AGENT SKILL

infinite-you

Routes InfiniteYou-FLUX identity-preserving photo recrafting tasks across self-contained local inference, model/demo setup, and pipeline-internal debugging.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
infinite-you
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Routes InfiniteYou-FLUX identity-preserving photo recrafting tasks across self-contained local inference, model/demo setup, and pipeline-internal debugging.
disable-model-invocation
true

02 资源文件 13

03 SKILL.md 正文

InfiniteYou Repo Skill

Use this skill when a task is about InfiniteYou / InfU / InfiniteYou-FLUX: identity-preserving image generation from a face image, FLUX.1-dev model setup, local Gradio operation, optional LoRA use, CUDA memory reduction, or debugging the custom InfuseNet pipeline.

Start here

  1. Read references/repo-provenance.md when checking staleness against a current checkout.
  2. Read references/bundled-runtime.md when you need to know what implementation code is bundled with this skill.
  3. Read references/installation-and-models.md before installing dependencies, resolving model paths, or running generation.
  4. Run scripts/check_infinite_you_environment.py for a safe bundled-runtime/import/CUDA/model-layout preflight that does not download models or run generation.
  5. Use the route map below to enter the most specific sub-skill.
  6. If the task involves identity images, public demos, model downloads, or generated outputs, check references/safety-and-licenses.md.

Route map

User task Read
Build or debug a local generation command from identity image, prompt, optional control image, LoRA, seed, model version, and memory flags. sub-skills/local-inference/SKILL.md
Use the Python InfUFluxPipeline API, understand inference parameters, or validate CLI flag names. sub-skills/local-inference/SKILL.md
Prepare or validate model directories, Hugging Face access, InsightFace support files, optional LoRAs, or FLUX base model paths. sub-skills/demo-and-model-setup/SKILL.md
Launch, customize, or troubleshoot the self-contained Gradio demo, model switching, cache behavior, or server binding. sub-skills/demo-and-model-setup/SKILL.md
Modify/debug face detection, ArcFace embeddings, Resampler projection, InfuseNet ControlNet integration, scheduler guidance, quantization, offload, or adapter internals. sub-skills/pipeline-internals/SKILL.md
Diagnose cross-cutting dependency, CUDA, model-access, Gradio, LoRA, or policy failures before choosing a workflow route. references/troubleshooting.md

Installation baseline

This generated skill bundles the InfiniteYou implementation modules under runtime/pipelines/, plus the dependency pins in runtime/requirements.txt. In a fresh runtime, create an isolated Python environment, install those pinned dependencies, and run the bundled checker from the skill directory.

python -m pip install -r runtime/requirements.txt
python scripts/check_infinite_you_environment.py --require-cuda

Full generation requires CUDA and external model files/access. The memory-reduction flags --cpu-offload and --quantize-8bit reduce peak CUDA memory but do not make generation CPU-only. By default, the bundled generation and demo entry points require local model directories; pass --allow-downloads only after explicit approval for network/model-license consequences.

Model and workflow facts

  • InfiniteYou-FLUX v1.0 has two documented variants: aes_stage2 for default aesthetics/text alignment and sim_stage1 for higher identity similarity.
  • The primary input is an identity image containing a detectable face; an optional control image supplies face keypoints.
  • If multiple faces are present, the pipeline selects the largest detected face.
  • Optional Realism and Anti-blur LoRAs are examples and are off by default.
  • Full local execution needs the FLUX.1-dev base model plus InfiniteYou and InsightFace support files, or explicitly authorized download access.

Safe operating rules

  • Use the bundled runtime and scripts by default; do not require the original checkout for normal generation, demo launch, or signature inspection.
  • Run dry-run/preflight helpers before heavy model execution or demo launch.
  • Do not download gated or non-commercial-use models, expose a demo server, or process identity images without confirming the user's authorization and policy constraints.
  • If source code, dependency pins, model variants, CLI flags, or pipelines/ APIs changed since the provenance snapshot, refresh this skill and rebuild the bundled runtime before relying on it.

Useful bundled helpers

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351