InfiniteYou: Flexible Photo Recrafting While Preserving Your Identity
Abstract: Achieving flexible and high-fidelity identity-preserved image generation remains formidable, particularly with advanced Diffusion Transformers (DiTs) like FLUX. We introduce InfiniteYou (InfU), one of the earliest robust frameworks leveraging DiTs for this task. InfU addresses significant issues of existing methods, such as insufficient identity similarity, poor text-image alignment, and low generation quality and aesthetics. Central to InfU is InfuseNet, a component that injects identity features into the DiT base model via residual connections, enhancing identity similarity while maintaining generation capabilities. A multi-stage training strategy, including pretraining and supervised fine-tuning (SFT) with synthetic single-person-multiple-sample (SPMS) data, further improves text-image alignment, ameliorates image quality, and alleviates face copy-pasting. Extensive experiments demonstrate that InfU achieves state-of-the-art performance, surpassing existing baselines. In addition, the plug-and-play design of InfU ensures compatibility with various existing methods, offering a valuable contribution to the broader community. sim_stage1 – Prioritizes higher identity similarity for more accurate face preservation. aes_stage2 – Focuses on better text-image alignment and enhanced aesthetics. ComfyUI InfiniteYou: Your Face, Your Rules!
(And No, It's Not About Making Infinite Copies of Yourself...Probably.)
- What's This "InfiniteYou" Thing Anyway?
Imagine a world where you can take a photo of yourself (or your grandma, no judgement!), throw it into a magical AI blender, and ask it to turn you into a cyberpunk warrior, a renaissance painting, or...well, anything your imagination cooks up! That's the power of InfiniteYou!
Built using some seriously cool tech from ByteDance (think "FLUX diffusion transformer" and "InfuseNet"—sounds like something out of Star Trek, right?), InfiniteYou lets you generate images that are both wildly creative AND still recognizable as you. Think of it as giving your face a super-powered makeover, while still retaining what makes you, you.
This toolkit isn't a one-size-fits-all wonder. It comes with two awesome workflows:
Face Combine: Mashup two faces into one crazy hybrid! (Ever wondered what you'd look like with Brad Pitt's jawline? Now's your chance!)
Zero-Shot Task: Unleash the power of text-based prompts to transform a single face into any role or style you can dream up!
- Why Should You Care About InfiniteYou?
Here's the lowdown:
Identity Protection Program (For Your Pixels): Forget looking like a generic AI-generated blob. InfiniteYou clings to your unique facial features like a lovesick koala.
Beauty Mode Activated: The "aes_stage2" mode is like a digital spa day for your images. It makes everything look more aligned, prettier, and generally more "wow"-inducing.
Workflow Wonderland: With both Face Combine and Zero-Shot Task, you've got options galore! It's like having a Swiss Army knife for face-morphing fun.
Total Control Freak Approved: Tweak those knobs and dials! You can adjust everything from how much influence the AI has (guidance) to the timing of the morph (fusion weights).
ComfyUI's New Best Friend: Slap this baby into your existing ComfyUI workflows – plug and play!
- How to Actually Use This Thing (Because Tech Can Be Confusing!)
IMPORTANT: When you fire up InfiniteYou for the first time, you'll see the Zero-Shot Task AND Face Combine options. Here's the deal: they're like two separate apps on your phone. They can technically run at the same time, but they're meant to be used independently. Don't try to wire them together like Frankenstein's monster – it won't end well!
3.1. InfiniteYou Generation: Step-by-Step, Simplified
Let's get you started. Here's a quick rundown:
Step 1: Load Up the Good Stuff (The Images)
Face Combine: Grab two photos of faces. The AI is gonna blend them.
Zero-Shot Task: Pick a single photo of a face. This one's getting the full makeover.
Step 2: Tell the AI What You Want
CLIP Text Encode nodes: type a prompt (e.g., "a boy, 10 years old, handsome in the classroom")
A negative prompt can be included
Step 3: Queue It Up! Hit that "Queue Prompt" button and watch the magic happen.
Step 4: Voila! Head to the Save Image node and see the transformation!
Diving Deeper: Choose Your Adventure!
Face Combine Workflow (The Face-Melting Fun Zone)
Best For: Creating totally new faces by mixing two existing ones. Think of it as genetic engineering, but with pixels!
Key Features:
Blends two faces together like a digital smoothie.
Fine-tune the blending with those handy "weight" sliders.
Control when the blending happens with "start" and "end" timing.
Zero-Shot Task Workflow (The Extreme Makeover)
Best For: Turning a single face into ANYTHING your heart desires, guided by a text prompt.
Key Features:
Keeps your identity super intact with "sim_stage1".
No need for a second face – just a single hero shot.
Uses your text prompt to completely reshape the image.
3.2. Parameter Playtime: Tweak It 'Til You Make It!
Here are the key knobs and dials you'll be playing with:
Face Combine Node:
adapter_file: Pick your blending model (like "aes_stage2_img_proj.bin").
weight: How strong should the blending be? Crank it up for a crazy hybrid!
balance: Which face gets more influence?
start_at and end_at: Control when the fusion kicks in.
fixed_face_pose: Lock the original pose, or let it move around.
Apply Node:
adapter_file: Which "stage" model to use? (More on this later!)
weight: How much identity to preserve.
start_at and end_at: When does the effect kick in?
fixed_face_pose: Keep the pose rigid, or let it be flexible.
FluxGuidance / BasicGuider:
guidance: More = More control, lower = More variety
Samplers:
sampler_name: Algorithm to generate the image
steps: Number of iterations to refine the image.
denoise: How much noise is removed
3.3. Advanced Ninja Tricks: Optimizing for Maximum Awesome
Model Switching: The Secret Sauce
aes_stage2: For better text-image coherence and styles
sim_stage1: For accuracy face identity retention
Important: Always update adapter_file and control_net files together!
Prompting Like a Pro
Identity Cues FTW: Add phrases like "a woman", "an elderly man", "a child" to nail the look.
Be Specific: Crystal-clear prompts are your friend. Describe the subject and the setting precisely.
So, go forth and create! InfiniteYou is your playground for facial experimentation. Just remember to have fun and don't blame us if you accidentally create a new meme!
