. only thing i noticed was that the seeds for your samplers arent set so they end up being two different seeds. Doesn't ruin it but it can have some weird effects.
Thanks! Actually (if I understood your criticism correctly), I made the "sampler" seed random in the second pass. The reason is that I tie sageattention to the first one. I don't tie it to the second one. Because sageattention distorts the faces. The faces that are distorted by sageattention in the first one are corrected in the second one. The reason is very simple. I gain 5 seconds.
@ugurdoyduk341 Ooooh, nice!
can you replace links in description with actual links not google search, please?
Updated
I have the same problem
same
my bad , i forgot to load the krea edit lora
Very nice workflow. I get great results with horizontal format, but with horizontal pictures i get in more then half of the generations twins (2 characters). Even prompting onegirl doesn't help. Any idea?
I have the same problem(((
Yes, this is disturbing. It does this even though there is only one clothing section selected in the Gemini node.
Hi, its possible to add pose or change them?
Certainly. Turn off Comfy. Back up the .py file you downloaded first. Then open it with notepad. Go to the # New Poses Category section. I think it starts from line-166. You can add a new pose there. Every pose you write must start with " and end with ". Quotation marks are important. Otherwise the node gives an error. save it. Then restart comfy.
Or upload the entire file to Gemini and type "I want to add new poses to the pose category." Write down your poses. Type "Give it to me as a .py file". Type "Don't change anything else." It will handle it :)
wow, thanks man:)
@Cukrova_Vila You're welcome
Not sure what I'm doing wrong, I have all the correct nodes as well as models and vae and clip but when it generates an image it doesn't do anything besides create a jumbled up mess.
Everything works with version 1.1 of LoRa
eager to try it I'm setting all for it. The pic before the last one looks odd, hopefully is cos bad prompt pairing a ruffle camisole with tight shorts