Using Creality 2 for AI Influencer Image Editing
This guide walks you through a new community workflow that transforms Creality 2 into a powerful image editing model for AI influencers. You’ll learn how to set up single and multi-image editing, stack LoRAs for better results, and generate high-quality influencer images locally or on cloud GPUs.
What You'll Learn
Prerequisites
Before starting, ensure you have:
- ComfyUI installed and configured (see our ComfyUI Installation Guide)
- The Creality 2 image edit workflow (available through the creator’s Patreon – linked in video description)
- The identity edit LoRA (Hugging Face post linked in the video description)
- Models needed:
- Turbo FP8 scaled model (main Creality 2 model)
- Qwen2 VL 4B FP8 scaled clip
- V 1.2.1 VAE (recommended for better quality)
- Ability to run on a GPU with at least 8GB VRAM (author used a 5090)
If your hardware isn’t powerful enough or you want faster generations, consider using RunPod’s cloud GPU service.
Special Offer - $5 Credit Included!
When you sign up for RunPod using our affiliate link, you'll receive a $5 credit that can be used to generate up to 9,000 images and 300 videos. This gives you plenty of resources to explore ComfyUI and AI image/video generation without any upfront cost!
Step‑by‑Step Process
Step 1: Set Up the Creality 2 Image Edit Workflow
- Download the workflow from the creator’s Patreon (link in video description).
- Load the workflow in ComfyUI – it should appear with clearly labeled sections.
- Download the required models if you haven’t already: the Turbo FP8 scaled model, Qwen2 VL 4B FP8 scaled clip, and V 1.2.1 VAE.
- Place the models in the appropriate ComfyUI model folders (check your ComfyUI installation directory).
- Download the identity edit LoRA from the Hugging Face post (also linked in the video description) and place it in the LoRA folder.
Why V 1.2.1 VAE?
The creator found that using V 1.2.1 VAE produces noticeably better quality than the default VAE. You can experiment, but this is the recommended choice.
Pro Tip
The workflow includes a fast group bypasser that lets you toggle between single‑image and multi‑image editing. Keep it on “single” for testing.
Step 2: Prepare Your Reference Images and Prompts
- Load a base image of your character (the “image to edit”) into the Load Image node.
- Set your output resolution and total megapixels using the provided controls.
- Write your positive and negative prompts in the designated sections. For example:
- Positive: “A brand new image of the woman in image 1. She’s standing in an elevator wearing a red dress and a white hair band, high‑angle smartphone selfie with one arm extended, looking up at the camera with a soft smile.”
- Negative: (leave empty or use standard negative prompt if you have one)
Prompt Tips
When using a full‑body reference, add this to the end of your prompt: “Use image 1 as a character and body proportion reference only.” This prevents background bleed from the reference image.
Pro Tip
The model runs only 8 steps with a CFG of 1 when using the Turbo FP8 model. This makes generations extremely fast – seconds per image on a 5090.
Step 3: Run a Single‑Image Edit with Identity LoRA
- Enable the identity edit LoRA in the workflow (the one you downloaded from Hugging Face).
- Keep all other LoRAs disabled for now.
- Click Queue Prompt to generate.
- Review the result – you should see a new image that follows your prompt while preserving the character’s identity.
What Makes This Powerful
The identity edit LoRA combined with Creality 2’s image editing capability allows you to change pose, outfit, and scene while keeping facial features consistent – even subtle traits like hair color splits are handled automatically without extra prompt engineering.
Pro Tip
If body proportions look off, switch to a reference image that shows more of the character’s body (e.g., a full‑length shot). The model will infer the proportions from that reference.
Step 4: Stack Additional LoRAs for Better Aesthetics
- Enable the Realistic Snapshot LoRA (or similar style LoRA) in the workflow.
- Optionally enable the Filter Bypass LoRA – the creator found it improves overall quality even for non‑promiscuous content.
- Leave your prompt the same or make minor adjustments.
- Queue Prompt and compare with the previous result – the stacked LoRAs should produce a more polished, influencer‑style image.
LoRA Stacking Works Extremely Well
This method supports stacking multiple LoRAs on top of the identity edit LoRA. For the best results, add a character LoRA (if you have one trained on your subject). The creator notes that with a character LoRA, the output can surpass “Nano Banana” quality entirely within ComfyUI.
Warning
Stacking too many LoRAs may reduce fidelity. Start with the identity LoRA and add one or two more for style. Always check the face consistency.
Step 5: Multi‑Image Editing – Swapping Clothing and Backgrounds
- Switch the fast group bypasser to “multi‑image” editing.
- Keep your base character image in the first image slot.
- Load a second image (e.g., a dress photo or a location screenshot) into the second image slot.
- Adjust your prompt to specify the desired change. For example:
- “Replace the red dress of the woman in image 1 with the yellow sundress in image 2.”
- Or for a background swap: “Generate a new image of the woman posing in front of the London Eye using the environment in image 2. Influencer photograph looking off camera to the left, hair caught in the wind.”
- Enable your LoRAs (identity + style).
- Queue Prompt and review.
Environment Clashing
When using a photo for a background, the model may blend the two environments. To avoid this, add a Remove Background node in your workflow. Preview the mask on the background image, then connect the masked image to the Load Image input instead of the raw photo. This forces the model to only use the new background environment.
Pro Tip
For clothing swaps, you may need to run a few generations – the model is not 100% accurate but gets very close (e.g., floral patterns on a dress will match the reference).
Step 6: Iterative Editing – Changing Poses and Creating Carousels
- Save the output image from a previous generation.
- Load that saved image as the new base image in the workflow (still in single‑image mode).
- Write a new prompt to change the pose, for example: “Tilt the woman’s head slightly to the left. Eyes closed with a soft smile. Raise the camera height – high‑angle selfie.”
- Queue Prompt – you’ll get a new pose while keeping the environment and clothing.
- Repeat to build a carousel of different poses for your influencer.
Quality Remains High Without Upscaling
The raw outputs from Creality 2 (even without any upscaling) are already excellent. The creator did not use any upscaling in the examples shown.
Pro Tip
If you notice artifacts (e.g., dress patterns becoming too punctuated), you may need to run additional generations or reduce the CFG slightly. Some brute‑forcing is normal.
Related Guides
- ComfyUI Installation Guide – Complete installation and setup of ComfyUI, including custom nodes and workflow loading.
- Running ComfyUI on RunPod – Cloud GPU setup, network volumes, and cost optimization for running ComfyUI.
- Deploy Your First RunPod – Basic RunPod account creation and first GPU pod deployment.
Next Steps
Now that you can edit AI influencer images with Creality 2:
- Train a character LoRA for your specific subject to achieve even better facial consistency.
- Experiment with multi‑image editing to create background‑swapped product shots or location‑based content.
- Combine this workflow with upscaling (like Ideogram 4‑based upscaling) for higher resolution outputs if needed.
- Check the creator’s AI influencer course (linked in video description) for advanced techniques, including creating the base character images.