InfiniteYou FLUX LoRA
Generates new images of a person from one face photo and a prompt. Built on ByteDance's InfiniteYou for FLUX.1-dev, with your own FLUX LoRAs applied on top for style.
Overview
InfiniteYou FLUX LoRA runs ByteDance's InfiniteYou, an identity-preserving image generator built on FLUX.1-dev, with FLUX LoRAs on top. From one photo of a person and a text prompt, it produces new images of that person in the scene the prompt describes. A LoRA pushes the look toward a trained style: the samples include comic-book art, action-figure packaging and a retro advertisement illustration.
InfiniteYou feeds identity features from the face photo into the FLUX.1-dev diffusion transformer through InfuseNet, a network attached with residual connections. The paper was an ICCV 2025 highlight. This page runs the aes_stage2 checkpoint, which ByteDance describes as better at text-image alignment and aesthetics than sim_stage1, the variant tuned for identity similarity.
What you can build
- Research prototypes of avatar and profile picture generators where a user uploads a selfie and picks a style
- Personal, non-commercial portraits such as toy-box, comic-book and retro-poster versions of yourself
- Experiments on how a style LoRA changes the look while the identity stays fixed
- Studies that place the same consenting person in many different scenes
Inputs
- Face photo: one image of the person, uploaded or given as a URL. It decides whose face appears; the scene comes from the prompt.
- Prompt (required): the scene, clothing, pose and style of the new image.
- LoRA: one FLUX LoRA, either one you trained on Wiro or a direct link to a .safetensors file on Google Drive, AWS S3 or any public file host. Leave it empty to run InfiniteYou on its own.
- Advanced: 40 inference steps by default (1 to 100), guidance scale 7.0 (1 to 20), 1 to 8 images per run, a seed, and width and height (1024 x 1024 by default).
Recommended settings
- ByteDance's reference script defaults to 30 steps and a guidance scale of 3.5, lower than this page's 40 and 7.0. Those values are a reasonable first test.
- If the generated person's gender does not match, ByteDance suggests adding "a man" or "a woman" to the prompt.
- Include a LoRA's trigger word in the prompt when it has one.
- Generate several images per run and keep the closest likeness.
Limitations
- One identity per run: the model takes a single face photo.
- Likeness can be weaker than with sim_stage1, which this page does not use.
Safety & compliance
ByteDance releases the InfiniteYou weights under CC BY-NC 4.0 for academic research only. It asks that the FLUX.1-dev base model and the InsightFace face models it relies on also be used under their own licenses and only for academic research, and FLUX.1-dev itself is published under a non-commercial license. Check these terms against your use case, use face photos only of people who have agreed to it, and do not use the results to impersonate anyone.
API quick start
Run InfiniteYou-flux-lora with a single API call.
{
"prompt": "Your prompt here",
"selectedLoraModel": "...",
"selectedLoraModelPrivate": "...",
"selectedLoraModelPrivateUrl": "..."
}curl -X POST "https://api.wiro.ai/v1/Run/wiro/infiniteyou-flux-lora" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "Your prompt here",
"selectedLoraModel": "...",
"selectedLoraModelPrivate": "...",
"selectedLoraModelPrivateUrl": "..."
}
JSONOther models you might like
Explore similar image generation models.
