AGENT SKILL

dreamomni2

Routes DreamOmni2 image generation, image editing, and Gradio demo workflows.

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

01 前言 Front Matter

name
dreamomni2
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Routes DreamOmni2 image generation, image editing, and Gradio demo workflows.
disable-model-invocation
true

02 资源文件 7

03 SKILL.md 正文

DreamOmni2

Use this skill when the task names DreamOmni2, FLUX.1-Kontext, two-image instruction-based image editing, two-image image generation, or the bundled Gradio demos.

What this skill covers

  • Two-image editing and generation on a CUDA GPU
  • The VLM prompt stage that converts instructions into a diffusion prompt
  • Gradio launchers for the editing and generation demos
  • Shared model-path, environment, and troubleshooting guidance

Prerequisites

  • NVIDIA CUDA hardware with enough VRAM for the DreamOmni2 / Qwen2.5-VL / FLUX.1-Kontext stack
  • Python 3.11 or another supported modern CPython
  • Runtime dependencies from requirements.txt plus Gradio
  • Local model assets or hub IDs for the VLM and LoRA weights

Start here

  1. Read references/model-setup.md for the expected model layout.
  2. Run scripts/check_env.py to verify the Python, torch, and GPU setup.
  3. Run scripts/check_models.py to confirm the model paths you intend to use.
  4. Choose the matching sub-skill:
    • sub-skills/inference/ for CLI generation and editing.
    • sub-skills/web-demo/ for the Gradio launchers.

Shared runtime helpers

  • scripts/dreamomni2_common.py contains the bundled image-resize and prompt-format helpers used by both sub-skills.
  • references/troubleshooting.md collects the shared install, GPU, model-path, and import failures.

Route map

sub-skills/inference/

Use this route for:

  • editing one source image with a second reference image
  • generating a new image from two references
  • prompt-stage and diffusion-pipeline questions
  • inference_edit.py and inference_gen.py-style command-line runs

sub-skills/web-demo/

Use this route for:

  • launching the editing or generation Gradio UI
  • choosing ports, server names, and upload order
  • UI-specific troubleshooting such as port conflicts and browser launch issues

Do not use this skill for

  • training, FSDP, or distributed optimization helpers
  • dataset preparation or benchmark infrastructure
  • generic diffusion research that is not DreamOmni2-specific

Routing notes

  • The CLI and web demo workflows both depend on the same model layout and the same CUDA-capable stack.
  • The editing workflow requires the source image to be first and the reference image to be second.
  • The generation workflow uses the same two-image VLM prompt stage but a different LoRA adapter.
  • If the user only needs installation or model-layout validation, start with the root helper scripts and the model-setup reference before going into a sub-skill.

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